doi
stringlengths 28
28
| title
stringlengths 19
311
| abstract
stringlengths 217
5.08k
| plain language summary
stringlengths 115
4.83k
| article
stringlengths 3.87k
161k
|
|---|---|---|---|---|
10.1371/journal.pgen.1002930 | Phylogenetic and Transcriptomic Analysis of Chemosensory Receptors in a Pair of Divergent Ant Species Reveals Sex-Specific Signatures of Odor Coding | Ants are a highly successful family of insects that thrive in a variety of habitats across the world. Perhaps their best-known features are complex social organization and strict division of labor, separating reproduction from the day-to-day maintenance and care of the colony, as well as strict discrimination against foreign individuals. Since these social characteristics in ants are thought to be mediated by semiochemicals, a thorough analysis of these signals, and the receptors that detect them, is critical in revealing mechanisms that lead to stereotypic behaviors. To address these questions, we have defined and characterized the major chemoreceptor families in a pair of behaviorally and evolutionarily distinct ant species, Camponotus floridanus and Harpegnathos saltator. Through comprehensive re-annotation, we show that these ant species harbor some of the largest yet known repertoires of odorant receptors (Ors) among insects, as well as a more modest number of gustatory receptors (Grs) and variant ionotropic glutamate receptors (Irs). Our phylogenetic analyses further demonstrate remarkably rapid gains and losses of ant Ors, while Grs and Irs have also experienced birth-and-death evolution to different degrees. In addition, comparisons of antennal transcriptomes between sexes identify many chemoreceptors that are differentially expressed between males and females and between species. We have also revealed an agonist for a worker-enriched OR from C. floridanus, representing the first case of a heterologously characterized ant tuning Or. Collectively, our analysis reveals a large number of ant chemoreceptors exhibiting patterns of differential expression and evolution consistent with sex/species-specific functions. These differentially expressed genes are likely associated with sex-based differences, as well as the radically different social lifestyles observed between C. floridanus and H. saltator, and thus are targets for further functional characterization. Our findings represent an important advance toward understanding the molecular basis of social interactions and the differential chemical ecologies among ant species.
| Chemical communication is an important factor in the regulation of social interaction in animals. The family of eusocial insects commonly known as ants offers an almost unique opportunity for examining the genetic basis for the chemosensory pathways that underlie ant sociality. In order to address this issue, we have manually and comprehensively reannotated the chemoreceptor repertoire in a pair of evolutionarily and behaviorally divergent ant species, Camponotus floridanus and Harpegnathos saltator. In addition, we have used next-generation RNA sequencing to examine the chemosensory receptor transcriptome between males and females within these species. Our analysis demonstrates rapid gene birth-and-death for the ant odorant and gustatory receptor gene families, as well as clear differences in the expression of particular subsets of chemoreceptor genes between males and females. Finally, we have begun to examine the odor space within these discrete social units by heterologous characterization of the first C. floridanus odorant receptor that also exhibits sex-specific differential expression. Taken together, our results provide a foundation for future studies of the genetic basis for the chemical signaling and chemical ecology underlying the dramatically different social lifestyles exhibited by these and other species of ants.
| The family of insects commonly known as ants (family Formicidae) originated during the Cretaceous period, approximately 140 million years ago [1]. Since that time, they have established a global presence, with only the most remote locations lacking ant species [2]. Indeed, in some cases, such as lowland tropical rainforest canopies, ants have come to dominate the biomass [3], [4]. Their ecological success is reflected in the number and diversity of ants, of which there were 283 known genera [5].
There is a wide diversity in the behavior and morphology of different ant subfamilies that includes both the level and complexity of social organizations. For instance, Camponotus floridanus (the Florida Carpenter Ant), is a Formicine ant from the South-Eastern United States which belongs to one of the most globally prevalent ant genera [6]. These ants feature a rigid caste structure, with strict division of labor between the reproductive queens and the non-reproductive workers that is primarily regulated through pheromones [7], [8], [9]. Workers have a high threshold to lay eggs, and regulation of their reproduction through aggressive interactions does not occur [10]. Furthermore, the worker caste is divided into two classes: minor workers and major workers, which differ in size and morphology [2], [6]. On the other hand, Harpegnathos saltator, a predatory species of Ponerine ant endemic to India and Sri Lanka is characterized by a more flexible reproductive system. H. saltator colonies are relatively small (averaging 65 to 225 individuals, depending on season and region) [11], and queen to worker dimorphism is weak [11], [12]. When a H. saltator colony loses its queen, one or more of the workers will begin laying eggs and become functional reproductives (referred to as gamergates) [12] and this behavioral transition is initiated with strong aggressive interactions [13].
Sociality in ants is considered to be a simple model for complex behaviors in humans and other mammals [14]. The success of ants is thought to have arisen in large part from their well-developed eusociality, wherein individuals live together in colonies with one or several highly fertile female “queens” surrounded by a host of non-reproductive female “workers.” These workers then support and defend the queen and her progeny. The fact that the workers are the queen's own daughters is thought to provide the evolutionary advantage for the workers to protect and support the queen [6].
While it is generally accepted that a variety of chemical signals mediate many of the interactions between these castes, as well as interactions between individuals from competing colonies, there is great interest in determining the particular pheromones and their cognate molecular receptors that mediate these interactions [2]. It is likely that these semiochemicals are initially detected in peripheral sensory neurons by members of three major insect chemosensory receptor gene families: odorant receptors (Ors) [15], [16], [17], [18], [19], gustatory receptors (Grs) [15], [20], [21], [22], [23], and the more recently discovered variant ionotropic glutamate receptors (Irs) [24], [25], [26].
Ors and Grs belong to the same superfamily and both encode seven-transmembrane-domain proteins [17], [22]. Ors are mainly expressed in olfactory receptor neurons (ORNs) within sensory appendages such as antennae and maxillary palps, where they are responsible for the perception of volatile chemical signals [17], [19]. Conventional insect Ors (so-called “tuning” Ors) are associated with odorant specificity. They are typically highly divergent and their orthologous relationships are usually difficult to determine even within order (e.g. Drosophila vs. Anopheles [27], and Nasonia vs. Apis [28]). In contrast, one member of this gene family, which is now uniformly known as Orco, is both highly conserved across insect orders and widely expressed in a majority of ORNs [29], [30]. Orco is necessary and sufficient for the proper localization and retention of other tuning Ors at the dendritic membrane, and is required for proper function of tuning Ors [29], [31]. Rather than playing a role in odorant specificity, Orco forms an essential part of a heteromeric ion channel in cooperation with a tuning Or that is gated by its cognate odor ligand [32], [33], [34], [35], [36].
In contrast with the Ors, Grs are highly expressed in gustatory organs [20], [21], [22], and a large portion of these receptors respond to soluble tastants [37], [38], [39] and pheromones [40], [41], [42], leading to the “gustatory” designation for this group of chemoreceptors. However, there are some exceptions; for example, one unusual group of Grs respond to the volatile chemical carbon dioxide [43], [44], demonstrating that members of this receptor family are not necessarily limited to gustatory or pheromonal responses. This is further supported by the expression of some Grs in non-gustatory organs such as the arista and Johnston's organ [45].
Irs are homologous to ionotropic glutamate receptors (iGluRs) and thus are evolutionarily unrelated to Ors and Grs [24], [26]. The role of IRs as chemosensory receptors has recently been uncovered based on multiple lines of evidence, including their divergence from conventional iGluRs at sequence level and the expression of several Irs in chemosensory neurons [24]. While Irs are generally thought to mediate responses to acids and amines [25], members of this family of chemosensory receptors may also sense other classes of chemicals.
We hypothesize that the striking contrast between C. floridanus, with its strict queen-worker dimorphism and largely pheromone-regulated reproduction, and H. saltator, with its flexible reproductive system that is associated with behavioral and pheromonal regulation of reproduction, is correlated with distinctive semiochemical and chemoreceptor profiles, which in turn generate differences in their chemical ecologies. The same is likely to be true of caste- or sex-based differences in behavior within each species.
To test these hypotheses, we first developed a custom gene annotation pipeline to comprehensively describe the chemosensory receptor repertoires of C. floridanus and H. saltator. We then investigated the evolutionary patterns (e.g. gene gain-and-loss) of these chemosensory receptor genes, in order to gain insight on their functional diversification. Furthermore, we performed RNAseq analyses of caste- and sex-specific antennal transcriptomes to identity chemoreceptors that are differentially expressed between males/females and between species. We found multiple clades of chemosensory receptor genes that show differential expansion/contraction among ant species. In addition, a large number of chemosensory receptor genes exhibited sex-specific expression or male/female-enrichment. These chemosensory receptor genes exhibiting interesting evolutionary and expression patterns may have potentially contributed to the different chemical ecology between sexes/species. We also successfully identified agonists for two Or genes to further validate these annotations. The findings of this study inform us as to the genetic basis for the differences in chemical ecology between C. floridanus and H. saltator, as well as the potential role of chemosensory receptors in the biology and evolution of eusociality in ants.
The automated genome annotations of C. floridanus and H. saltator revealed about 100 Or and about 10 Gr genes [46], which is substantially fewer than the number of Or and Gr genes in two other sequenced ant genomes (e.g. argentine ant: Linepithema humile [47], and harvester ant: Pogonomyrmex barbatus [48]; Figure 1). These low numbers were not surprising because the annotation of Or/Gr genes in other insect genomes has been difficult and usually requires extensive manual efforts [47], [48]. In order to address this potential discrepancy and comprehensively elucidate the genomic repertoire of chemosensory receptor genes in C. floridanus and H. saltator, we rigorously re-annotated Or, Gr, and Ir genes in these two ant species using a custom automated pipeline followed by careful manual inspection.
To maximize the sensitivity of our re-annotation, we collected reported Or, Gr, and Ir gene sequences from other sequenced Hymenoptera and insect relatives of C. floridanus and H. saltator, including Apis mellifera, Acyrthosiphon pisum, Drosophila melanogaster, Nasonia vitripennis, L. humile, and P. barbatus. These insect chemosensory receptor genes were used to identify putative Or/Gr/Ir coding regions within the C. floridanus and H. saltator genomes and to guide homology-based gene prediction. As a result, we discovered a large number of previously unannotated chemosensory receptor genes and corrected several previously reported gene models [46]. All these annotations were manually inspected in multiple sequences alignments to identify and correct for potential errors (e.g. missing exons, unrelated sequences). This analysis indicates that C. floridanus contains 407 putative Or coding loci, of which 352 loci encode intact Or genes, which is similar to those newly annotated in H. saltator, with 377 loci in total and 347 intact loci (all chemosensory receptor genes annotated in this study are available in Dataset S1). The number of Ir predictions is also similar between the two ants, with 31 Ir genes in C. floridanus and 23 in H. saltator. On the other hand, C. floridanus contains 46 intact Gr genes, which is significantly higher than the 17 intact Gr genes found in H. saltator (Figure 1). Moreover, all three families of chemosensory receptor genes exhibited high degrees of sequence divergence among family members (Table S1).
In addition to the chemosensory receptor genes listed above, we also found a large number of incomplete gene models in these two ant genomes. For example, in C. floridanus and H. saltator, there are respectively ∼100 and ∼80 Or gene models encoding proteins shorter than 300 amino acids. In parallel to the difference in intact Gr genes, only three fragmented Gr gene models were found in H. saltator, while C. floridanus has ∼30 short Gr genes. Close examination of their genomic sequences revealed two principal mechanisms apparently leading to these fragmented Or/Gr gene models: 1) the presence of multiple frame-shift mutations and premature stop-codons, suggesting that they represent pseudogenes; and 2) their locations around undetermined genomic regions (e.g. edges of contigs/scaffolds), indicative of incomplete assembly as expected from a draft genome. The latter mechanism explains about 80% of the incomplete gene models.
Furthermore, similar to other insects [28], [47], [48], [49], [50], [51], most chemosensory receptor genes are tandemly arrayed in the C. floridanus and H. saltator genomes. In both cases, about 75% of Or genes are located in gene clusters of 4 to about 40 genes, and these occur in 24 and 20 Or gene clusters (n≥4) in C. floridanus and H. saltator, respectively (Figure S1). Although to a lesser degree than the Ors, half of the Gr and Ir genes in both ants have at least one neighboring homolog.
To better understand the evolutionary history of chemosensory receptor genes in the two ant species, we performed Hymenoptera-wide phylogenetic analysis on each of the OR, GR, and IR gene families. Additional analyses including D. melanogaster and Tribolium castaneum showed that most relationships among hymenopteran and non-hymenopteran sequences were not resolved within the OR and GR families (see below). In this study, while they are generally categorized as belonging to the same receptor superfamily [22], we elected to analyze the OR and GR families separately due to their high level of divergence.
To further understand the evolutionary dynamics of chemosensory receptor genes, we quantified the gene birth and death events and estimated the number of ancestral gene copies in each family using both the maximum-likelihood (ML) and the parsimony based methods implemented in CAFÉ [62] and Notung [63], respectively. For all three families, the ML method suggested relatively high copy numbers in the ancestor of Hymenoptera (Figure 5). For instance, it estimated a repertoire of 266 Or genes in the hymenopteran ancestor, which was expanded in all ant lineages, but significantly contracted in both N. vitripennis and A. mellifera. A similar pattern was also observed in both the GR and IR families. Moreover, the ML analysis suggested that the low number of Gr genes in H. saltator is due to a significant gene loss in this lineage.
On the other hand, the parsimony approach gave conservative estimates of ancestral copy numbers and showed that many more gene-gain events occurred during later stages of hymenopteran evolution. According to the parsimony analysis, the number of Or genes increased from 25 in the last common ancestor of Hymenoptera to about 200 in N. vitripennis and A. mellifera, and more than 300 in all four ants (Figure 5A). Most notably, the repertoire of Or genes increased by three-fold in the ancestor of ants (from 51 to 204 copies), after the separation of A. mellifera, and continued to expand greatly along each ant lineage. Interestingly, although to a lesser degree, the ML method also identified significant expansion on the branch leading to the ant ancestor. In addition to the large number of gene gains, substantial gene losses also occurred in all ants. On the other hand, most duplications of ant Grs occurred in C. floridanus, L. humile, and P. barbatus, while there were only one gene gain and four gene loss events on the lineage to H. saltator (Figure 5B). Similar to the OR and GR families, the number of Ir genes also doubled in the ancestor of ants after its separation from other Hymenoptera (Figure 5C). Subsequent increase of Ir gene number was only observed in C. floridanus and L. humile.
Overall, the ML and parsimony analyses gave different estimates of the ancestral copy numbers and gene gain and loss events. The ML method assumes a random gene birth and death process [64], which is significantly violated by both the OR and GR families (p-values<0.01). On the other hand, the parsimony approach aims to minimize the number of gene gain and loss events, and thus might underestimate the number of ancestral copies. Nonetheless, both analyses support the hypothesis that chemosensory genes have distinct evolutionary dynamics in ant lineages in comparison to the other two hymenopterans.
In insects, most Ors and some Grs/Irs are expressed in antennal ORNs [18], [24], [49], [65]. As best illustrated in studies of the Drosophila olfactory system, each ORN expresses a single tuning Or which is responsible for the odorant response profile and all the ORNs expressing that singular tuning Or send axonal connections to a single antennal lobe glomerulus thereby providing a mechanistic basis for the initial stages of odor coding [18]. Therefore, we analyzed antennal transcriptomes of workers and males for both C. floridanus and H. saltator, to identify chemosensory receptor genes that are differentially expressed between castes (minors and majors in C. floridanus) and between different sexes, and which might play salient roles in social communication (see Table S2 for information on transcriptome datasets).
We performed pairwise comparisons between males and females within C. floridanus and H. saltator (Dataset S5). At the whole transcriptome level, there was a very high similarity between major and minor worker of C. floridanus (r2 = 0.99; Figure S10A), while greater diversity was found between workers and males (r2 values around 0.85 for all comparisons), largely due to mild up-regulation of many genes in males (Figure S10B, S10C). Similar trends were also observed for chemosensory receptor genes (Figure S10D).
In order to validate our bioinformatic annotations and in an attempt to link functional data to the antennal expression data, we have cloned a small subset of 14 C. floridanus and H. saltator Or genes, drawn from 6 subfamilies in the Or phylogeny (D, E, H, L, V, and 9-exon). These include four genes (CfOr263, HsOr212, HsOr213, and HsOr279) that display significant differential expression in our transcriptome analysis (see Methods and Materials for full list). This allowed us to carry out deorphanization studies to decipher the odorant response profiles of these receptors through the use of two-electrode voltage clamp recordings in Xenopus oocytes heterologously expressing ant Ors [44], [66]. After first confirming that the C. floridanus and H. saltator Orco proteins showed coreceptor function in combination with a previously deorphanized mosquito tuning Or (Figure 8A, 8B), candidate ant tuning Ors were screened against a panel of 73 unitary and complex stimuli (Table S4). These stimuli consisted of a variety of general odorants, as well as hydrocarbons known to be produced by H. saltator or C. floridanus.
Out of the 14 tuning Ors initially screened, CfOr263 (from OR subfamily D; Figure 2), which is highly expressed in workers as compared to males (Figure 6A), produced specific and dose-dependent responses to 2,4,5-trimethylthiazole (Figure 8D, 8F), a naturally occurring odorant found in cooked beef and pork [67] found in the library of general odorants that we screened. An additional Or from H. saltator, HsOr55 (from OR subfamily L; Figure 2), showed a dose-dependent response to another odorant from our general odorant library, 4-methoxyphenylacetone (Figure 8E, 8G), which is a naturally occurring odorant found in anise essential oil [68]. However, this particular Or has not been shown to be differentially expressed between males and females. It should also be noted that, as is the case for most ant Ors, both receptors have multiple closely related homologs that may possess similar chemosensory functions (Figure 6A).
We have developed and used a dedicated annotation scheme to comprehensively elucidate the repertoire of chemosensory receptor genes in both C. floridanus and H. saltator. Through exhaustive homology search and careful manual curation, we significantly improved upon previous studies to identify roughly equivalent numbers of Or/Gr/Ir genes in the genomes of C. floridanus and H. saltator as compared to two other sequenced ant genomes [47], [48], providing a solid foundation for subsequent study.
It is striking that, in general, ants have the most expanded repertoire of chemosensory receptor genes in Hymenoptera (Figure 1). The numbers of ant OR and IR family members are much greater than those of the other two hymenopteran genomes currently available. Indeed, thus far, ant genomes have the largest number of Or genes among all insects [69]. Furthermore, although the number of the Gr genes varies greatly among hymenopterans and also within ants, L. humile carries the largest Gr family; it has about 2- and 10-fold more Grs than N. vitripennis and A. mellifera, respectively. Interestingly, although ants and honey bees are both social insects, ants have much larger repertoires of all three chemosensory receptor gene families than honey bees, possibly indicative of a more sophisticated communication system relying on chemicals [70].
Our phylogenetic analyses of hymenopteran chemosensory receptor genes reveal distinct evolutionary patterns among gene families. Among chemosensory receptors, the OR family shows the most dramatic birth-and-death evolution, with many OR subfamilies displaying diversified patterns of gene gain-and-loss. For example, the 9-exon subfamily and others have experienced rapid gene duplications at almost all stages of Hymenoptera evolution, followed by numerous losses of duplicates. In contrast, there are 35 subclades that have only one ortholog in all four ants. Further, the IR family has maintained relatively stable copy numbers in ants; lineage-specific expansion only occurred in C. floridanus and L. humile for two of the 13 “divergent IRs”. In between these extremes is the GR family that has expanded moderately in N. vitripennis and three of the four ants.
Recent studies of chemosensory receptors in mammals and Drosophila, as well as other genes with important regulatory and physiological functions, have suggested a possible correlation between functional requirements and the variations of gene numbers [52], [71], [72]. Genes with conserved roles tend to have relatively stable copy numbers while those with diversified functions have higher rates of birth-and-death, although the degrees of copy number changes are somewhat random. Our results suggest that this pattern could also hold true for the evolution of the hymenopteran chemosensory receptor genes. For example, as an obligatory co-receptor for all other Ors [29], Orco is the most conserved insect Or gene and also the only one that has maintained unambiguous orthology in all insects studied to date, including ants [69]. Similarly, orthologs of most “antennal IRs” [26] have also maintained strict single-copy in Hymenoptera. It has been proposed that these conserved “antennal IRs” represent the earliest insect chemosensory receptors and perform functions important for all insects [26]. Therefore, we suggest that the chemosensory receptor genes that have constant copy numbers in ants (e.g. the 35 single-copy tuning Ors) are likely to carry out important functions common for all ants.
On the other hand, prevalent rapid expansions in chemosensory receptor gene families could allow for diversification in ligand specificity/sensitivity among duplicated receptor genes. Such functional divergences would offer tremendous opportunities for organisms to explore different chemical niches, thus facilitating the adaption to new environments and/or the evolution of novel life styles such as sociality. In all three gene families, we found either retention of the complete ancestral repertoire (according to the ML method) or dramatic increases in gene numbers (according to the parsimony method) in the ancestor of ants (Figure 5), which might have contributed to the success and subsequent diversification of this group.
In addition, there are many cases of unbalanced expansions/contractions among lineages in specific (sub-)families, suggesting that the chemosensory receptor repertoire has been differentially exploited among ants, which might shed light on the evolution of different lifestyles of ants. For example, our results indicate expansions of Grs in C. floridanus, L. humile, and P. barbatus, but not H. saltator, which are likely to reflect differences in their feeding behaviors. In this view, scavengers like C. floridanus might require a highly expanded repertoire of taste receptors to discriminate nutritious food sources from spoiled, contaminated, or poisoned substrates. In contrast, H. saltator workers likely rely more on visual cues to track down prey, as suggested by their large eyes and expanded number of ommatidia [73]. Furthermore, Grs which act as contact chemoreceptors would be far less useful for identifying and capturing prey. In fact, ponerine ants in general rarely use liquid food sources, since they normally lack the ability to exchange liquids stored in their crop [74] which further reduces the potential benefit of a large Gr repertoire.
Another intriguing possibility is that Grs are involved in the contact chemosensation of species-specific, nonvolatile CHCs (e.g. queen pheromone, nestmate recognition signals, etc.), and that C. floridanus has more Grs precisely because they utilize a greater number and variety of pheromones to support their more rigid and complex social lifestyle. Presumably, these Grs would be in addition to the large number of worker enhanced Ors that are likely to be involved in the same process. Furthermore, C. floridanus has expansions in multiple GR subfamilies, including 5 homologs of the DmGr43a/AmGr3 gene, which has been recently shown to be a fructose receptor [75]. Taken together, our results indicate a correlation between the expanded GR family and the more complex chemical ecology of C. floridanus.
The antenna is perhaps the most important chemosensory organ for ants, where a variety of ant species have been observed to closely inspect their environment and each other by touching their antennae in a process known as antennation [2]. This makes it likely that most of the behaviorally important chemosensory neurons (and their corresponding chemosensory receptors) are located in this organ. Our comparative analysis of antennal transcriptomes of workers and males in both C. floridanus and H. saltator reveal differential expressions of chemosensory receptor genes both within and between species, providing important clues on their functional divergence.
One major pattern revealed by our results is the substantial sexual dimorphism in chemosensory receptor gene expression in ants. For both C. floridanus and H. saltator, almost all Ors were expressed in workers, but only one third were expressed in male. Similarly, workers consistently had more expressed Grs and Irs than males. In contrast, expression of chemosensory receptor genes was highly similar between major and minor workers in C. floridanus. Previous studies have shown that the antennal lobes of males from both C. floridanus and H. saltator lack a large subset of glomeruli relative to workers [76], [77], [78], which may explain the low number of chemosensory receptor genes expressed in males. Given that the number of glomeruli in insects generally correlates with the number of functional odorant receptors [18], [65], it is likely that most of the Ors that are only expressed in C. floridanus and H. saltator workers project to these female-specific glomeruli. Furthermore, it has been shown in another Camponotus species (Camponotus japanicus) that females exclusively possess the olfactory sensilla necessary to detect non-nestmate CHCs, [79], [80]. It is therefore likely that the CHCs receptors are encoded by some of the worker-specific Ors in C. floridanus. In particular, the 9-exon subfamily represents the largest expansion of Ors in all ants and it harbors close to 100 worker-specific Ors in both C. floridanus and H. saltator. These results strongly support previous hypothesis that members of the 9-exon subfamily are likely candidates for ant CHCs receptors [47], [48]. These Ors are potentially involved in detecting CHCs involved in worker-to-worker or worker-to-queen intracolonial social communication.
Interestingly, we also noticed discrepancies between the overall number of Ors and the number of glomeruli in the adults of these two ant species. H. saltator workers and males both have far more expressed Ors than the number of glomeruli in the adult antennal lobe (approximately 78 in the adult male and 178 in the adult worker [77]). The discrepancy in H. saltator could possibly be the result of co-expression of multiple tuning Ors in the same ORN and/or the projection of ORNs expressing different, but related tuning Ors to the same glomerulus, which have both been observed for a small number of Ors/ORNs in D. melanogaster [81], [82], [83], [84].
However, given that the number of expressed Ors is about twice the number of observed glomeruli, this would mean that each glomerulus received input from, on average, two odorant receptors. Although co-expression of tuning Ors has not been observed to such a broad extent in any insect olfactory system studied to date, it should be noted that many of the receptor pairs that are co-expressed in Drosophila appear to be the result of tandem duplication events [84]. Therefore, it is possible that the extensive tandem duplication of H. saltator Or genes may also result in the co-expression of closely related odorant receptors from the same clusters. All of these are highly interesting hypotheses that may be examined in future studies.
In contrast to H. saltator, C. floridanus has approximately 80 fewer Ors than the number of adult worker glomeruli (about 454 [76]). In this instance it is possible that many of those glomeruli receive projections from Gr and Ir expressing ORNs, as there is precedence for this in Drosophila [24], [43] and the number of predicted Grs and Irs would be enough to fill the gap. Moreover, it could be that several Ors have been missed by the current analysis due to incomplete genome assembly; some of the fragmented Or gene models might represent genuine genes, and further genomic/transcriptomic data would help address this possibility.
Although chemosensory receptor genes in general had higher expression in workers, our studies have nevertheless identified a single Or (CfOr267, in subfamily 9-exon) and a single Ir (CfIR8a) in C. floridanus, as well as 4 Ors (HsOr32, HsOr35, and HsOr37, in subfamily L; and HsOr224, in subfamily E) and 2 Irs (HsIR8a and HsIR75u.2) in H. saltator that were significantly male-enriched. The male-enrichment of a receptor gene could be due to elevated expression of the gene in ORNs of males relative to workers, and/or increased number of ORNs expressing the gene in males. No matter which of the possibilities is indeed the case, our results indicate higher overall abundances of these chemosensory receptor genes in male antennae. These genes are viable candidates for receptors that are specifically tuned for male-specific social cues, including queen pheromones. In fact, at least one male-specific honeybee odorant receptor that responds to a queen-specific pheromone has already been revealed through microarray analysis and subsequent functional characterization in Xenopus oocytes [85]. It would not be surprising to see that similar results will be found with the male-enriched ant Ors.
In insects, the co-receptors IR8a and IR25a are the two most conserved Irs [26]. Although a systematic profiling of sexual dimorphic Ir expression is still lacking, a previous study has shown that the Anopheles gambiae orthologs of both IR8a and IR25a have higher expression in female than male [49]. Interestingly, IR8a was the most male-enriched Ir in both C. floridanus and H. saltator. While IR25a also displayed higher expression in C. floridanus male, it was not expressed in the male of H. saltator. These results could possibly indicate a functional divergence of IR8a and IR25a between Diptera and Hymenoptera. In addition, the high expression of IR25a in males of C. floridanus, but not H. saltator, suggests that IR25a-mediated signaling might have contributed to the more expanded roles for males within the colony of the former species. It may be that C. floridanus males are more involved in intracolonial interactions than H. saltator males, since males from other Camponotus species are known to participate in food exchange in the colony [86], which has not observed in H. saltator males.
We have also found diversified expression of closely related Ors within and between species. For example, in the basal clades of the 9-exon OR subfamily, closely related C. floridanus and H. saltator Ors showed opposite sexual dimorphism in their expression (Figure 6B). Although the well-supported monophyletic clade within the 9-exon OR subfamily mostly consists of worker-enriched genes, it also harbors a few genes that are highly enriched in male (Figure 6C). Thus, while our expression results are generally (and strongly) consistent with the idea that members of the 9-exon OR subfamily are involved in the detection of CHCs by workers [47], a subset of these receptors have apparently been adapted for use in males, possibly for detecting queen mating pheromones.
Taken together, these results indicate that ant Or genes have experienced not only extensive gain-and-loss, but also rapid changes in their expression, once again highlighting the highly dynamic nature of chemosensory receptor gene evolution. Our phylogenetic and transcriptomic analyses, in combination, have identified ant chemosensory receptor genes that exhibit evolutionary and expression patterns indicative of species/sex-specific functions. Ultimately, deorphanization of these receptors will greatly facilitate our understanding of the chemical ecology of social lifestyle in ants.
In our heterologous studies of ant tuning Ors, we have identified chemical agonists for a single receptor from each of the two species analyzed. These data provide conclusive validations for our bioinformatic-based annotations. Although a honeybee odorant receptor has been previously shown to respond to the queen substance 9-oxo-2-decenoic acid [85], we believe that this represents the first published report of ligand activators for odorant receptors from ants.
In these studies, HsOr55 from H. saltator, display significant responses to 4-methoxyphenylacetone, a naturally occurring odorant found in anise essential oil [68]. Since anise essential oil has been shown to have a repellent and/or insecticidal effect on at least some species of insects [87], [88], 4-methoxyphenylacetone might represent a general insect repellent, with HsOr55 acting as the detector for this repellent in H. saltator. Whatever HsOr55's role may be, it is likely to be a very general one, since HsOr55 transcripts do not appear to be differentially expressed between workers and males.
The other odorant receptor characterized in this study, CfOr263 from C. floridanus, displayed sensitivity to 2,4,5-trimethylethiazole, a naturally occurring odorant found in cooked beef and pork [67] that has been previously shown to induce strong responses in the CpC neuron of the maxillary palp in the mosquito Anopheles gambiae [44]. While the relevance of this chemical to C. floridanus remains unclear, the fact that CfOr263 transcripts are enriched in workers relative to males suggests that this odorant may be an important volatile semiochemical for C. floridanus workers. Regardless, the successful identification of odors that activate CfOr263 and HsOr55 strongly validates the role of ant Ors as chemosensory receptors. Furthermore, the large differential expression of CfOr263 between workers and males indicates that it is detecting a sex- specific signal that is relevant to workers but not to males, and testing a broader panel of odorants in the future will provide a better understanding of what that signal might be.
We have revealed a greatly expanded repertoire of chemosensory receptor genes for a pair of divergent ant species, including about 400 Ors and an order of magnitude smaller number of Grs and Irs. Phylogenetic analysis of these newly annotated genes indicates that there are likely to be vast differences in the importance of particular chemoreceptor families and subfamilies between the four ant species examined, which is likely to reflect the variety of ecological and social demands experienced the members of each species. These analyses also reveal high rates of gene birth-and-death evolution among the olfactory and gustatory receptor genes, suggesting that some factor (such as changes in the complex CHC profiles that control ant social behavior) is driving rapid evolution in their chemical response profiles. The large repertoire of ant chemosensory genes might be either due to preferential retention of ancestral genes or rapid expansions in the ant ancestor and during later stages of ant evolution. To further complement these phylogenetic results, we have generated and analyzed antennal-specific RNAseq expression data to identify ∼40 C. floridanus and ∼120 H. saltator chemosensory receptors that exhibit significant sexual dimorphism in expression. This expression data has, in turn, informed studies towards the identification of odorant ligands for socially relevant receptors, a process that we have already successfully accomplished in a heterologous system for one of the differentially expressed C. floridanus Ors. Taken together, our evolutionary analysis, transcriptome profiling, and heterologous characterization provide new insights into the roles of the chemosensory receptors in inter-sex behavioral and social differences of ants.
The assemblies of C. floridanus (version 3.5) and H. saltator (version 3.5) were downloaded from the Hymenoptera Genome Database [89]. Protein sequences of reported chemosensory gene were also collected from Apis mellifera, Acyrthosiphon pisum, Drosophila melanogaster, Nasonia vitripennis, L. humile, and P. barbatus [15], [26], [28], [47], [48], [50], [54]. An in-house bioinformatics pipeline was developed to identify candidate chemosensory genes in C. floridanus and H. saltator. First, all collected chemosensory gene sequences were searched against the two ant genomes using TBLASTN [90] with an e-value cutoff of 1e-5. Resulting High-scoring Segment Pairs (HSPs) were sorted by their blast bit-scores, and an average bit-score of the top 75% HSPs were calculated. Any HSPs with a bit-score less than 25% of the average was discarded. Chains of HSPs were than created from retained HSPs. Two HSPs were chained together if the following criteria were met: 1) they are derived from the same query; 2) they are located within 3 kb on the same strand of a scaffold/contig; and 3) the corresponding query region of the upstream HSPs must also be N-terminal to that of the downstream HSPs. The third criterion was applied to avoid artificial concatenation of neighboring chemosensory genes. Genomic regions covered by HSPs chains were considered putative chemosensory gene coding regions. For each putative gene, we then selected the query corresponding to the highest scoring HSPs at that region as reference sequence for homology-based gene prediction using GeneWise (version 2.2.0) [91]. All predictions were sorted by ORF length and the lowest 25% was filtered. This pipeline was iterated by adding results of previous run to input until no additional genes were found.
Multiple sequence alignments (MSAs) of predicted OR/GR/IRs were constructed using MUSCLE (version 3.8) [92] and manually inspected. Attempts to improve annotations were made whenever an obvious problem was identified (e.g. missing exon, incorrect exon-exon junction). In addition, in the OR and GR families, we observed many fragmented gene models, likely due to pseudogenization and incomplete genome assembly. For the convenience of subsequent analyses, a minimum size cutoff of 300 amino acids was used for the ORs and GRs. For IRs, we screened all predicted protein sequences with InterProScan (V4.8) [93] and filtered the ones without characteristic domains of IR (PF10613 and PF00060) [26].
We included in our phylogenetic analysis chemosensory receptor genes in six hymenopteran species, including A. mellifera, C. floridanus, H. saltator, N. vitripennis, L. humile, and P. barbatus. For each of the OR/GR/IR families, all family members were firstly aligned at once using MUSCLE (version 3.8) and a preliminary phylogenetic tree was built using RAxML (version 7.2.8) [94]. Sequences were then divided into groups corresponding to highly supported clades in the preliminary phylogeny. Groups were aligned individually using PROBALIGN (version 1.4) [95] and then combined together using the profile alignment function of MUSCLE. The complete alignment were further manually inspected and adjusted using GeneDoc (version 2.6) [96]. In addition, poorly aligned regions in the alignment were removed using trimAl (version 1.4) [97]. The final maximum-likelihood tree was constructed using RAxML with Le-Gascuel (LG) substitution model [98] and GAMMA correction for rate variation among sites. Reliability of tree topology was evaluated by 100 bootstrap replicates. To estimate the number of gene gain and loss events, we used a maximum-likelihood based approach implemented in CAFÉ (version 2.2) [62] with default settings. As an alternative approach, we also used the parsimony based “modified reconciliation method” [99]; we first collapsed branches with bootstrap support lower than 70 in phylogenies of OR/GR/IR families and then reconciled condensed trees with known organismal relationships using Notung (version 2.6) [63].
Samples originated from C. floridanus colonies that had been founded in the Liebig lab from queens captured in southern Florida between 2002 and 2009 and from H. saltator colonies collected in Karnataka, India between 1995 and 1999. Antennae were collected from each of five groups of adult ants: H. saltator workers and males and C. floridanus major workers, minor workers, and males. Whole ants were flash-frozen in liquid nitrogen and kept on dry ice as 100 antennae from each group were removed with forceps. Antennae were placed directly into RNAlater ICE (Ambion) that had been pre-chilled on dry ice in a conical, ground-glass, tissue homogenizer. RNAlater ICE was replaced with 1 ml Trizol (Invitrogen), in which antennae were homogenized. Total RNA was isolated following Trizol manufacturer instructions; briefly, after addition of 200 µl of a chloroform∶isoamylalcohol mixture (24∶1), each sample was mixed vigorously and the RNA-containing aqueous layer was isolated with centrifugation. RNA was further purified and DNAse-treated with the RNeasy Miniprep kit (Qiagen). After ethanol-precipitation, the RNA pellet was resuspended in 30 µl nuclease-free water. Male samples were sequenced using Illumina HiSeq2000 at the NYULMC Genome Technology Center, generating ∼33 million 50 bp single-end reads for C. floridanus male and ∼164 million 51 bp single-end reads for H. saltator male. All worker samples were sequenced at Hudson Alpha, generating more than 20 million 50 bp paired-end reads for each sample (sum of two technical replicates).
Reads of C. floridanus male sample were trimmed to 34 bp (8 bp trimmed from both ends) to remove low-quality positions. In addition, for all worker datasets, we treated each paired-end read as two single-end reads. Therefore, all datasets in our subsequent analyses consist of only single-end reads. Alternative strategies for data processing led to highly similar estimations of gene expression values (Table S5). For each dataset, reads were mapped to the corresponding ant genome using TopHat (version 1.3.3) [100] with default setting. Gene annotations for C. floridanus (version 3.5) and H. saltator (version 3.5) were downloaded from the Hymenoptera Genome Database and used in combination with our annotation of chemosensory genes to guide the reads mapping. Gene expression levels (in FPKM values) and differentially expressed genes were determined using Cuffdiff v1.3.0 [101] with frag-bias-correct, multi-read-correct, and upper-quartile-norm options turned on.
Predicted Or coding sequences were amplified, by PCR, from H. saltator and C. floridanus worker antennal cDNA samples obtained from colonies established at Arizona State University (Tempe, AZ). The PCR-amplified sequences were then TOPO cloned into the Gateway Entry vector pENTR/D-TOPO (Life Technologies), followed by an additional cloning step into a destination vector derived from pSP64T. To obtain cRNA for each Or, the pSP64T vector containing the appropriate coding sequence was linearized by restriction digest and used as a template for cRNA synthesis using the mMessage mMachine Sp6 Kit (Ambion). Heterologous expression of ORs was accomplished as described previously [66]. Briefly, mature oocytes were surgically extracted from Xenopus leavis adult females, treated with 2 mg/mL collagenase II in 1× Ringer's solution (96 mM NaCl, 2 mM KCl, 5 mM MgCl2, and 5 mM Hepes, pH 7.6) for 30–45 minutes at room temperature, and then injected with 27.6 nL of a 1∶1 mixture (by mass) of a given tuning Or in combination with the appropriate Orco ortholog (either HsOrco or CfOrco). After injection, oocytes were stored in Incubation Medium (10% dialyzed horse serum in 1× Ringer's solution) at 18C for 3–7 days before testing. Responses to odorants were measured by recording whole-cell currents in Clampex 10.2 (Molecular Devices) using a two-electrode voltage-clamp setup (OC-725C, Warner Instruments) maintained at a −80 mV holding potential. Odorants were first dissolved in DMSO, and then further diluted into Ringer's solution before being introduced to the oocyte recording chamber using a perfusion system. For the hydrocarbons that were tested, 0.01% Triton X-100 (Sigma) was also added to the Ringer's solution to aid in dissolving the odorant. The following odorant receptors were tested with the odorants listed in Table S4: CfOr183, CfOr215, CfOr263, HsOr19, HsOr55, HsOr132, HsOr170, HsOr175, HsOr212, HsOr213, HsOr234, HsOr239, HsOr279, HsOr287.
Odorant chemicals were purchased from commercial sources at the highest purity available. Henkel 100, a mixture of 100 different volatile chemicals, was obtained from Henkel (Düsseldorf, Germany), and the C7–C40 saturated alkane mixture was purchased from Supelco (Bellefonte, PA, USA).
|
10.1371/journal.pntd.0001024 | Strategies for Introducing Wolbachia to Reduce Transmission of Mosquito-Borne Diseases | Certain strains of the endosymbiont Wolbachia have the potential to lower the vectorial capacity of mosquito populations and assist in controlling a number of mosquito-borne diseases. An important consideration when introducing Wolbachia-carrying mosquitoes into natural populations is the minimisation of any transient increase in disease risk or biting nuisance. This may be achieved by predominantly releasing male mosquitoes. To explore this, we use a sex-structured model of Wolbachia-mosquito interactions. We first show that Wolbachia spread can be initiated with very few infected females provided the infection frequency in males exceeds a threshold. We then consider realistic introduction scenarios involving the release of batches of infected mosquitoes, incorporating seasonal fluctuations in population size. For a range of assumptions about mosquito population dynamics we find that male-biased releases allow the infection to spread after the introduction of low numbers of females, many fewer than with equal sex-ratio releases. We extend the model to estimate the transmission rate of a mosquito-borne pathogen over the course of Wolbachia establishment. For a range of release strategies we demonstrate that male-biased release of Wolbachia-infected mosquitoes can cause substantial transmission reductions without transiently increasing disease risk. The results show the importance of including mosquito population dynamics in studying Wolbachia spread and that male-biased releases can be an effective and safe way of rapidly establishing the symbiont in mosquito populations.
| Wolbachia are symbiotic bacteria that are found in many insect species. Recent laboratory studies show that certain strains of Wolbachia can reduce the capacity of mosquito species to transmit diseases such as dengue fever and malaria, either by directly inhibiting the pathogen or by shortening lifespan. However, little is known about how easily these bacteria will spread in natural mosquito populations or the impact of deliberate Wolbachia introduction on disease transmission. We use a simple model of Wolbachia-mosquito interactions to explore the design of field releases of infected mosquitoes to initiate symbiont spread. A particular concern is how Wolbachia can be introduced while releasing only small numbers of female mosquitoes which may bite humans and transmit disease. The models include explicit mosquito population dynamics including seasonal fluctuations in population size and different forms of population regulation. We find that rapid Wolbachia establishment is possible by releasing predominantly male mosquitoes, though the number of insects introduced may need to be large. This strategy requires the introduction of considerably fewer females compared to equal sex-ratio releases and is unlikely to increase disease transmission throughout the intervention. We demonstrate that once Wolbachia has become established, substantial reductions in disease transmission are possible.
| Mosquito-borne parasites and viruses cause some of the world's most important diseases, disproportionately affecting poor communities and representing a major public health challenge. Biological control techniques aimed at suppressing mosquito populations or reducing their capacity to transmit disease may be a useful addition to traditional vector control strategies, especially if resistance to chemical insecticides in mosquito populations continues to rise [1]. Recently there has been increased interest in the use of certain strains of Wolbachia bacteria to reduce transmission by mosquito vectors of human diseases [2]–[7]. Wolbachia are maternally-inherited endosymbiotic bacteria that are common in many insect species including mosquitoes. Wolbachia spread in mosquito populations by manipulating the host's reproduction using a mechanism known as cytoplasmic incompatibility (CI) [8]. CI occurs when Wolbachia in infected males modify the sperm of their host such that arrest of embryonic development occurs unless the egg also carries the bacterium. Uninfected females are therefore at a disadvantage, and the Wolbachia spreads by a process of positive frequency-dependent selection. Models of Wolbachia dynamics show that spread will occur if the proportion of infected hosts exceeds a threshold that is higher for Wolbachia that cause stronger reductions in host fitness [9].
Recent studies indicate that infecting mosquito populations with certain strains of Wolbachia may lower their rates of disease transmission for two reasons. First, the bacteria may reduce mean adult lifespan [6], [10]. Because most vector-borne pathogens have a relatively long extrinsic incubation period in the mosquito a reduction in average longevity disproportionately affects infectious individuals, with beneficial consequences for disease transmission [11], [12]. However, a reduction in longevity also lowers the fitness of Wolbachia carriers and hence increases the threshold infection frequency required for spread to occur [13]. An ideal strain would increase mortality only late in life as this would (i) particularly affect pathogen-carrying individuals; (ii) have a lesser effect on host fitness and thus require fewer individuals to be introduced to pass the threshold infection frequency; and (iii) lead to less selection for modulation of the harmful effects of these Wolbachia. Second, Wolbachia can inhibit the development, replication or dissemination of important mosquito-borne pathogens, including filarial nematode parasites [5] and dengue and chikungunya viruses in Aedes aegypti [2], [7], and Plasmodium malaria parasites in Aedes aegypti [7] and Anopheles gambiae [5]. The capacity of Wolbachia-infected mosquitoes to transmit these diseases may thus be much reduced.
However, the ability of Wolbachia to assist in the control of mosquito-borne diseases will depend on their dynamics in natural mosquito populations. Understanding the ecology of Wolbachia infections in mosquito populations is important as programmes to establish Wolbachia in wild Ae. aegypti are currently under consideration [3]. Recently we developed a modelling framework that allows the spread of Wolbachia that reduce the longevity of their insect hosts to be analysed [14]. The models allow the study of the demographic consequences of releasing the significant numbers of individuals often needed to breach the threshold for Wolbachia to spread. They can be used to explore different schedules of Wolbachia introduction (for example few large or many small introductions of infected insects), the effects of different types of density-dependent mortality in the host population on Wolbachia dynamics and the timing of introductions in a seasonal environment.
Here we employ this modelling approach to investigate practical questions concerning the use of Wolbachia for mosquito-borne disease management. Because female mosquitoes bite people and so constitute a nuisance, and because they can potentially transmit disease, it is desirable that only a minimum number of female insects are released as part of a Wolbachia introduction. This may be possible by applying methods of sex separation by pupal size sorting to reared insects to create releases with a highly male-biased sex-ratio [15], [16]. We develop theory for male-biased release strategies and explore their feasibility and how releases may be optimised when the mosquito population size shows strong seasonal fluctuations. We then extend the model to include a simple representation of a mosquito-borne disease. The model is sufficiently general to represent a wide range of mosquito species and the diseases they transmit; here we chose parameters derived from the literature on Anopheles mosquitoes for illustration. This is used to estimate how the rate of disease transmission changes over time following male-biased Wolbachia releases. Different assumptions about mosquito population dynamics and the effects of Wolbachia on vectorial capacity are explored.
The model of mosquito and Wolbachia dynamics used here is an extension of that in [14] with separate adult sexes and the inclusion of egg and pupal stages (Figure 1). It is phrased as a system of integral equations describing the numbers of infected and uninfected larvae and adults of different ages; full details are given in Text S1.
The mosquito life cycle is divided into three juvenile stages (egg, larva and pupa) and an adult stage. The population is assumed to be regulated by density-dependent mortality experienced during the larval stage described by a power function, , where is larval density and and β are constants. Higher values of the parameter β denote a steeper response to increasing density (which we shall refer to as strong density dependence). Mortality in adults is assumed to be age-dependent and is modelled by a Weibull function whose parameters may depend on infection status (see Text S3 and [14]). Adult fecundity is assumed to be constant with age (but see the Discussion).
Wolbachia may increase adult mortality, particularly in older age-classes, and the proportional reduction in average adult longevity caused by Wolbachia is denoted sg. Wolbachia-infected individuals may also have reduced fecundity (by a proportion sf). Mating is assumed to occur at random, and an uninfected female mating with an infected male will lose a fraction sh of her offspring. Infected females fail to transmit Wolbachia to their offspring with probability . We assume here that Wolbachia does not affect survival during, or length of, the juvenile stages.
For a closed population (no immigration, deliberate introduction, or emigration), the position of the equilibrium threshold frequency above which Wolbachia spreads through the population depends on the magnitude of the fitness effects of the bacterium on its host, and the probability of non-transmission (see Figure 2). For the basic model analysed here, Hancock et al. [14] showed that the threshold frequency p*is(1)where and . This expression is closely related to the classic condition for spread derived for discrete-generation, purely genetic models by Turelli and Hoffmann [17].
Mosquito populations are very sensitive to patterns of seasonal rainfall, and often show strong annual fluctuations in abundance [18]–[20]. We model this by assuming that larval carrying capacity (the parameter in the expression for larval density dependent mortality) varies over the year. Two seasonal abundance patterns are considered which we refer to as A and B. These patterns were chosen to represent attributes of mosquito population dynamics that we have found to be important to Wolbachia spread; strong temporal variation in adult abundance and varying rates of seasonal population growth and decline. In pattern A there is a six-month season of high mosquito abundance generated by setting = 0.05 for six consecutive months and = 0.1 for the rest of the year (Figure 3A; solid line). In pattern B the seasonal increase and decline in mosquito abundance is more gradual (Figure 3B; solid line). This pattern is produced by setting the larval carrying capacity to α = 0.055, 0.055, 0.05, 0.05, 0.053 and 0.06 respectively for the six months of the year when mosquitoes are abundant and = 0.1 otherwise.
In exploring different release strategies, for operational reasons we restrict deliberate introductions to the wet season. Although the size of the resident population is lowest in the dry season, which would appear to facilitate population replacement, Anopheles and other mosquitoes are highly sensitive to desiccation. Mosquitoes may aestivate during the dry season [19], or rest in microhabitats with higher than average humidity. It is likely that if introductions were made at this time of year any introduced mosquitoes would experience very high mortality before locating relatively rare, suitable resting sites (or conspecifics with which to mate).
We extend the age-structured model of mosquito and Wolbachia dynamics to include a vector-borne pathogen. The infected and uninfected adult classes are divided into susceptible, exposed and infectious (SEI) stages, with the exposed stage assumed to be of fixed duration (the extrinsic incubation period). A fraction x of the human population is assumed to be infectious, and this parameter as well as the total number of humans is assumed to be constant over time. The model makes assumptions about the frequency of blood feeding, and the probability of the mosquito being infected during a blood meal. Our treatment of the adult stages is based on Hancock et al. [21] (a model of the interaction between Anopheles, Plasmodium and a pathogenic fungus). Full details of the model are given in Text S2. We assume here that a proportion cw of mosquitoes that are infected with both Wolbachia and the pathogen do not become infectious. This represents the reduction in disease transmission that has been shown to occur in mosquitoes infected with Wolbachia.
A critical quantity in the epidemiology of mosquito-borne diseases is the Entomological Inoculation Rate (EIR), the number of bites on humans by infectious mosquitoes per person per day. This is simply the total number of infectious mosquitoes (both carrying and not carrying Wolbachia) per human multiplied by the daily rate of biting [22]. Analytical expressions for the equilibrium EIR can be obtained for a constant environment where Wolbachia is absent or at equilibrium frequencies (Text S2).
The different parameters included in the models and their default values are shown in Table 1. Parameter estimates obtained in the field for Anopheles mosquitoes have been used where possible, though for some such as those governing density dependent mortality little information is available. Data on age-dependent mortality rates of laboratory colonies of Anopheles were used to parameterise the Weibull function describing adult age-dependent mortality [23]. We assume that mosquitoes experience additional age-independent background mortality at rates observed in field populations (see Text S3).
Parameters describing the effect of Wolbachia on longevity derive from field cage studies of the life-shortening Wolbachia strain wMelPop infecting Ae. aegypti [6]. We calculated the age-dependent increase in the rate of adult mortality caused by wMelPop infection in Ae. aegypti and assumed that it would have a similar proportional effect on Anopheles (Text S3). Mosquitoes in cages tend to live longer than those in nature and this can lead to overestimation of the fitness consequences of late-acting mortality. Including background field mortality, the overall reduction in average adult lifespan caused by Wolbachia infection is assumed to be 16% (sg = 0.16) (Text S3). Both strains of Wolbachia that have to date been successfully introduced into Ae. aegypti, wMelPop and wAlbB, inhibit the development and transmission of human pathogens in this host [2], [4], [7], and unlike wMelPop the wAlbB transinfection had no observable impact on longevity in the lab [2]. We also explore the effects of introducing a Wolbachia that causes a 5% reduction in adult lifespan in the field (sg = 0.05). We address the implications of our imperfect knowledge of different parameters in the Results and Discussion.
When transmission is perfect (), Wolbachia spreads when it reaches an infection frequency in the population such that an average infected female has more offspring than an average uninfected female. The latter are disadvantaged through mating with infected males which causes them to lose a fraction sh of their offspring. This picture is slightly more complex when transmission is not perfect, or when immigration, introductions or emigration are occurring [14], but again spread is caused by the presence of infected males giving an indirect, relative advantage to Wolbachia-bearing females. This advantage can be made greater simply by increasing male (and not female) infection frequency. Of course infected females must be present for the infection (which is not transmitted through males) to spread, but once the threshold is exceeded the frequency in females will increase from an arbitrarily low start. An infection can thus be established even though relatively few females are released.
A simple way to model sex-biased releases is to assume that newly-emerged infected males and females are introduced into an uninfected population at constant rates IM and IF. For simplicity we assume that the larval carrying capacity does not vary with time (no seasonality) and that the uninfected population is at equilibrium prior to the introduction. Figure 2 illustrates how introducing males at a relatively high rate allows the infection to invade when females are introduced at a much lower rate (1% of the rate of male introduction). The time it takes for the infection to be established is longer when introduction rates are low.
When the rate at which infected females are introduced is very small (), it is possible to calculate the unstable equilibrium male infection frequency above which Wolbachia spreads, and the threshold rate of male introduction required to exceed this frequency (Figure 2 and Text S1). However the expressions are complicated because the introduction of infected males reduces the fecundity of resident uninfected females and this lowers the density dependent mortality experienced by the juvenile population. The effect of the introduction on the rate of recruitment of uninfected adults will thus depend on the strength of juvenile density dependence. This is illustrated by comparing the threshold male introduction rates required for Wolbachia spread to occur in populations with relatively strong ( = 0.3, = 0.05) and weak density dependence ( = 0.1, = 0.2). Values of the larval carrying capacity were chosen so that the equilibrium adult abundance in the absence of Wolbachia is the same in both cases. The required rate of male introduction is approximately 50% higher in the case of strong (IM = 0.44 day−1) as opposed to the weak (IM = 0.28 day−1) density dependence. This occurs because the reduction in density-dependent mortality caused by the introduction of males is greater when density dependence is stronger, and so less suppression of the (uninfected) adult population occurs. It can thus be important to consider demographic as well as genetic processes in models of Wolbachia dynamics.
A more realistic scenario for the release of Wolbachia-infected mosquitoes is that the insects are released in separate batches rather than continuously, and that mosquito population size fluctuates seasonally. The total and relative numbers of male and female mosquitoes that need to be released for spread to occur were studied in a population whose seasonal dynamics are described by pattern A. We compare releases consisting of equal numbers of the two sexes and 95% males and calculate the minimum numbers that have to be liberated at different times of the season to ensure Wolbachia becomes established. The release strategy we model is of 30 daily releases, each containing the same number of mosquitoes.
The results are shown in Figure 3A. First note that for all strategies releases early in the season when the resident population is small require fewer mosquitoes to be introduced, a result we explore in more detail below. Overall, for any particular release date, the total required release size is 3–4 times larger for the 95% male-biased strategy compared to the equal sex-ratio strategy, and so the number of mosquitoes that must be reared (prior to separation of the sexes) assuming a 50∶50 sex-ratio is 6–8 times greater. However, although more mosquitoes in total must be produced, fewer females need to be released with the male-biased strategy. In the present example, which is typical of others we have explored, the total number of females introduced is approximately ⅓–½ the numbers required in the equal sex-ratio strategy.
There are two reasons why the male-biased strategy requires the release of fewer females. First, releasing a large number of males causes a high frequency of incompatible matings and so reduces the size of the resident (uninfected) population. Second, the high frequency of infected males means that infected females have a strong relative fitness advantage. However, the dynamics of releasing mosquitoes in a finite number of separate batches are not the same as those assuming introduction at a constant rate. Figure 4 shows the male infection frequency as a function of time over a 3 year period following 30 daily 95% male releases made in the second month of the season of high mosquito abundance. Although male infection frequencies are initially very high they decline rapidly after the final release as the introduced males die. At this stage there are still relatively few infected females present and hence recruitment of Wolbachia-carrying individuals is low. To prevent the loss of Wolbachia in this transient period, the releases must attain a temporary male infection frequency that is considerably higher than the threshold calculated in the continuous release case. Enough females must also be introduced so that they produce sufficient infected sons that the male infection frequency does not fall below the threshold following the final release.
As in the case of continuous release, we found that the minimum required number of insects for Wolbachia establishment depended on the assumed form of juvenile density-dependent mortality. Further details are given in Text S4 but we again found that larger releases were needed when density dependence was strong. However, for all the forms of density dependent mortality we studied, consistently ⅓–½ the number of females was required for male-biased (95%) compared to equal sex-ratio releases.
These results suggest that the establishment of Wolbachia using highly male-biased releases is feasible, provided comparatively large numbers of mosquitoes can be reared and the sexes separated. We explore this issue further in the Discussion.
We explored the effect of seasonal variation in mosquito abundance on the release size necessary for Wolbachia to spread for the two seasonal patterns described in the Model Development. Again we assume that 95% of the insects introduced are males, and that 30 daily releases of the same size are made. Figures 3A & B show the minimum required release size for different release times.
In both cases releases early in the season require fewer mosquitoes. At this time the resident population is small and so a fixed number of introduced infected insects constitute a greater proportion of the total. Releases made towards the end of the season when the population is beginning to decrease may also require fewer insects, but this depends on the rate of population decline. In case A the decline is abrupt and the size of releases required for spread increases steadily through the season. Late season releases here are a poor strategy because the decline in larval carrying capacity drastically reduces recruitment to the adult stage so that the proportion of infected individuals is chiefly determined by adult mortality rates that with our parameter assumptions particularly penalise Wolbachia-carrying individuals. However, in case B, where the population declines more gradually, the required release size decreases towards the end of the season.
Insight into the seasonal dynamics of Wolbachia spread following the releases can be gained by plotting the temporal change in male infection frequency for seasonal pattern A (Figure 4). The male infection frequency falls following the final release and then starts to rise again towards the end of the season as the progeny of the first female introductions reach the adult stage. However, the collapse in carrying capacity acts to reduce the infection frequency as the season ends, for the reasons described above. Wolbachia only becomes established if the infection frequency towards the end of the season is high enough that it does not drop below the threshold (eqn. 1) when the population enters the steep decline. This is a further reason why Wolbachia strains that reduce host longevity require large releases before they can become established.
Models of the introduction of Wolbachia with equal numbers of males and females show that the number of releases made can significantly affect the total number of insects that need to be introduced to establish Wolbachia in the population [14]. In particular, introducing large numbers of females at one time that then reproduce can increase the juvenile density-dependent mortality, which disadvantages the progeny of these females. Introducing the insects in a larger number of smaller releases is therefore sometimes more effective, particularly for Wolbachia strains that incur a high fitness cost and thus require large releases to allow spread.
In the case of highly male-biased releases, this effect does not occur, because relatively few females are added and the number of larvae declines due to the high frequency of incompatible matings. However multiple releases may still be beneficial because they prolong the period over which the male infection frequency is artificially elevated, so sustaining the fitness advantage of infected females and allowing their numbers to increase from an initial low level.
For seasonal pattern A, we compared the minimum number of introduced insects required for spread for strategies where different numbers of equal-sized, daily releases are made, again assuming that the sex-ratio of the releases is 95% male (see Text S5). If releases are made towards the middle of the season, after the period of rapid population growth, the total required number of introduced insects is smaller if the insects are distributed across a larger number of batches. This is not the case for releases made close to the start of the season, when there is a slight advantage in releasing the mosquitoes in a single batch. At the start of the season the benefit of prolonging the elevation of the male infection frequency over multiple releases is lost because the population is increasing rapidly and so later releases cause a smaller increase in the proportion infected. These results indicate that seasonal changes in mosquito abundance are a much stronger determinant of the required release size than the number of releases made (Text S5).
Here we examine the effects of Wolbachia introduction on the abundance of female mosquitoes and the rate of disease transmission for a release strategy that introduces the minimum number of mosquitoes required for spread in 30 daily equal-sized batches, and a strategy that releases a larger number, three times the minimum required, in 90 daily equal-sized batches. The sex-ratio of the releases is 95% male. We assume that seasonal abundance dynamics follow pattern A, and that releases are made one month into the season of high mosquito abundance.
Figure 4 shows the daily entomological inoculation rate (EIR), the female population size, and the male infection frequency for a three year period where releases of the minimum size required for spread are made in the second year. During the releases the total number of females (including wild and released individuals), and likewise the EIR, are quickly reduced compared to the level in the previous year, although at the start of the releases there are slightly more females present than there would be in the absence of the intervention. The reduction in both quantities is due mainly to the suppression in population abundance caused by the high frequency of incompatible matings between uninfected females and infected males. The reduction also partly results from the lower fitness of Wolbachia-infected females, although this does not have a strong effect in the year of release because the Wolbachia infection frequency in females remains relatively low. In this example the Wolbachia does not reach its stable frequency until the year following the releases and its establishment results in much greater reduction in EIR than in the female population size (Figure 4). This is because the reduction in longevity brought about by Wolbachia infection causes a disproportionate reduction in the abundance of individuals that live long enough to transmit the pathogen.
We now compare these dynamics to those produced when the total number of insects released is three times the minimum required (Figure 5A). In this case the Wolbachia spreads much more rapidly and reaches its final frequency in the year of release. Although more insects are introduced there is still only a very slight initial increase in the female population size at the start of releases, followed rapidly by a net reduction in both population size and EIR. An advantage of releasing more mosquitoes is that the EIR declines more quickly due to faster Wolbachia spread.
In addition to reducing adult longevity, Wolbachia can also directly inhibit pathogens within the mosquito. The insets in Figure 5 show the combined effects of life-shortening and pathogen inhibition on the EIR once Wolbachia has become established. We assume that Wolbachia reduces average adult longevity by either 16% (Figure 5A) or 5% (Figure 5B). Reducing longevity has a major impact on the EIR but in both cases direct pathogen inhibition gives a further substantial decrease in the EIR. Our results show that a 16% reduction in longevity with no effect on transmission is similar to the joint effect of a 5% reduction in lifespan with a 50% reduction in transmission. However, it would require far fewer mosquitoes to be released to establish a Wolbachia with the second phenotype.
We explored how these conclusions were affected by the nature of the assumed density-dependence (see Text S6). When density-dependence is strong the releases cause less reduction in the female population size, both transiently due to incompatible matings and in the long term due to the fitness costs of Wolbachia infection. However the reduction in the EIR was similar for all the forms of density dependence we considered, particularly in the longer term once the Wolbachia has reached a high infection frequency. This is because the EIR is much more sensitive to changes in adult mortality than to changes in the rate of adult recruitment.
Introducing Wolbachia into mosquito populations can lead to a reduction in the transmission of mosquito-borne diseases. The normal way in which establishment has been envisioned is through the release of equal numbers of male and female mosquitoes [3]. However, as females transmit disease and are responsible for nuisance biting, it is important to minimise the numbers of females released, and this may be critical in obtaining regulatory permission and public support for introductions. It is shown here that establishment can occur following releases composed very largely of males provided this causes the Wolbachia infection frequency in males in the field to exceed a threshold. The numbers of females in the population decline rapidly following the initial male-biased releases, and only for a relatively brief period at the commencement of releases are female numbers slightly higher than they would have been in the absence of the intervention. However, for male-biased releases the numbers of insects that must be reared is considerably higher than when releases are composed of equal numbers of the two sexes, and a reliable method must exist for separating males and females. These may not be major barriers to the strategy. Some mass-rearing facilities have the capacity to produce over 1 million mosquitoes per week [15], which is more than 30 times the estimated size of some village-scale natural mosquito populations [3], [24]. For Aedes aegypti, male and female pupae can be rapidly separated by size and when larval rearing conditions are optimal over 99% males can be achieved [15], [16]. Transgenic sex separation methodologies also exist (e.g. Alphey et al. [15]), although their use would lose the advantages of Wolbachia intervention not involving genetically modified organisms.
However in situations where the number of insects that can be reared and released is more strongly limited, such as in isolated rural areas, it may be preferable to use equal sex-ratio releases which will provide more rapid Wolbachia spread for a given available release size and number. This would improve the likelihood of achieving Wolbachia establishment, for example in the case of unexpected losses of released insects. Equal sex-ratio releases result in considerably less suppression of female numbers as well as of the EIR during the release period compared to male-biased releases. They also lead to higher densities of biting females, though the increase over natural levels is not always very marked (see Text S7). Release programmes that have the capacity for bigger release sizes and greater numbers of releases are likely to gain stronger benefit from using male-biased releases to limit the addition of females and suppress the vector population. Engagement with local communities will also reveal the extent to which the possibility of modest and temporary increases in biting female numbers would be a significant impediment to their support for the program.
A further strategy for reducing the risk of increased biting or disease transmission associated with the introduction of females is artificially to suppress the mosquito population prior to release, for example by insecticidal fogging or larval control. Fewer released mosquitoes would then be required to surpass the threshold infection frequency that allows Wolbachia to spread. Suppression measures would be stopped immediately prior to mosquito release so as to minimise the time for population numbers to rebound, and so as not to affect the introduced insects. The efficacy of different types of pre-release suppression will depend on the population dynamics of the mosquito species, in particular the form of density dependence, and can be explored using the type of model developed here (see Text S8 for examples).
Once Wolbachia becomes established in the population, the model indicates that the rate of disease transmission can be substantially reduced due to the bacteria both reducing adult mosquito longevity and inhibiting pathogen transmission. The pathogen inhibition phenotype of Wolbachia described for the wMelPop strain in Ae. aegypti [5], [7]and An. gambiae [4] is also produced by some, but not all, other strains of the bacterium, both in Drosophila [25]–[27] and Ae. aegypti [2]. To date only wMelPop has been shown to produce a significant reduction in lifespan, but it seems reasonable to predict that other strains will also produce some degree of life shortening when moved into a naïve mosquito host, particularly if costly immune pathways are activated [4], [5], [7]. Our results indicate that even a small reduction in adult longevity acts together with the direct effects of Wolbachia on the pathogen to produce a considerably greater reduction in pathogen transmission.
In general Wolbachia strains that induce strong pathogen inhibition with minimal or no associated life-shortening would be the optimal choice for use in disease control strategies, since this would reduce the level of releases that are required, improve spread dynamics, and minimize selective pressure for modulation of phenotypes that reduce pathogen transmission. Whether the additional Wolbachia-associated mortality occurs only in late life or throughout adulthood will also determine the strength of the selective pressure for modulation of the phenotype, and thus how long-lasting the strategy is likely to be in providing disease control. Sub-lethal effects could for example affect the capacity of young adults to escape predation in the wild. Ultimately it may only prove possible to obtain a full understanding of how cage survival dynamics translate to natural conditions, and relative mortalities in captive-bred versus wild insects, once field releases are actually underway.
Other aspects of mosquito biology that have not been considered here may also be important to Wolbachia spread dynamics. For example, in Aedes aegypti old females have been shown to have lower fecundity [28], [29], and this may reduce the fitness costs of Wolbachia infection if its effects on the host are strongest late in life. An extended version of the model that incorporates age-dependent fecundity can be analysed using the methods presented in Text S1. However we currently know little about the interactions between mosquito age and the effects of Wolbachia infection on fecundity for any mosquito species. This emphasises the need for detailed empirical study of the effects of Wolbachia on mosquito demography.
In conclusion, our results show that the establishment of Wolbachia in natural mosquito populations using male-biased releases is feasible provided that the mass-rearing capacity is available for the larger number of insects that need to be reared. Successful establishment of Wolbachia strains which reduce mosquito longevity or interfere with the pathogen in its vector are predicted to have substantial long-term benefits in terms of reduced disease transmission, and employing male-biased introductions minimises the risk of any biting or disease transmission during the release period.
|
10.1371/journal.pbio.1000176 | β1 Integrin Maintains Integrity of the Embryonic Neocortical Stem Cell Niche | During embryogenesis, the neural stem cells (NSC) of the developing cerebral cortex are located in the ventricular zone (VZ) lining the cerebral ventricles. They exhibit apical and basal processes that contact the ventricular surface and the pial basement membrane, respectively. This unique architecture is important for VZ physical integrity and fate determination of NSC daughter cells. In addition, the shorter apical process is critical for interkinetic nuclear migration (INM), which enables VZ cell mitoses at the ventricular surface. Despite their importance, the mechanisms required for NSC adhesion to the ventricle are poorly understood. We have shown previously that one class of candidate adhesion molecules, laminins, are present in the ventricular region and that their integrin receptors are expressed by NSC. However, prior studies only demonstrate a role for their interaction in the attachment of the basal process to the overlying pial basement membrane. Here we use antibody-blocking and genetic experiments to reveal an additional and novel requirement for laminin/integrin interactions in apical process adhesion and NSC regulation. Transient abrogation of integrin binding and signalling using blocking antibodies to specifically target the ventricular region in utero results in abnormal INM and alterations in the orientation of NSC divisions. We found that these defects were also observed in laminin α2 deficient mice. More detailed analyses using a multidisciplinary approach to analyse stem cell behaviour by expression of fluorescent transgenes and multiphoton time-lapse imaging revealed that the transient embryonic disruption of laminin/integrin signalling at the VZ surface resulted in apical process detachment from the ventricular surface, dystrophic radial glia fibers, and substantial layering defects in the postnatal neocortex. Collectively, these data reveal novel roles for the laminin/integrin interaction in anchoring embryonic NSCs to the ventricular surface and maintaining the physical integrity of the neocortical niche, with even transient perturbations resulting in long-lasting cortical defects.
| The developing cerebral cortex contains bipolar neural stem cells that span the cortical layers between the inner ventricular surface and the outer pial surface of the embryonic brain. The nuclei of these cells remain near the ventricular cavity, a microenvironment or niche thought to provide vital signals. It is not known how this inner end of the bipolar stem cell is held in place, or what would happen if its attachment to the inner surface were lost. Genetic manipulation can be used to disrupt candidate molecules involved in this adhesion, but this will affect all adhesion points and complicate the results. We have therefore developed an approach to target the stem cell attachments specifically at the ventricular surface by placing blocking antibodies directly into the ventricles of mouse embryos and then expressing fluorescent markers in the stem cells to see the effects of losing this attachment in living tissue. We examined laminins and integrins, whose expression and properties make them excellent candidates. Blocking integrin signalling by antibody application caused the inner end of the stem cells to rapidly detach and then undergo aberrant cell division. We also showed that a transient block of integrins (for <2 hours) resulted in permanent malformations of the cortical layers and disrupted neuronal migration.
| The cues responsible for maintaining the physical and molecular architecture of the stem cell niche of the developing mammalian brain are not well known. In the mammalian neocortex, the radial glia neural stem cells (NSC) that generate neurons are bipolar and have a radial morphology that spans the developing neocortical wall [1],[2]. These NSC have their soma located within the ventricular zone (VZ) adjacent to the ventricle, and their apical and basal processes make contact with the ventricular surface and the pial basement membrane, respectively [3],[4]. The basal (pial) process is important for informing the fate of the NSC daughter cells and then acting as a guidepost for their migration [5]–[7]. In contrast, the apical process contains cilia that extend into the ventricle and are thought to be important for morphogen signalling within the VZ microenvironment or niche [8],[9]. In addition, the apical process is necessary for the interkinetic nuclear migration (INM), which takes place during VZ cell proliferation [4],[10], and its transection results in translocation of the NSC soma away from the ventricular surface [7]. The apical processes of adjacent radial glia cells are attached to one another via cadherin-based adherens junctions that are Numb and Numbl-dependant [11],[12]. Recent studies have also highlighted the importance of Cdc42 [13], αE-catenin [14], β-catenin [15],[16], and the adenomatous polyposis coli protein (APC) [17] in the maintenance of this morphology. Deletion of Cdc42 as well as Numb/Numbl in NSCs disrupts apical adherens junctions resulting in defects in cell proliferation and disorganized cortical lamination [12],[13]. Likewise, αE-catenin and β-catenin, components of adherens junctions, regulate NSC cell cycle progression and thereby cerebral cortical size [14]–[16],[18]. Recently, APC has been shown to regulate the development and maintenance of the radial glial scaffold during corticogenesis [17]. However the specific adhesive molecules required for anchorage of these interconnected apical processes within the ventricular niche and their impact on neocortical development have not yet been determined.
The integrin α6β1 heterodimer is expressed at high levels in the apical regions of NSC [19]. Laminins, which serve as ligands for integrins in the extracellular matrix (ECM), are also present in the VZ niche [19], suggesting a possible role of laminins and integrins in providing these adhesive signals for NSC within the VZ. However, previous studies examining conditional deletion of β1 integrin in NSC [20] or α6−/− mice [21] did not report abnormalities of NSC behaviour in the VZ. We reasoned that this might reflect compensation for the long-term loss of one integrin by other heterodimer combinations, as has been described for β subunit integrin mutants in Drosophila midgut development [22]. To test the potential role of laminin/integrin binding in VZ maintenance and proliferation, we circumvented this possible compensation by transiently disrupting β1 integrin/laminin binding specifically in the VZ using blocking antibodies injected into the ventricle of the embryonic mouse brain. We also developed a novel ex vivo multiphoton time lapse imaging method that enables the effect of targeting of the blocking antibody to the cortical niche to be seen in real time. Furthermore, we analyzed VZ cell morphology and proliferation in laminin α2 deficient embryos. Together, our data demonstrate a novel role for laminin/integrin binding in the regulation of NSC proliferation and adhesion within the embryonic VZ, as well as its requirement to maintain the architecture of the neocortical niche.
While β1 integrin (accession number Swiss Prot P09055, http://www.ebi.ac.uk/swissprot) has previously been shown to be present in the VZ of the developing cortex [19],[20],[23], we confirmed the expression levels in the neocortical wall on the embryonic days at which we performed the perturbation studies. At E13.5, there is a high level of β1 integrin in the VZ, as shown by double labelling with a mitotic marker of M-phase, phospho histone 3 (PH3, Figure 1A and 1B). The high level of β1 integrin continues into the cortical subventricular zone (SVZ) as marked by the second layer of PH3+ cells, and β1 integrin is also highly expressed at the pial surface and in blood vessels (Figure 1A and 1B). Importantly, there are particularly high levels of β1 integrin on the apical surface of the VZ and on radial glia apical fibers (as assessed by double labelling with RC2, Figure 1E–1J). Analysis of the subcellular localization of β1 integrin within the ventricular processes reveals that this receptor is mainly located immediately basal to the adherens junctions (Figure S1). At E16, as large numbers of neurons begin to differentiate in the cortex, the level of β1 integrin remains high in the VZ/SVZ but decreases in the neuronal layers (Figure 1C and 1D).
Because β1 integrin is also expressed at the pial surface, where it is involved in the organization of the cortical marginal zone [20], one major challenge was to preferentially inactivate β1 signalling only at the apical surface to determine the particular contribution of β1 integrin in the cellular dynamics that take place in the VZ during formation of the neocortical wall. To accomplish this, we delivered a blocking antibody (Ha2/5) [24] into the cerebral ventricle in utero to specifically block β1 integrin function in cells bordering the ventricle. To determine the efficacy of this approach, we first assessed the in vivo dynamics of the antibody by injecting fluorescently conjugated Ha2/5 into the ventricles of E14 mice in utero. We observed widespread localization of the antibody within the VZ and SVZ after 6 h (Figure S2B) and 24 h (unpublished data). The antibody penetration was confined to apical regions and did not reach the pial surface. To confirm in vivo that the antibody inhibited integrin signalling, we evaluated phospho-Akt 1 (p-Akt 1) levels in the cortex 30 min after injection. p-Akt-1 is known to be highly expressed in NSC and is a well-recognized downstream signalling molecule in the integrin pathway [25],[26]. Brain lysates were prepared from E12 and E15 embryos injected either with the Ha2/5 or with an isotype control (ITC) antibody. Western blotting revealed a reduced level of p-Akt 1 expression 30 min after injection (Figure S2C). By 2 h after antibody injection, the differences in p-Akt 1 levels were absent (unpublished data) demonstrating that the perturbation of β1 integrin signalling is transient.
We took advantage of the movements of the NSC soma, which take place during cell cycle progression, to determine whether β1 integrin signalling affects the positioning of the NSC. Mitosis in the VZ normally occurs on the ventricular surface, after which the NSC soma transitions to the abventricular side of the VZ before entering S-phase (which can be identified by the incorporation of 5-bromo-2-deoxyuridine [BrdU] into the newly synthesized DNA). This cell cycle-dependent nuclear movement is known as INM [27]. Injection of 10 ng of the β1 integrin blocking antibody into the lateral ventricles of E12.5 and E15.5 embryos disrupted this pattern 18 h after injection, with mitotic PH3+ cells now scattered throughout the VZ (Figure 2A and 2B). Injection of a higher concentration of Ha2/5 (100 ng) produced identical results (unpublished data). Quantitative analysis (Figure 2C and 2D) revealed a significant increase in the number of PH3+ cells away from the ventricular surface (nonventricular surface or nVS) in both E12.5- (Figure 2C; p<0.01, unpaired two-tailed t-test) and E15.5-injected brains (Figure 2D; p<0.001, unpaired two-tailed t-test) without any changes seen in the number of PH3+ cells at the ventricular surface. Due to INM, a short 1 h pulse of BrdU normally labels cells clustered in S-phase at the abventricular boundary of the VZ (Figure 2E). Indeed, a maximum labelling index (calculated by the percentage of BrdU+ cells) was observed 60–70 µm away from the ventricular surface in the embryos injected with ITC antibodies (Figure 2G). In contrast, perturbation of β1 integrin signalling shifted the maximum labelling index 80–110 µm away from the ventricular surface (Figure 2F and 2G; p<0.05, two-way ANOVA), and also resulted in an overall increase in the number of BrdU+ cells. Thus in addition to the PH3+ mitotic cell ectopias, supernumerary S-phase cells were found in abnormal positions following β1 integrin blockade. To investigate somal translocation towards the VZ surface (i.e., M-phase reentry), E15.5-injected embryos were pulsed with BrdU 6 h prior to sacrifice, allowing the majority of proliferative cells to transit through S-phase and be on the ventricular surface either in or approaching mitosis at the time of analysis. In the embryos injected with the ITC antibody, this nuclear movement was indeed observed with the highest labelling index occurring in bin 1 nearest the ventricle (Figure 2H). While bin 1 also contained the highest labelling index in Ha2/5-injected brains, there was a significant increase in the labelling index of abventricular bins (bins 7–20) compared to controls. Thus, although continued ventricular divisions were apparent following blockade of β1 signalling, the abventricular dividing progenitor population was significantly increased.
To determine whether changes in the SVZ may be related to the mode of division in the VZ, we analyzed the distribution of cleavage plane angles. It has previously been shown that the mitotic spindle undergoes significant rotation during metaphase [28]–[30], leading to changes in the cleavage orientation of mitotic figures, which may be an indication of cell fate [29]–[33]. In addition, β1 integrin signalling has previously been shown to affect mitotic spindle formation in Chinese Hamster Ovary (CHO) cells in vitro [34]. We therefore assessed the orientation of cell divisions in the VZ 18 h after antibody injection and found that the β1 integrin blocking antibody caused a significant change in the pattern of cell divisions (Figure 3). In the developing telencephalon, the majority of mitoses occur vertically with cleavage angles greater than 60 degrees relative to the ventricular surface [30],[31], although a small percentage (15%–20%) can be seen with lower degrees of cell division (i.e., horizontal divisions). Indeed, this is what was observed with the ITC-injected embryos 18 h after antibody injection on E12.5, E13.5, E14.5, and E15.5 at rostral, medial, and caudal levels of the telencephalon (Figure 3A, 3C–3E). However, there was a reduction in the amount of (horizontal) cell divisions with cleavage angles below 60 degrees in the β1 integrin antibody injected embryos in the medial and caudal regions of the dorsal telencephalon (Figure 3C–3E). Using a statistical model to analyze the distribution of the VZ cleavage angles throughout neurogenesis (from E13 to E16), we found that the proportion of horizontally dividing VZ cells (0–30 degrees) is significantly lower at the medial and caudal levels of the forebrain after disruption of β1 signalling (Figure S3A). Interestingly, the effect of blockade of β1 integrin signalling was not seen in neural precursors located rostrally (Figures 3C, 3E, and S3). Together, these data indicate that the ventricular divisions that remain following β1 integrin blockade exhibit altered cleavage parameters coincident with the increased number of abventricularly proliferating cells.
Our BrdU and cell division studies clearly demonstrate that disruption of β1 integrin signalling leads to the presence of ectopic mitotic cells. We considered the possibility that these positioning defects lead to precocious differentiation. First, we determined the effects of β1 integrin blockade on the number of intermediate progenitor cells (IPC), which express the transcription factor T-brain 2 (Tbr2) [35]. IPC are neuronal progenitors that are generated from the NSC in the VZ, and which undergo further rounds of division just outside of the VZ in the SVZ [2],[35]. We found no difference in the number of Tbr2+ cells between ITC and Ha2/5-injected brains at E13 (Figure S4A–S4D). In addition, no premature neuronal (β3 tubulin, Figure S4E and S4F) or glial (NG2, unpublished data) differentiation was detected in the neocortical wall. Thus, β1 integrin blockade did not lead to abnormalities in cell differentiation within 18 h, and notably, although the number of proliferating cells in abventricular positions was increased, we found no increase in the number of Tbr2+ IPCs.
To determine whether the cell positioning defects following β1 integrin blockade are due to disruption of NSC morphology, we simultaneously performed electroporation of an RFP-expressing plasmid (CAG-RFP) with the Ha2/5 antibody injection to fluorescently label a population of VZ cells at the time of β1 integrin inhibition (Figure 4). We used electroporation parameters previously shown to transfect only VZ cells [36] and determined whether cells had detached from the ventricle surface within 18 h of co-electroporation/injection (Figure 4A and 4B). To do this, sections were stained with phalloidin to label the actin ring at the border of the NSC apical membranes so that the apical processes attached at the ventricular surface could be unambiguously identified (Figure 4C–4F; further 3-D examples can be seen in Figure S5 and Video S1). Volumetric reconstructed slices were created by image analysis and both the numbers of cell soma and apical processes were counted to generate a soma∶process (S∶P) ratio (Figure 4C–4H), and the percentage of apical processes still attached at the ventricular surface was also determined (Figure 4I and 4J). There was a significant difference between the two groups (*, p<0.05, unpaired two-tailed t-test), with ITC-injected embryos having a lower S∶P ratio and a higher percentage of apical processes in contact with the ventricular surface compared to the brains injected with the β1 integrin blocking antibody at both E13.5 (Figure 4G and 4I) and E15.5 (Figure 4H and 4J). This 3-D analysis therefore identified a morphometric change in neocortical VZ cells following Ha2/5 injection, with β1 integrin blockade resulting in detachment of apical processes from the ventricular surface as shown in Figure 4A. Furthermore, we also identified many dystrophic ascending basal processes emanating from the VZ cells (Figure 4A and 4B) indicating that apical detachment has widespread morphological effects on VZ cells and may therefore adversely affect neuronal migration.
To visualize the impact of β1 integrin signalling blockade on VZ cell morphology in real time, we performed time lapse imaging of neocortical VZ cell dynamics in living slices following electroporation with farnesylated enhanced green fluorescent protein (eGFP-F) (Figure 5). To specifically block β1 integrin at the apical surface without disrupting its function at the pial surface, we applied a drop of growth factor-reduced matrigel containing the antibody (β1 integrin blocking or ITC control) inside the lateral ventricle of the living slices prepared from E14.5 embryos electroporated 24 h earlier (Figure 5A). Analysis of the diffusion of β1 integrin blocking antibody-FITC from the drop of matrigel into the neocortical wall via pixel intensity profiles revealed that β1 integrin blocking antibody is mainly present in the 1/5 of the neocortical wall next to the ventricle in living slices; this corresponds to the VZ/SVZ compartment and indicates that, as with the in utero experiments, the blocking antibody does not reach the pial surface (Figure S6). This enabled us to monitor the effect of localized antibody blockade on eGFP-F+ cell morphology in the slices. After 10 h of contact with the antibody, we noted the progressive bending of both basal and apical processes (red arrows, Figure 5C and Video S3) as well as the detachment of apical end-feet from the ventricular surface (red arrowheads, Figure 5C and Video S3). In contrast, in the control experiment, NSC apical and basal processes are unaffected (Figure 5B and Video S2). Collectively, these data demonstrate that β1 integrin signalling disruption at the ventricular surface results in a progressive destabilization of the VZ architecture due to the simultaneous loss of both NSC bipolar morphology and apical end-feet at the ventricular surface.
The laminin α2 chain (accession number Swiss Prot Q60675) mediates cell adhesion through β1 integrins [37] and is expressed at the embryonic ventricular surface [19]. Thus, laminin α2 chain-β1 integrin interactions may be involved in the NSC adhesion at the ventricular surface during corticogenesis. To test this possibility, we analyzed cell proliferation and mitotic cleavage parameters in the VZ of laminin α2 deficient mice (Lnα2−/− mice) [38]. As with β1 integrin blockade, more proliferating cells were present outside the VZ/SVZ after a 1 h BrdU pulse in Lnα2−/− embryos (Figure 6A–6C). Furthermore, the angle of VZ cell division was also altered in Lnα2−/− embryos with the proportion of horizontal divisions (0–30 degrees) significantly lower in the medial region of the telencephalon (Figures 6D and S3B). Using the same experimental paradigms as in the β1 integrin blockade experiments, we performed in utero electroporation of the CAG-RFP plasmid in E15.5 Lnα2−/− mutant embryos (Figure 6F) and control wild-type littermates (Figure 6E). We then quantified the numbers of cell soma and apical processes and determined both the S∶P ratio (Figure 6G) and the percentage of apical processes (Figure 6H) still in contact with the ventricular surface. There was a significant difference between the two groups (*, p<0.05, unpaired two-tailed t-test) with a higher S∶P ratio in the Lnα2−/− mutants compared to the controls, consistent with an apical detachment of electroporated NSC. These results show that disruptions of either β1 integrin or of a ligand expressed in the VZ lead to identical alterations in cell position, NSC proliferation, orientation of cell division, and apical process detachment. These results with the Lnα2−/− mice therefore identify laminin α2 as a key ligand for the integrins expressed in the VZ and thus provide a genetic corroboration of our antibody perturbation studies.
To investigate the long term consequences of β1 integrin blockade and detachment of VZ cells on neocortical morphogenesis and layer formation, we utilized the co-electroporation/antibody injection strategy to mark cells at E15.5 and then allowed cortical development to proceed until postnatal day (P) 4. We reasoned that VZ cell detachment may lead to disruption of cortical layering since the detached NSCs with dystrophic radial fibers that we observed in the short-term experiments would not generate the proper amount of committed neurons and would alter the migration route to the cortical plate. Indeed, we found a reduction in the width of cortical layers I-V (Figure 7A–7C), as well as in the radial distribution of RFP+ cells in the somato-sensory cortex following β1 integrin antibody injection (Figure 7D–7F). Interestingly, in keeping with the rostro-caudal differences in β1 integrin blockade described in Figure 3 spatial discrepancies were also found in the postnatal cortex of animals injected with β1 integrin blocking antibody at E15.5; cortical layer thickness was reduced in somato-sensory but not in the primary motor cortex, although these results did not reach statistical significance. These results therefore provide evidence that proper maintenance of apical process attachment during embryogenesis is critical not only for INM and NSC proliferation, but also for neuronal migration and cortical cell layer formation, as a result of which transient disruption of β1 integrin signalling can have long lasting effects.
Using a multidisciplinary approach that includes cellular/molecular analysis and multiphoton time lapse imaging, we have revealed a hitherto unsuspected role for β1 integrin during neocortical development. Previously, β1 integrin has been suggested to be important for neocortical formation through its regulation of the radial glial contacts on the pial basement membrane [20],[39]. In our study, we combined in utero electroporation and injection of a specific blocking antibody to specifically inactivate the β1 integrin receptor by preventing binding to its ligand (laminin) at the ventricular surface. Compared to transgenesis or siRNA knock down, which cause widespread effects throughout both cell and tissue, this novel approach resulted in a focused and transient disruption at the subcellular level that resulted in detachment of the apical processes of many NSC from the ventricular surface and led to increased numbers of ectopic proliferating cells as well as perturbations to INM. Confirming that integrins act at least in part through interactions with laminins in the neocortical VZ, we found similar abnormalities in the laminin α2-deficient mouse. Together, our data clearly demonstrate for the first time in vivo, to our knowledge, that integrin/laminin interactions at the apical VZ surface play a critical role in the adhesion that maintains the stem cells within their niche and preserves the architecture of the VZ.
The adhesion of stem cells to their niche is critical for the molecular programmes that promote maintenance. For example, altered expression of adhesion-related genes is known to cause depletion of haematopoietic and epidermal stem cell niches [40],[41]. A recent report has also shown that niche-supporting gonad cells in Drosophila also require integrin signalling to ensure niche integrity [42]. Although the VZ lacks a basal lamina, which is well recognized as a principal site of cell/extracellular matrix interactions, we have shown previously that both laminins (α2, α4, and α2 chains) and integrins are expressed at the apical surface of the neocortical wall in the embryonic mouse VZ [19]. Our present observations suggest that integrin/laminin interactions are necessary to enable the retention of apical processes seen for at least 5 h after mitosis, and which may be critical for key cell-cell interactions that instruct behaviour [6]. So, while no perturbation of cell differentiation following β1 integrin blockade has been detected in our study, premature loss of these interactions resulting from apical process detachment has profound consequences on other aspects of NSC behaviour, including dysregulated proliferation of the NSC and altered allocation to the developing cortical plate. Interestingly a recent investigation of the role of α6β1 integrin in the adhesion of adult SVZ progenitor cells to endothelial cells using the same in vivo blocking antibody paradigm demonstrated two similar phenotypes to those we observed in the embryo—separation of SVZ progenitor cells from their normal location (adjacent to blood vessels) and enhanced proliferation [43]. Integrin/laminin interactions may therefore play similar roles in the regulation of neural stem and progenitor behaviour in embryonic and adult central nervous system. Whether blood vessels in the embryonic NSC niche provide some of these laminins as they do in the adult remains unknown, but recent studies do suggest a critical role of the developing cortical vasculature in regulating cortical neurogenesis [44],[45] and laminin α2 is expressed in blood vessels in the embryonic VZ [19]. Further work to test the hypothesis that laminin/integrin interactions in the vicinity of blood vessels contribute to the embryonic niche as they do in the adult is therefore required.
The two other phenotypes we observed after disruption of integrin signalling in the VZ, the loss of the subset of VZ cells that divide with horizontal cleavage planes and abnormal cortical layer formation may not simply be explained by an effect solely on VZ adhesion. Under normal circumstances, horizontal cleavages are the minority, and daughter cell fate cannot be predicted solely by the cell division orientation of its parent cell [31],[46],[47], because cells undergoing vertical cleavage during mitosis can give rise to either identical (via symmetric division) or different (via asymmetric division) daughter cells [48]. However, several recent studies have suggested an important link between the precise regulation of mitotic spindle orientation and the fate of neocortical neural progenitors. In particular, disruptions of centrosomal proteins such as Aspm [33], Nde1 [49], doublecortin-like kinase [50], and Cep120 [51], all of which play crucial roles in mitotic spindle function, affect the neural progenitor pool size and lead both to alterations in INM [51] and reductions in cerebral cortical size [49]. Although the link between mitotic spindle orientation and daughter cell fate is still debated, several recent studies demonstrate the close relationship between VZ cell cleavage angle and the location of the resulting daughter cells (i.e., ventricular surface versus abventricular location) [46],[48; present study]. While the detailed molecular mechanisms by which the integrin/laminin interaction influences the NSC cleavage orientation are still not known, compelling data linking integrin signalling to spindle assembly have already been reported for Chinese Hamster Ovary cytokinesis in vitro [34]. Thus our present results extend the importance of this role by demonstrating that β1 integrin signalling is required for the regulation of NSC mitotic spindle dynamics for the cells that normally undergo oblique cleavages during neocortical neurogenesis in vivo (Figure 8). Furthermore, our data suggest regional differences in that medial and caudal telencephalic progenitors are most sensitive to β1 integrin signalling. In keeping with this, it is interesting to note that human congenital muscular dystrophy caused by deficiency of the laminin α2 chain has been associated with significant abnormalities of cortical development in the occipital but not frontal regions of the telencephalon [52].
The thinning of the cortical layers that we observed in the postnatal mouse brain following transient blockade of integrin signalling in the embryo might reflect the alterations in the plane of cell division and subsequent effects on neurogenesis, but a key observation argues against this. We found that the NSC proliferation and morphological defects occurring after β1 integrin blockade had long-term consequences on the migration of both the newly formed neurons as well as those previously generated before the antibody injection. For example, the deep cortical layers (IV and V), which contain neurons born before the perturbation on E15.5, were also thinner than in controls. This deep layer defect is not likely to be caused solely by a disruption in neurogenesis at midgestation, since the earlier born neurons would be expected to establish proper laminar positions. Rather, it points to a phenotype resulting from the dystrophic radial glia processes we observed in the antibody-injected tissue, with cells born several days prior to the injection and still en route to the cortical plate affected by the morphological changes to the radial glia in the VZ. The most parsimonious explanation is that the loss of apical adhesion leads to NSC detachment and shortening and dystrophy of the basal process, and this in turn perturbs the migration of the cells on these processes. Supporting this proposed mechanism, several reports have demonstrated that mechanical forces play an important role in shaping the developing brain [7],[53],[54]. The present data suggest that anchorage of the apical endfeet provides the physical tension required for maintenance of position and morphology of radial glia cells during corticogenesis. Thus, our data using short-term blocking approaches reveal functions not shown by knock-out experiments and clearly define the novel contribution of integrins to neocortical development by elucidating a number of key roles in the regulation of NSC behaviour in the mammalian VZ.
Control ICR mice were produced in the Children's National Medical Center (CNMC) animal facility. ICR (CNMC), C57/BL6 (National Institute on Aging, NIA), and laminin α2 deficient mice (Oregon Health and Science University, OHSU and also SUNY Stony Brook, NY) were housed under standard conditions with access to water and food ad libitum on a normal 12 h light/dark cycle. Genotyping for the laminin α2 deficient mice was performed as previously described [38].
6- or 20-µm thick coronal sections from frozen Tissue-Tek embedded embryonic brains were harvested from three different levels along the rostro-caudal axis depending on the experiment. Most of the studies focused on the medial region of the forebrain, corresponding to E13/E14 plate 4 (for E13 to E14 embryos) or E15/E16 plate 5 (for E15 to E16 embryos) in the atlas by Jacobowitz and Abbott (1998). The analysis of the cleavage plane angle was also performed on the rostral (E13/E14 plate 3; E15/E16 plate 3) and caudal (E13/E14 plate 5; E15/E16 plate 8) regions of the embryonic forebrain [55].
Intraventricular injections were done with approval from the NIA and CNMC Institutional Animal Care and Use Committees using methods described previously [36]. Briefly, timed-pregnant mice (from E12 to E15) were anesthetized with ketamine/xylazine and a midline laparotomy was performed exposing uterine horns. The lateral ventricle in the brain of each embryo was visualized with transillumination and the injections were performed with a glass capillary pipette (75–125 µm outer diameter with bevelled tip) driven either by a Sutter micromanipulator (Sutter Instrument Company) equipped with 20-µl Hamilton gas-tight syringe or a nitrogen-fed Microinjector (Harvard Apparatus). For integrin blocking studies, approximately 1 µl of either β1 integrin blocking antibody (10 ng or 100 ng; Ha2/5, with or without FITC conjugation, BD Pharmigen) or an ITC antibody (anti-hamster, BD Pharmigen) solution (combined 3∶1 with sterile fast green dye to enable monitoring of the injection into the cerebral ventricles, Sigma) was injected alone or mixed with DNA. Two different plasmid vectors were used: a plasmid encoding red fluorescent protein under the control of the chicken β actin promoter (CAG-RFP) and a plasmid expressing eGFP-F (Clontech). For the in utero electroporation procedure, the anode of a Tweezertrodes (Genetronics) was placed above the dorsal telencephalon and four 40-V pulses of 50 ms duration were conducted across the uterine sac. Following intrauterine surgery, the incision site was closed with sutures (4-0, Ethicon,) and the mouse was allowed to recover in a clean cage. Mice were humanely killed 8–24 h after the injection unless indicated otherwise and embryonic brains were harvested.
Organotypic slices were prepared 24 h after in utero electroporation performed on E14.5 brains with an eGFP-F plasmid as described previously [30],[56]. Briefly, 300-µm-thick slices containing EGFP-F+ cells were collected in ice-cold Complete Hank's Balanced Salt Solution using a vibrating microtome (Leica VT1000S) and transferred into serum-free medium (SFM; neurobasal medium supplemented with B27, N2 and glutamax; Invitrogen). After 1 h of recovery, the slices were placed in a 35-mm glass bottom culture dishes. ITC control or β1 integrin blocking antibodies were diluted (1∶100) in growth factor-reduced matrigel (BD Biosciences) and a drop (0.5 µl) of this solution carefully introduced in the ventricular space of the embryonic brain slices. A slice holder immobilized the slices and 3 ml of SFM were added. The 35-mm glass bottom culture dishes containing the slices with matrigel were positioned in a heated micro-incubation chamber (DH-40i; Warner Instruments). Preheated SFM was pumped over the slices for the length of the imaging experiment (usually 10 h), the slice temperature was maintained at 37°C and the imaging preparation was maintained in 5% CO2/95% air for the entire period. All multiphoton imaging was performed on a Zeiss LSM 510 Meta NLO system equipped with an Axiovert 200M microscope (Zeiss) direct coupled to a Mira 900F laser pumped by an 8-W Verdi laser (Coherent Laser Group). EGFP was excited at 850 nm and time-series experiments were conducted under oil-immersion with 25× objective. Time-series images consisted of 40-µm-thick z-stacks and were collected at multiple locations at 2 min intervals to repetitively record both β1 integrin blockade and control slices. The experiments were analyzed with LSM 510 software. For the presentation of videos, each z-stack was projected onto one optical slice per time period and the resulting frames were assembled and compressed using Volocity software (Improvision).
For analysis of the diffusion of the blocking antibody in the experiments using matrigel to deliver antibodies within organotypic slices, these slices were prepared as above. After 10 h of incubation with the ITC control or FITC-labelled β1 integrin blocking antibodies in a drop of growth factor-reduced matrigel placed in the ventricular cavity, slices were fixed in 4% PFA and nuclei stained with dapi. 20-µm-thick z-stacks were collected and were analyzed with LSM 510 software. Each neocortical length was divided in five bins, each representing 20% of the total cortical thickness. The pixel intensity calculated by the LSM software was summed for each bin and then averaged and plotted in a graph (Figure S6).
20-µm-thick coronal sections were imaged using a Zeiss LSM 510 NLO system direct coupled to an inverted Axiovert 200M microscope (Zeiss). 25× (DIC, 0.8 na; Zeiss) image stacks (1-µm intervals) containing the region/cells of interest were collected with conventional detectors and then analyzed post hoc. For the orientation of cell division, studies were conducted with a 40× oil-immersion lens (DIC, Plan Neofluar, 1.3 na; Zeiss). Each frame of the series, consisting of a z-stack of images, was reconstructed in 3-D using Zeiss LSM software and was then rotated around the y-axis to bring the edge of the mitotic figures at the VZ surface into view so that the mitotic spindle plane was parallel to the computer screen. The angle of the mitotic spindle was then measured by projecting a line through the spindle to a reference line parallel with the ventricular surface. This procedure was repeated for each mitotic figure in each frame from the beginning of metaphase (a discrete organized metaphase plate) until the beginning of chromatid separation in anaphase. The spindle angles were then documented manually and graphed using SigmaPlot software.
The 3-D reconstruction of CAG-RFP cell attachment to the ventricular surface labelled with phalloidin was performed using Volocity software (Improvision).
For integrin signalling validation, half the litter was injected with the blocking antibody and the other half with the control antibody for 30 min. Telencephali were isolated by rapid dissection 30 min after injection and then flash frozen. Total brain lysates were prepared by resuspending the tissue in cell lysis buffer. Tissue protein was extracted using T-PER tissue protein extraction buffer with protease inhibitor cocktail (Sigma) and protein concentration was determined by the BCA protein assay kit (Pierce). 50 µg of protein was separated by SDS-PAGE (8%–12%) and transferred to nitrocellulose membranes. The membranes were blocked in 5% nonfat milk for 1 h at room temperature, followed by an overnight incubation at 4°C with antibodies raised against p-Akt1 (BD Pharmigen), total Akt (T-AKT, BD Pharmigen), or β-actin (Sigma). Membranes were then washed and incubated with secondary antibodies for 1 h at room temperature. Protein bands were visualized using a chemiluminescence detection kit (Amersham Biosciences).
Embryonic brains were fixed in 4% paraformaldehyde (PFA) in PBS overnight at 4°C before being transferred to sequential 20% and 30% solutions of sucrose (w/v) and left at 4°C overnight or until the brains equilibrated. The brains were then embedded in TissueTek (Sakura) prior to cryostat sectioning (Leica CM3050S). For immunofluorescence, sections were blocked for a minimum of 30 min in PBS containing 0.1% Triton X-100 and 10% normal goat serum (Sigma). Sections were incubated overnight with primary antibodies at 4°C. After incubation with the appropriate secondary antibodies and counter-staining with 4′,6-diamidino-2-phenylindole dihydrochloride (Dapi, Sigma) to visualize the DNA, images were acquired using an Olympus IX50 fluorescence microscope. Images were processed using MagnaFire and Photoshop 6.0 (Adobe) and adjusted such that the entire signal was in the dynamic range. The following antibodies were used for immunofluorescence: anti-β1 integrin (used 1∶100 in blocking buffer), anti-Tbr2 (used 1∶200 in blocking buffer following a 5 min boil in 10 mM sodium citrate, Millipore), anti-RC2 (used 1∶5 in blocking buffer, Developmental Studies Hybridoma Bank), anti-phospho histone H3 (used 1∶500 in blocking buffer, Millipore), anti-BrdU (used 1∶5 in BrdU blocking buffer, Accurate Chemicals), anti-β3 tubulin (used 1∶500 in blocking buffer, Sigma). For BrdU staining, the blocking buffer consisted of DMEM (Sigma) supplemented with 1% tween-20 (Sigma) and 7 mg of DNAse (Sigma) per 1 ml of blocking solution.
The angle of the cleavage plane was determined in cells in anaphase identified by propidium iodide staining performed on 20-µm-thick sections from three different levels (rostral, medial, and caudal) of the forebrain from E13 to E16 embryos. F-actin filaments were visualized at the ventricular surface using Alexa Fluor 488 phalloidin (165 nM final concentration, Molecular Probes) by incubation for 1 h after a prior 5 min incubation in 0.1% Triton X-100 in PBS and 30 min in 10% normal goat serum in PBS for blocking. Nuclear counterstaining was performed by 10 min incubation at room temperature in TOPRO-3 iodide (1∶100, Molecular Probes).
For the subcellular localization analysis of β1 integrin, first a postfixation with methanol at −20°C for 10 min was performed on the cryosections before a blocking step in a solution containing bovine serum albumin 3% and Tween 0.05% for 1 h at room temperature (RT). Then, anti-β1-integrin antibody (clone MB1.2 from Chemicon Int., 1/100 dilution) and Alexa Fluor 546 coupled phalloidin (Molecular Probes, 1/200 dilution) were incubated overnight in the blocking buffer at RT. Alexa Fluor 488-conjugated donkey anti-rat antibody (Molecular probes, 1/250 dilution) was incubated for 1 h at RT along with Hoechst for nuclei staining. Images were captured with a Zeiss confocal using an oil immersion 63× objective with a zoom of 2. The profile function of the Zeiss acquisition software was used to determine the fluorescence intensity of each marker at a defined xy position.
For PH3 analysis performed 18 h following antibody (ITC or β1 blocking) injection at E12.5, positive cells were calculated from the average of three sections from five separate embryos from two litters. For E15.5 injected embryos, positive cells were calculated from the average of three sections from 11 (ITC) or 17 (β1 integrin blocking antibody) separate embryos from nine litters. For BrdU analysis, labelling index was calculated on the basis of three sections from three embryos from 1 litter for both E12.5 and E15.5 injected embryos. For Tbr2 expression analysis, a 200 µm×100 µm (width×height) region adjacent to the ventricular surface was analyzed and an average was calculated on the basis of three sections from three embryos from one litter. All statistical analysis was performed using GraphPad Prism version 4.00 for Windows, GraphPad Software (www.graphpad.com). The specific statistical test is indicated in both the text and figure legends.
Apical process quantification was performed on 150-µm vibrating microtome-cut coronal sections of co-antibody injected/CAG-RFP electroporated E13.5 and E15.5 brains stained with phalloidin to label the ventricle surface. 80-µm z-stacks were collected in 2-µm steps with a 1,024×1,024 pixel frame size and each z-stack was analyzed with LSM examiner (Zeiss) and Volocity software (Improvision) to determine the number of RFP+ cell bodies within 200 µm from the ventricle surface and to determine the number of apical processes attached to the ventricle (colocalized with the phalloidin staining). Both of these counts were calculated from the average of three embryos for each condition. The ratio soma/apical processes (S∶P) representing the total number of cell bodies divided by the number of apical processes was determined and the results were analyzed by unpaired two-tailed t-test.
The postnatal phenotype of embryos co-injected/electroporated at E15.5 was assessed at P4 by determining the thickness of the individual cortical layers (I, II/III/IV, V, VI as shown in Figure 7A) and by the radial distribution (i.e., layer specification) of RFP+ cells in multiple areas at both the primary motor and somato-sensory cortical levels. The data from two ITC and three β1 integrin blocking antibody injected animals were analysed by unpaired two-tailed t-test.
In preparation of performing statistical analysis we checked assumptions of normality and homogeneity of variance, and found that the data for the orientation of cell division were not normally distributed and could not be transformed to achieve acceptable levels of normality to permit linear regression analysis. Thus, an ordinal logistic regression model was developed to estimate the tendency toward having greater angles of cleavage in one group (β1 integrin blocking antibody injected brains or Lnα2−/− brains) compared to another (controls: ITC injected brains or wild type littermates from Lnα2−/− embryos). To perform these analyses, angles of cleavage were stratified as, <30 degrees, 30 to <60 degrees, and 60–90 degrees. The model, which included covariates to account for brain region and study group, enabled the study to estimate and compare differences in the frequency of angle of cleavage strata in one study group compared to another. The model adjusted variance estimates to account for the correlation between repeated measurements on the same embryo.
|
10.1371/journal.pgen.1000747 | Mutations in GDF5 Reveal a Key Residue Mediating BMP Inhibition by NOGGIN | Signaling output of bone morphogenetic proteins (BMPs) is determined by two sets of opposing interactions, one with heterotetrameric complexes of cell surface receptors, the other with secreted antagonists that act as ligand traps. We identified two mutations (N445K,T) in patients with multiple synostosis syndrome (SYM1) in the BMP–related ligand GDF5. Functional studies of both mutants in chicken micromass culture demonstrated a gain of function caused by a resistance to the BMP–inhibitor NOGGIN and an altered signaling effect. Residue N445, situated within overlapping receptor and antagonist interfaces, is highly conserved among the BMP family with the exception of BMP9 and BMP10, in which it is substituted with lysine. Like the mutant GDF5, both BMPs are insensitive to NOGGIN and show a high chondrogenic activity. Ectopic expression of BMP9 or the GDF5 mutants resulted in massive induction of cartilage in an in vivo chick model presumably by bypassing the feedback inhibition imposed by endogenous NOGGIN. Swapping residues at the mutation site alone was not sufficient to render Bmp9 NOG-sensitive; however, successive introduction of two additional substitutions imparted high to total sensitivity on customized variants of Bmp9. In conclusion, we show a new mechanism for abnormal joint development that interferes with a naturally occurring regulatory mechanism of BMP signaling.
| The development of the human skeleton is regulated by intricate signaling pathways involving secreted molecules that bind to cell surface receptors to elicit a response in the target cell. Bone morphogenetic proteins (BMPs) are an important part of this process. Their signaling capacity is regulated on several levels including the extracellular space where inhibitors such as Noggin prevent BMPs from binding to their cognate receptors. We here describe that specific mutations of a single amino acid in GDF5, a member of the BMP family, cause congenital fusion of joints. Investigating the effect of the mutation in detail, we can show that the mutant GDF5 is no longer inhibited by Noggin, thus providing a functional explanation for the patients' phenotype. Furthermore, we show that the mutated residue (N445) is conserved throughout the BMP family, with the exception of BMP9 and BMP10, that carry the same amino acid at this position as the mutant GDF5. Both are, just as the mutant, resistant to inhibition by Noggin. Variants of BMPs that are insensitive to antagonists may induce bone formation more effectively, providing a source for effective, low-dose therapeutics for clinical applications.
| Bone Morphogenetic Proteins (BMPs) and the related Growth & Differentiation Factors (GDFs) are phylogenetically conserved signaling proteins that belong to the Transforming Growth Factor beta (TGFβ) superfamily. Originally identified for their ability to induce bone, they were subsequently shown to be involved in multiple aspects of body patterning and morphogenesis [1]–[2]. Mutations in BMPs and their receptors can cause a wide variety of congenital and postnatal diseases [3]–[4]. Despite their different functions, all BMPs share a common signaling mechanism. They are translated as precursor proteins consisting of a prodomain, which is released proteolytically by members of the subtilisin-like proprotein convertase family [5], which is important to activate signaling that is conferred through the mature domain [6]. The highly conserved mature domain is characterized by seven cysteine residues. Six of them form an intramolecular cysteine knot whereas the fourth of the seven cysteines is important for the dimerization of the BMP monomers [7]. BMPs are secreted peptides that act as homo- or heterodimers and bind to two different types of membrane-spanning serine/threonine kinase receptors, the type I BMP receptors (ACVRL1, ACVR1, BMPR1A, BMPR1B) and type II BMP receptors (BMPR2, ACVR2). Binding of BMPs to heterotetrameric receptor complexes activate the SMAD as well as other intracellular signaling pathways, like the MAPK pathway [8]. BMP signaling is precisely regulated by a large number of antagonists, which act extracellularly, on the membrane level as well as intracellularly [9]. Noggin (NOG; MIM *602991), one of the best characterized extracellular BMP antagonists, was first isolated as a dorsalizing factor secreted by Spemann's organizer of Xenopus embryos [10] and later shown to be required for patterning of the neural tube and in skeletogenesis [11]–[12]. The crystal structure of NOG bound to BMP7 provided the structural basis of inhibition of BMP signals, revealing the clamp-like grip of the covalently linked homodimer on the homodimeric ligand that blocks both pairs of receptor binding interfaces [13]. The importance of NOG for the development of joints was shown by the identification of mutations in NOG in patients with symphalangism (SYM1, MIM #185800) and multiple synostosis syndrome (SYNS1, MIM #186500) [14]. We and others have shown that certain mutations in GDF5 (also known as CDMP1; MIM *601146) are also associated with SYM1 and SYNS1 [15]–[18]. SYM1/SYNS1 associated mutations result in over-active GDF5 due to altered receptor binding specificity. In contrast, loss of function mutations in GDF5 result in brachydactyly types C or A2 (BDC, MIM #113100; BDA2 MIM #112600) [6], [19]–[20]. Homozygous loss of function mutations cause acromesomelic chondrodysplasia of the Grebe (MIM #200700), Hunter Thompson (MIM #201250) or Du Pan (MIM #228900) types, conditions characterized by extremely short limbs and digits [21]. A recent and comprehensive review on the respective genotype-phenotype correlations has just been compiled and published [4]. Here, we describe two novel GDF5 point mutations (N445K and N445T) that are associated with a pronounced form of SYNS1. Functional characterization of the mutants demonstrated increased biological activity when compared to wtGdf5 due to a resistance against inhibition by Nog. The importance of the N445 residue for NOG function was further substantiated by identification of two other BMPs, BMP9 and BMP10, that share the same replacement at this site and that are also insensitive to inhibition by NOG.
We identified a heterozygous missense mutation c.1335T>G leading to an exchange of the hydrophilic asparagine to basic lysine (p.N445K) in three family members with SYNS1. The mutation segregated in an autosomal dominant manner. The clinically unaffected mother of two affected children did not show the p.N445K mutation arguing for the presence of a germline mosaicism. In addition, we identified a de novo asparagine to threonine (p.N445T) mutation (c.1334A>C) in another patient with SYNS1. All four affected patients exhibited the characteristic features of severe SYNS1 (Figure 1). The radiographs showed bilateral fusions of carpal and tarsal bones as well as proximal symphalangism in fingers and toes. Distal phalanges of fingers and toes II to V were hypoplastic and some nails appeared small. Fusions between humerus and radius were present in all four patients leading to fixation of the elbow joints in a flexed position. Mild cutaneous syndactyly was present in some individuals. The phenotype was congenital and appeared to be non-progressive.
A 3D structural model of GDF5 in complex with BMPR1B (PDB 3EVS; [22]) with homology-based binding epitopes for the type II receptors superimposed on that for NOG shows that asparagine 445 co-localizes within both the type I receptor and the NOG interaction sites (Figure 2A and 2B). Sequence alignments showed that the N445 residue is conserved in the related BMP/GDF group, with the exception of BMP9 (K372), BMP10 (K368), GDF9 (V398) and GDF15 (M253) (Figure 2C).
Retroviral expression of Gdf5 in micromass cultures induces whereas Nog inhibits chondrogenesis, which can be quantified by Alcian blue staining of the extracellular matrix. The N445K and N445T mutants induced chondrogenesis to a similar extent as wtGdf5. Co-expression of Nog prevented induction of chondrogenesis by wtGdf5, but had only a minor effect on the Gdf5 mutants, demonstrating the critical role of the asparagine residue in the interaction of Gdf5 with Nog (Figure 3A and 3B).
BMP9 and BMP10 share the same amino acid residue at this position as one of the SYNS1- associated mutants. We analyzed Bmp9 and Bmp10 in the micromass system and observed an even stronger chondrogenic effect than for wtGdf5, indicating that they are potent inducers of chondrogenic differentiation. Co-expression of Nog did not suppress chondrogenesis significantly, demonstrating that both ligands are naturally Nog-insensitive unlike most other Bmps. To test whether the substitution shared by Bmp9 and Bmp10 conferred resistance to Nog, we swapped the wildtype lysine residue of Bmp9 with asparagine (K371N), the residue shared by most Bmp family ligands, including Gdf5. This exchange was not sufficient to transform Bmp9 into a Nog-sensitive Bmp (Figure 4A, 4B, 4C, and 4D), indicating that additional residues must contribute to diminished binding. Hence two other dissimilar sites within the NOG interface, shown in schematic and 3D surface models of BMP9 (Figure 4A and 4B), were targeted for exchange. Introducing a second swap (YH415Q) at a contact with the NOG N-terminal extension or “clip” diminished biological activity substantially upon co-expression of Nog. A third successive mutation (K348L), located on the periphery of the type II receptor interface, rendered Bmp9 completely Nog-sensitive relative to the control, without significantly affecting the prochondrogenic potential of the custom variant in the absence of Nog (Figure 4C and 4D).
To analyze the relative resistance to NOG more quantitatively, recombinant protein of one of the mutants (N445T) was produced and dose-dependent inhibition assays performed. For this C2C12 cells that stably express Bmpr1b, the high affinity type I receptor for GDF5, have been chosen because all analyzed proteins induced a significant amount of ALP activity, making it a convenient and reliable assay to analyze the inhibitory effect of NOG [16]. Dose response curves were generated to determine optimal concentrations for subsequent inhibition assays (Figure 5A). C2C12-Bmpr1b cells were incubated with rhGDF5, rhN445T GDF5, rhBMP2 or rhBMP9 in combination with rhNOG at increasing concentration and ALP activity determined after three days. Although rhBMP2 was completely inhibited by equimolar rhNOG, the apparent affinity of rhGDF5 for the antagonist was markedly less pronounced, requiring super-stoichiometric ratios to achieve total inhibition. Consistent with the effects in micromass, rhN445T GDF5 and rhBMP9 were unresponsive to rhNOG at any concentration (Figure 5B).
RhGDF5 only slightly induced ALP activity, but effectively inhibited myogenic differentiation of the myoblastic cell line C2C12. In contrast, the rhN445T GDF5 mutant induced ALP activity in a dose dependent manner in C2C12 cells, which indicates an altered signaling effect (Figure 6A and 6B). To test possible alterations of the interaction between the N445T mutant and the type I receptor, we co-stimulated C2C12 cells with recombinant proteins with soluble ectodomains (ecd) of TGFβ superfamily type I receptors and used the luciferase reportergene assay as a read out (Figure 6C, 6D, 6E, and 6F). A direct comparison of rhGDF5 and rhN445T GDF5 revealed that addition of BMPR1Aecd as well as BMPR1Becd had a slightly stronger inhibitory effect on rhN445T GDF5 than on rhGDF5. As anticipated, rhBMP2 was almost completely inhibited by the ectodomain of rhBMPR1A and partially by the ectodomain of rhBMPR1B [23], whereas rhBMP9 was antagonized by the ectodomain of rhACVRL1 [24]. To identify possible differences between rhGDF5 and rhN445T GDF5 with respect to their binding affinities to the two BMP type I receptors BMPR1A and BMPR1B we analyzed them by using Biacore. However, no significant changes between rhGDF5 and rhN445T GDF5 could be observed in these experiments (Table 1).
Stimulation of micromass cultures with recombinant proteins revealed stronger chondrogenic differentiation with rhN445T GDF5 than with rhGDF5 or rhBMP2. Because of the high potency of rhBMP9, a 1∶20 dilution was employed to prevent saturation effects (Figure 7A and 7B). Quantitative real time PCR showed that treatment of micromass cultures with rhGDF5, rhN445T GDF5, rhBMP2 or rhBMP9 resulted in an upregulation of Nog (Figure 7C). Upregulation of Nog by exogenous rhBMP was also examined in vivo. Heparin beads soaked with recombinant protein were implanted into chicken limb buds and in situ hybridization for Nog was performed. Implantation of rhBMP9 soaked beads (0.5 mg/ml) at day four (HH25–27) resulted in early lethality (n = 40). To avoid this effect, the beads were implanted at a later stage (HH29/30) containing less protein (0.25 mg/ml). In agreement with the in vitro results, we observed a strong upregulation of endogenous Nog surrounding beads indicating that wildtype as well as mutant BMPs induce the expression of their inhibitor (Figure 7D). To further evaluate and compare activity of growth factors in vivo, we overexpressed wtGdf5, N445T Gdf5, N445K Gdf5 and Bmp9 in developing chicken hind limb buds. Overexpression of wtGdf5 caused a general thickening of skeletal elements, with fusions of joints as well as digits. Overexpression of N445K and N445T Gdf5 resulted in a much more severe phenotype. The entire limb morphed into a single cartilage element, without joints or even soft tissue. Injection of Bmp9 had a lethal effect; however the Bmp9-infected embryos that survived (7 out of 40) showed a phenotype similar to N445 Gdf5 mutants (Figure 8). Thus, the increased biological activity of the GDF5 mutants as well as BMP9 may arise in large part from insensitivity to endogenously upregulated Nog.
Transduction of temporally and spatially specific signals through the BMP pathway in a diverse array of cellular contexts is achieved in part by differential affinities of the receptors of the heterotetrameric-signaling complex for the array of ligands in the family. However, differential affinity to secreted inhibitors that act as ligand traps plays a reciprocal, equally important role in context-specific tuning of BMP signals. Here, we show that single point mutations (N445K,T) in the wrist epitope of the BMP-related ligand GDF5 lead to a resistance to Nog, and a subsequent dramatic increase in the intensity of the signaling output. Moreover, we show that wildtype BMP9 and BMP10, two divergent members of the family that harbor the asparagine for lysine substitution, are markedly resistant to Nog in limb bud micromass and C2C12 myoblast cell assays.
The GDF5 mutations N445K,T are associated with an inherited skeletal disease characterized by joint fusions in fingers, toes and elbows. This phenotype is similar to that of SYNS1, previously associated with hypomorphic alleles of NOG. SYM1, a less severe condition in which symphalangism is limited to the small interphalangeal joints of the fourth and fifth finger, is also caused by mutations in NOG. We have previously shown that a SYM1-like phenotype can also result from gain of function mutations in GDF5 [15]–[16].
GDF5 mutations described here affect the wrist epitope, which is the polypeptide segment between the third and fourth conserved cysteine residue that is most divergent as shown in the multiple sequence alignment of TGFβ superfamily ligands (Figure 3). Because the wrist epitope is a key element of the type I receptor interface, the divergence may reflect the differential affinity of the superfamily of signaling ligands for their specific receptors. In addition, the diversity may reflect differential affinity for the array of secreted antagonists. For example, TGFβ ligands are deposited extracellularly as an inactive latent complex not regulated by secreted antagonists, whereas BMP/GDF ligands can bind with picomolar affinity, or evade inhibition almost entirely, as we have shown here for the interaction of NOG with the divergent BMP9, BMP10 pair [25]–[26].
Functional studies of the N445K,T mutations suggest a dual effect of the mutations - an altered signaling effect on the one hand and resistance to NOG on the other. Whereas the insensitivity of N445K,T mutations was clearly demonstrated in retroviral overexpression studies or by stimulation with recombinant proteins, the altered signaling effect could not be finally clarified on a molecular level. The fact that - in contrast to rhGDF5 - the rhN445T GDF5 mutant induced significant amounts of ALP in C2C12 cells prompted us to speculate on altered receptor specificity. This hypothesis was further analyzed by neutralization assays of different TGFbeta type 1 receptors and Biacore analyses. RhN445T GDF5 seemed to be slightly more sensitive to the inhibition by BMPR1Aecd and BMPR1Becd, but no significant changes of receptor affinities were detected. The different signaling effect might be explained either by different association and dissociation rates of the ligand-receptor complex or by the involvement of other co-receptors.
Interestingly, the reported R438L mutation in GDF5 lies within the same domain, i.e. the wrist epitope (type I receptor interface as well as NOG binding interface). Yet it results in altered receptor specificity with increased affinity to the BMPR1A, and not affecting the inhibition by NOG. The R438 position is not conserved within the BMP/GDF subgroup. We demonstrated that the rhR438L GDF5 mutant has a higher affinity for the BMPR1A receptor than rhGDF5, giving this mutant BMP2-like properties. BMP2 is also inhibited by NOG, which indicates that this amino acid change seems to have an effect on the receptor specificity only, but otherwise it is well tolerated by the BMP antagonist NOG.
The similarity of the resulting phenotypes caused by either loss of function NOG mutations or activating mutations in GDF5 argues in favour of GDF5 representing the most important target for NOG during the critical phase of joint development and indicates that SYM1/SYNS1 are associated with an increased biological activity of GDF5.
The crystal structure of the NOG-BMP7 complex showed that NOG, a covalently linked homodimer, inhibits BMP signaling by blocking the molecular interfaces of the binding epitopes for both type I and type II receptors, sequestering the covalently linked homodimeric ligand in an inactive complex [13]. In the crystal structure of GDF5 [27], the conserved asparagine (N445) is located near the amino terminal end of the large helix (α3), its amide sidechain projecting tangential to, and hydrogen bonding with, the backbone of the long finger 2 at a main chain carbonyl (E491) across the dimer interface. Superposition of the GDF5 crystal structure on BMP7 in the NOG-BMP7 complex provides a model of the NOG-GDF5 complex for interpretation of the role of the conserved asparagine and the consequences of the mutations. In the model, just as in the NOG-BMP7 crystal structure, the amide nitrogen of the asparagine sidechain is hydrogen bonded to two main chain carbonyl groups, one across the dimer interface (GDF5 E491) and one along the extended N-terminal 14 clip of the antagonist (NOG A36). In addition to this triangulated interaction, the carbonyl oxygen of the N445 sidechain also forms hydrogen bonds with the main chain via the amide nitrogen of NOG A36. Furthermore, in a superposition model of a GDF5-type I receptor complex, the asparagine sidechain appears to interact with E81 and G82 across the ligand-receptor interface. Thus, loss of the amide group by substitution with a lysine or threonine side chain in the N445K,T mutants would on the one hand disrupt two stabilizing interactions within the NOG-GDF5 complex, and, on the other hand, alter the ligand-receptor interface resulting in a broader specificity. With respect to the differential effects of threonine substitution on interaction with NOG, superposition of GDF5 and BMP7 in the crystal structure of the BMP7-NOG complex reveals a major structural difference between the two ligands at the interface with the key residue of the N-terminal extension of NOG, P35, also a site of several human mutations linked to SYM1 [13],[28]. Due largely to major conformational differences in the backbone and side chains of the short finger 1 of GDF5 relative to the BMPs, the hydrophobic pocket that accommodates the proline side chain is predicted to be only half formed, which would disrupt the stabilizing interaction to a similar extent as mutation of P35. Thus, interaction between wildtype GDF5 and NOG is anticipated to be weaker than that between BMPs and NOG, as evidenced by our titration assays which required super-stoichiometric ratios (cf. Figure 5B), perhaps so much so that disruption of the adjacent interaction between the conserved asparagine and the backbone of the N-terminal extension (A36) is sufficient to abolish formation of the GDF5-NOG complex. BMP9 and BMP10, which are not responsive to inhibition by NOG, do not bear the homologous asparagine residue. However introduction of the asparagine in BMP9 was not sufficient to confer sensitivity to NOG. Successive introduction of two additional GDF5-like substitutions, removal of a BMP9-specific insert (YH415Q) and K348L, imparted near total sensitivity. In conclusion, while substitution of the single asparagine is sufficient to impart NOG sensitivity in GDF5, other BMPs require further structural alteration in order to impart sensitivity, or resistance. Because ligand-receptor and ligand-antagonist complexes are sufficiently structurally conserved within the BMP/GDF family, GDF5 function was successfully transferred to BMP9 by rounds of rational design.
Currently, rhBMP2, rhBMP7, and rhGDF5 are used in clinical applications for bone regeneration or are under investigation in clinical trials. Although BMPs are very potent in cell culture systems or small animal models (nanograms), significantly higher concentrations are needed for measurable effects in humans (milligrams). It was shown that BMP antagonists are regulated during fracture healing [29], indicating that endogenous regulation of the BMP signaling cascade may neutralize exogenously applied rhBMPs, necessitating higher doses than required in vitro. In keeping with this hypothesis, the in vivo and in vitro studies presented here demonstrated induction of Nog after treatment with rhBMPs, an apparent physiological response to regulate and constrict BMP action. The upregulation of inhibitors in response to exogenous growth factors may therefore also be applied for BMP treatment, which can only be partially compensated for by high dosage due to feedback control mechanisms [9], [30]–[31]. Thus, variant BMPs that are insensitive to antagonist may induce bone formation more effectively, providing a source for effective, low-dose therapeutics for clinical applications.
All clinical investigations have been conducted according to Declaration of Helsinki principles. The study was approved by the local institutional review board. Patients were investigated and radiographed by a clinical geneticist who diagnosed typical SYNS1. Informed consent was obtained for genetic analyses from all patients or the legal guardians. Mutation screening in GDF5 was carried out as previously described on purified DNA obtained from blood sample [32].
Protein sequence alignments comprising the highly conserved cysteine knot domains of the human TGFβ superfamily were aligned using CLUSTAL X (http://bio.ifomfirc.it/docs/clustal/clustalx.html) [33] and colored using CHROMA (http://www.lg.ndirect.co.uk/chroma/) [34]. GDF5-NOG complex modeling was previously published [28]. Images of the molecular structure were produced using the UCSF Chimera package (http://www.cgl.ucsf.edu/chimera/) [35].
The coding sequences of mouse (m) mBmp9, mBmp10 were amplified by PCR on mouse embryo E14.5 cDNA and cloned into pSLAX-13 using the following primer pairs: mBmp9_BsmBI_f ACGTCTCCCATGTCCCCTGGGGCCTTCCG and mBmp9_BamHI_r TGGATCCTACCTACACCCACACTCA, mBmp10_BsmBI_f GATACGTCTCCCATGGGGTCTCTGGTTCTGCC and mBmp10_XmaI_r GATCCCCGGGCTATCTACAGCCACACTCAGAC. In vitro mutagenesis for chicken (ch) chGdf5 and mBmp9 was performed with Quickchange Kit (Stratagene) according to manufacturer's recommendations. Cloning into retroviral vector, and production of viral supernatant in DF1 cells and concentration of viral particles was performed as described [36]. RCAS(BP)B-Nog was a kind gift of Andrea Vortkamp.
RhGDF5 (Biopharm GmbH) was dissolved in 10 mM HCl, rhNOG was a kind gift from A. Economides (Regeneron Pharmaceuticals Inc.) and dissolved in 0.1% BSA/PBS, rhBMP2 was a kind gift from W. Sebald and dissolved in 4 mM HCl, 0.1% BSA, rhBMP9 (R&D Systems) was dissolved in 4 mM HCl, 0.1% BSA. RhN445T GDF5 was prepared as previously described with minor modifications [16]. Inclusion bodies were dissolved in 8 M Urea, 20 mM Tris HCl, 5 mM EDTA, 50 mM NaCl, and 66 mM DTT, pH 8.3. Purification was carried out on a cation exchange column SP Sepharose FF (XK26/40, GE Healthcare) with a gradient from 100% eluent A (6 M Urea, 20 mM Tris HCl, 1 mM EDTA, 50 mM NaCl, 10 mM DTT, pH 8.3) to 100% eluent B (6 M Urea, 20 mM Tris HCl, 1 mM EDTA, 400 mM NaCl, 10 mM DTT, pH 8.3), flow rate 3.5 ml/min. Eluents were concentrated by ultrafiltration (Amicon, Omega 5) and adjusted to 0.5 M NaCl. This solution was dissolved 1∶10 in refolding buffer (150 mM NaGlycine, 500 mM NaCl, 20 mM 33 mM 3-(3-cholamidopropyl) dimethylammonio-1-propanesulfonate, 3 mM oxidized glutathione (GSSG), 1 mM EDTA, pH 9.8) under gentle agitation so that a final protein concentration of 1 mg/ml was reached. The solution was then isoelectric precipitated with 1.8-fold 20 mM NaH2PO4 at 4 °C for 1 hour, pH 7.4. After centrifugation the pellet was dissolved in 0.1% trifluoroacetic acid. The dimeric GDF5 N445T was separated from monomeric protein by RP-HPLC (reversed phase column Source 15 RPC, fine line pilot 19 35/20 cm, GE Healthcare) with a gradient from 100% eluent A (0.1% trifluoroacetic acid (TFA)) to 100% eluent B (0.1% TFA, 90% CH3N), flow rate 14 ml/min.
Micromass cultures were performed as previously described [16]. For each condition, four replicates were performed in parallel and every experiment was done three times. Stimulation of micromass cultures with rhGDF5, rhN445T GDF5, rhBMP2, and rhBMP9 was performed at day one for 12 or 72 h.
Micromass cultures were lysed in Trifast (peqlab) after 12 h of stimulations with recombinant human proteins and total RNA was isolated according to the manufacturer's recommendations. cDNA synthesis was performed with Taqman Kit (Applied Biosystems) according to manufacturer's guidelines. Gene expression was assessed by amplification of Nog and β-Actin as endogenous control on a Taqman 7500 (ABI) using SYBR green. The following primer pairs were used: chNog-526f TCTGTCCCAGAAGGCATGGT, chNog-590r CGCCACCTCAGGATCGTTAA, chActin-410f CAACAGAGAGAAGATGACACAGATCA, chActin-484r ACAGCCTGGATGGCTACATACA.
The myoblastic mouse cell line C2C12 (ATCC) and C2C12 stably overexpressing Bmpr1b (C2C12-Bmpr1b; [37]) were seeded in 96-well plates at a density of 1.5×104/well in growth medium (high-glucose DMEM, 10% FCS, and 2 mM L-Gln in 10% CO2). 24 h later they 20 were stimulated in starved medium (2% FCS) for 72 h with recombinant proteins. The measurement of ALP activity was performed by pNPP as previously described [16].
The BIA2000 system (Biacore) was used to analyze the binding affinity of rhGDF5 and rhN445T GDF5 to immobilized receptor ectodomains of BMPR1A and BMPR1B as described previously [16].
The luciferase reportergene assay was performed with minor changes as previously described [38]. C2C12 cells were grown in DMEM high glucose with 10% FCS. SBE-pGL3 [39] and pRL-Tk (Promega) transfected C2C12 cells were seeded at a density of 1*105 cells/96-well and stimulated one day later for 15 hours with recombinant proteins in DMEM high glucose with 2% FCS. Luciferase activity was determined using the Dual-Glo Luciferase Reporter Assay System (Promega).
Injections of RCAS(BP)A viruses into HH10 chick embryo hind limb fields was performed as previously described [36],[40]. Afterwards embryos were harvested at day 7.5 (HH31–33) and fixed in 100% ethanol. Skeletal preparation and staining was performed with Alcian blue and Alizarin red [40]. Every construct was injected at least in 60 embryos. The effect of rhGDF5, rhN445T GDF5, rhBMP2, and rhBMP9 in vivo on Nog expression was determined by implantation of heparin-acrylic beads (Sigma-Aldrich) soaked with recombinant human proteins in chicken limb buds. The beads were implanted between stages HH25–27, in case of rhBMP9 at stage HH29/30 as described [30]. After 20 to 22 h the embryos were harvested and fixed in PFA. Every protein was implanted at least in 10 embryos.
Fixation of chicken embryos was performed with 4% PFA. Whole mount in situ hybridization was performed as previously described [41]. DIG-labeled probe of chNog [42] was detected with BMPurple (Roche).
|
10.1371/journal.pntd.0002901 | Explaining the Host-Finding Behavior of Blood-Sucking Insects: Computerized Simulation of the Effects of Habitat Geometry on Tsetse Fly Movement | Male and female tsetse flies feed exclusively on vertebrate blood. While doing so they can transmit the diseases of sleeping sickness in humans and nagana in domestic stock. Knowledge of the host-orientated behavior of tsetse is important in designing bait methods of sampling and controlling the flies, and in understanding the epidemiology of the diseases. For this we must explain several puzzling distinctions in the behavior of the different sexes and species of tsetse. For example, why is it that the species occupying savannahs, unlike those of riverine habitats, appear strongly responsive to odor, rely mainly on large hosts, are repelled by humans, and are often shy of alighting on baits?
A deterministic model that simulated fly mobility and host-finding success suggested that the behavioral distinctions between riverine, savannah and forest tsetse are due largely to habitat size and shape, and the extent to which dense bushes limit occupiable space within the habitats. These factors seemed effective primarily because they affect the daily displacement of tsetse, reducing it by up to ∼70%. Sex differences in behavior are explicable by females being larger and more mobile than males.
Habitat geometry and fly size provide a framework that can unify much of the behavior of all sexes and species of tsetse everywhere. The general expectation is that relatively immobile insects in restricted habitats tend to be less responsive to host odors and more catholic in their diet. This has profound implications for the optimization of bait technology for tsetse, mosquitoes, black flies and tabanids, and for the epidemiology of the diseases they transmit.
| Tsetse flies and other blood-sucking insects spread devastating diseases of humans and livestock. We must understand the host-finding behavior of these vectors to assess their epidemiological importance and to design optimal bait methods for controlling or sampling them. Unfortunately, mysteries abound in the host-finding behavior of tsetse. For example, it is strange that visual cues are more important for species found in riverine habitats, where dense vegetation restricts the range of visual stimuli, whereas olfactory cues are more important for species occurring in open savannah. To explain this paradox, we used a deterministic model which showed that restricted riverine habitats can reduce tsetse movement by up to ∼70%. This, and the fact that movement increases with fly size, can explain why savannah tsetse, especially the larger ones, rely relatively greatly on olfactory cues, are particularly available to large stationary baits, are repelled by humans, and often investigate baits only briefly without alighting on them. The results also explain why tiny, inexpensive, and odorless baits can control riverine tsetse effectively, whereas larger odor-baited devices are needed against savannah tsetse. These findings have important bearings on the study of host-finding behavior in other blood-sucking insects, including mosquitoes.
| Tsetse flies (Glossina spp.) occupy about ten million square kilometers of sub-Saharan Africa [1]. They feed exclusively on vertebrate blood and, in so doing, transmit those trypanosomes, namely Trypanosoma brucei rhodesiense and T. b. gambiense, that cause sleeping sickness in humans. These trypanosomes, together with others such as T. vivax, and T. congolense cause the disease of nagana in domestic animals. Host location by tsetse [2], [3] is thus a key aspect of disease dynamics. Moreover, understanding the host-orientated behavior of tsetse has led to several cost-effective means of attacking the flies [1], [4], [5], and could have implications for current and prospective methods of controlling mosquitoes, such as the use of bed-nets [6], insecticide-treated livestock [7], odor-baited traps [8] and genetically-modified vectors [9].
The various species of tsetse divide into the so called “forest”, “riverine” and “savannah” groups, of which only the latter two groups are epidemiologically important. The savannah species occupy extensive blocks of deciduous woodland and transmit mostly nagana [1]. whereas the riverine species are important vectors of both nagana and sleeping sickness and typically occur in evergreen woodland near water bodies The two groups of main vectors differ in at least four important ways: (i) savannah flies displace by an average of about 1 km/day [10], while riverine flies displace only about a third as much [11]; (ii) savannah tsetse commonly feed on large hosts such as warthog, kudu and elephant, while small animals such as lizards form much of the diet of riverine tsetse [12]; (iii) the response of savannah tsetse to odor is several times greater than for riverine tsetse [13]; (iv) savannah tsetse are strongly repelled by humans [2], whereas riverine flies are not [14], [15], [16]. These contrasts have led to marked differences between the designs of insecticide-treated screens, called targets, used to control each group [16]. For savannah tsetse the targets are 1–2 m2 and baited with artificial ox odor [17]; for riverine tsetse the targets are as small as 0.06 m2 and used without odor [18].
The distinctions between the behavior of riverine and savannah tsetse seem anomalous. For example, the avoidance of humans by savannah flies is usually attributed to the high risks of feeding on a type of host adept at killing probing insects [2], but the risks should be high for riverine flies too, so why are riverine flies not equally averse to humans? If savannah tsetse rely heavily on odor attraction, why do riverine flies not do so? Moreover, since riverine tsetse feed off small animals and land on tiny targets, why do savannah tsetse disregard such baits [19]. To explain these anomalies we hypothesized that the distinctive responses of riverine and savannah tsetse to baits is associated directly with the way that the overall size and shape of different habitats affect fly mobility, devoid of any distinctions in the innate behavior of the two groups of tsetse. This hypothesis is an extension of the experimental and theoretical evidence that various arrangements of dense bushes inside the habitat restrict the movement of tsetse and so alter the catches at baits [20], [21]. It resonates with indications from studies with other creatures that habitat geometry can be important in a variety of matters such as speciation [22], species coexistence in predator-prey relationships [23], the dynamics of such relationships [24], and population abundance [25].
While much of the behavioral impact of dense bushes within tsetse habitat has been established by experiments in the real world, involving small-scale manipulations of bush arrangements [20], [21], manipulations on a much larger and impractical scale would be required for field tests of the hypothesis that the behavior of tsetse is governed also by the overall size and shape of the habitat. Hence, we used a deterministic model to simulate within a Microsoft Excel spreadsheet the impact that the overall shape and size of habitats, together with the arrangement of bushes within them, has on tsetse displacement, catches at experimental baits, feeding success, host selection, and the efficacy of various types of target.
There were no ethical issues since all work was theoretical.
The spirit of the modelling was that a cohort of flies that had started its feeding cycle moved about the habitat, encountering visual and/or odor cues from various natural or artificial baits and then fed on the baits, or was killed by them, with a probability appropriate for each bait type. Flies fed or killed at various times during the cycle were accumulated and removed from the simulation.
The ability to find stationary baits depends largely on displacement rate [19]. The principles applying to this rate were elucidated by seeding flies in the central cell of a band or block of good habitat and allowing them to execute the average daily allocation of 1000 steps, in the absence of natural death or removal by baits. Blocks were in a checker-board arrangement with poor habitat so that flies could diffuse between blocks of good habitat, albeit slowly. Bands were flanked by no-go areas to focus only on movement within the band. The results with different widths of blocks and bands indicate that at widths of 10 m the displacement was only 43–64% of the displacement in homogeneous habitat (Fig. 2). The figures increased with increasing widths, but were still only 76–85% at widths of 450 m. At any given width, the displacement in a block was less than in a band. The complex curve for blocks was associated with the change in the ratio of perimeter to area, and hence the proportion of flies located where they could step out of the block.
To assess the effect of heterogeneity within the overall shapes of habitats, cells of no-go vegetation simulating impenetrable bushes [20], [21], were located within habitats of various shape. Findings from simulations with a variety of bush arrangements are exemplified (Fig. 3) by data for a 50 m-wide band with either no bushes, or each of four different bush arrangements, and for a large block composed of such bands placed parallel and adjacent to each other, with the adjoining parts of each band being mirror images. The rate of displacement tended to decline as: (i) numbers of bushes increased, (ii) flight paths between dense vegetation became more tortuous, and (iii) the abundance of dead-ends rose, so that the flies expended much flight on retracing their steps. Although real bushes in the field are unlikely to show the sort of serially repeated arrangements modelled above, the overall effects are likely to be similar.
Allowing that riverine habitat occurs in bands or small blocks, and is often more densely bushed than savannah, the above results match field observations that tsetse displacement is greatest with savannah tsetse [10], [11], [39]. For simplicity, subsequent modelling assumed that all habitats contained no dense bushes. With that assumption the differences found between the efficacy of baits in riverine habitats and large blocks of savannah tend to be conservative indications of real differences.
The relative importance of visual and olfactory stimuli is commonly estimated in the field by comparing catches from a host animal with those from an odorless model animal of the same size [2], [40]. In simulating such experiments, the two types of bait were operated for six days in a crossover design, alternating between sites that were sufficiently far apart to ensure that the baits there did not compete with each other. The baits were present for half of the daily step periods each day, consistent with the fact that field catches of tsetse are often made in the afternoon only [2]. The simulated catch with each bait was expressed as a percent of the initial abundance of tsetse per square kilometer of the good habitat, and the efficacy of odor relative to visual stimuli was taken as the percent by which the addition of odor increased the catch above that with visual stimuli alone. As expected, catches and odor efficacy increased with bait mass (Table 2). Intriguingly, catches declined markedly on going from the large block of habitat to the bands, but the decline was greatest with the large baits and when odor was used. Consequently, bait size was relatively unimportant in the bands, and the percent efficacy of odor in the narrowest band was around a quarter of the efficacy in the large block. Similar indications were produced when the baits were operated in habitat restricted to small blocks. For example, when the block consisted of just one cell, the catch with the lizard was >99.9% of the catch with the elephant and percent efficacy of odor was <0.1% with either animal.
Outputs for the percent efficacy of odor in the large block accord well with field data for savannah flies. For example, for G. m. morsitans and G. pallidipes in the field, the relative efficacies of odor with an ox (454 kg), donkey (204 kg) kudu (136 kg) warthog (82 kg) and bushpig (73 kg) averaged 435%, 175%, 89%, 56% and 73%, respectively [2]. More remarkably, outputs for the bands or small blocks accord well with the limited field efficacy of odor against riverine tsetse [13], despite the model's provisions that the innate responsiveness and mobility of flies in the bands was exactly the same as in the large block. Hence, habitat geometry, irrespective of any innate behavioral distinctions, can account for most differences between patterns of field catches of savannah and riverine tsetse.
Simulations were made with various densities of large and tiny targets (Fig. 1) operated continuously in a large block or 10 m-wide band. As in field campaigns against riverine tsetse, tiny targets were used without odor, but large targets were modelled with and without artificial ox odor, according with the field use of large targets against savannah and riverine flies, respectively. In keeping with field catches at targets [17]–[19], the numbers of targets required to achieve a given rate of kill differed greatly between the large block and the band (Fig. 4). To interpret the outputs it can be taken that an imposed death rate of about 4% per day, or 12% per feeding cycle, reduces field populations of tsetse by 99.99% per year, leading to population elimination in the absence of invasion [16]. On that basis, outputs accord with field indications for the numbers of various sizes of target needed to control savannah [41] and riverine [42] tsetse, and for the efficacy of odor with targets in savannah [2] and riverine [43] habitats. Hence, the results offer further support for the hypothesis that habitat geometry, not differences in innate behavior, determines much of the distinctive availabilities of riverine and savannah tsetse.
To explore the abilities of various sizes and population densities of hosts to support the tsetse population, it was assumed that flies fed only on those stationary hosts that the model introduced, so no allowance was made for feeding on any other animals. Feeding success was scored after four days when fed flies had replenished their food reserves after an average of around three days, i.e., the normal length of the hunger cycle. It was also scored after six days, when flies were about to die of starvation. Since some flies died of causes other than starvation, percent feeding success could not reach a full 100%.
As expected from the above work with targets and simulated field catches, the host numbers required to allow a given level of feeding success were much greater in a narrow band than in the large block, and the efficacies of the various hosts differed greatly in the block but relatively little in the band (Fig. 5). Thus, in the large block, about 15–30 lizards led to the same feeding success as one elephant, but in the band only about 2–3 lizards were required.
As in other modelling [44], the number of flies discovering hosts decreased substantially when hosts were grouped instead of being singly and evenly distributed. Consider, for example, a population of lizards at an overall density of 100/km2 in a band of habitat 10 m wide. When the lizards were distributed singly and evenly the 4-day feeding success was 25%, but dropped to only 2% when the lizards occurred in evenly distributed groups of four, with each group involving a lizard in each cell of a line of four cells along the axis of the band of habitat. In a large block of habitat the comparable figures for feeding success were 65% for lizards distributed singly, as against only 11% for the grouped lizards.
The outputs (Fig. 5) are consistent with the abilities of known host populations to support tsetse. Thus, savannah tsetse at Sengwa, Zimbabwe, were maintained by a mixed population of hosts comprising an average of ten warhogs, plus two elephants and several kudu and other bovids per square kilometre [45]. Moreover, the model's indications that tsetse in restricted habitats can be supported largely by small hosts such as lizards, with population densities of around 50–100/km2 [46], agree with the frequency of lizards and other small creatures in the blood-meal identifications of riverine tsetse [12].
Mobility has thus far been assumed to be the same for all flies. However, female tsetse displace at a greater rate than males [10]; young flies with poorly developed flight muscles [47] and old flies with damaged wings displace relatively little, and daily flight times can double or halve according to seasonal temperature [26]. To simulate this variability, the daily number of flight steps was increased or decreased threefold.
As expected, the greater the mobility of flies the sooner they fed. However, it was more instructive to consider what this implied about the extent to which flies could afford to be selective about feeding on hosts they encountered. To explore this, the model's map was provided with an even spread of hosts. At different points in the feeding cycle, calculations were then made of the probability that flies that did not feed at that point would die of starvation. In any given habitat, and with any given size and abundance of host, this probability increased with the number of host-searching days completed. It increased also with a reduction in the number of step periods allowed per day and was greater in the narrow band than in the large block. The latter phenomena are illustrated by considering outputs with kudu at 16/km2, which represents roughly the abundance and mean size of the main hosts, i.e., warthogs, elephants and kudu, that sustained the tsetse population in the savannah at Sengwa [45], discussed above. Simulations were also made with host populations consisting of lizards at 100/km2, to be closer to a host situation more typical of riverine habitats [46].
The results show that tsetse in large blocks of habitat can afford to feed much more selectively than when they are in a restricted habitat carrying the same types and abundance of hosts (Table 3). The comparison between real riverine and savannah areas will depend crucially on the numbers and sizes of hosts present in each situation, and on the intrinsic mobility of the tsetse present. However, the principles are established that a reduction in the innate mobility of tsetse, and the limits that restricted habitats impose on host location, can greatly favor a strategy of feeding on any host encountered.
The host-oriented behavior of tsetse is arguably better understood than that of any other blood-sucking insect [13], [48], allowing models of bait-finding to employ a wealth of empirical data as inputs and for output validation. Our model indicates that distinctions between riverine and savannah tsetse in respect of daily displacement and availability to various sizes of visual bait and odor plume are due largely to the immediate circumstantial effects of habitat geometry, rather than evolved differences in innate behavior. This indication must arise with any model that approaches reality since output patterns will be set by the following five principles. First, in restricted habitats the full benefit of stimuli from large baits is lost because some of the ambit of the stimuli covers places devoid of flies. This problem is especially severe with small blocks, as against bands, since the stimuli can go out of the block on all four sides. Second, even if stimuli from large baits do not go out of a small patch of habitat, the effective advantage of seeking large hosts is reduced because random flight within the patch ensures that a relatively small host there can be discovered before long. Third, the more restricted the space that tsetse occupy the less readily can they diffuse from their start point, so reducing their probability of finding a distant bait. Fourth, at any given density of baits, the more attenuated the habitat the greater the mean distance between flies and the nearest bait. Thus, if bait density is 100/km2 the average distance between flies and the nearest bait in large blocks is about 40 m, as against 250 m in a band 10 m wide. Likewise, an extensive ambit of bait stimuli can reduce substantially the mean distance the flies must displace to detect the bait in the large block, whereas it can reduce this distance in the band by relatively little. Finally, the time taken to travel any given distance by diffusive movement is proportional to the square of the distance [11].
Despite the immediate importance of habitat geometry, different species are likely to have evolved some innate behavior patterns suiting the distinctive demands of finding food in their particular environments. Any innate differences might relate not so much to means of locating hosts but rather to the response adopted after discovering hosts of various type, particularly men as against more tolerant, and less dangerous, hosts. Modelling suggests that the high mobility of tsetse in homogeneous and extensive habitats, and the comparative ease of finding hosts there, means that unless savannah tsetse are about to die of starvation they should be anthropophobic, in accord with field observations [2], [3], [49]. The corollary is that the anthropophily of riverine tsetse [15] is due to the poor mobility of flies in restricted habitats and the associated difficulties of finding safer hosts. In any event, the less a fly displaces the more important it is to investigate any host thoroughly before rejecting it, implying that in such circumstances the flies will remain longer with a host and be less discerning about alighting on it. Moreover, flies with low movement rates must rely on ‘ambushing’ passing hosts, as against active searching.
Our results suggest the possibility of reducing the wide variety of host-orientated behavior to a unifying framework applicable to both sexes and all species of tsetse in all habitats, including the many forest-group species not modelled here. The development of such a framework requires further theoretical and experimental attention. Nevertheless, host location must depend largely on displacement rates which affect: (i) effectiveness of odor attraction, (ii) reliance on small, abundant and solitary hosts, (iii) performance of small targets relative to large, (iv) repellence of humans, (v) importance of stationary as against mobile baits, and, (vi) persistence near hosts and the strength of alighting responses.
The magnitude of each of these phenomena is expected to be governed by (i) the width and length of the overall habitat, (ii) proportion of habitat that allows free flight, (iii) fly size, since innate displacement potential increases with size and (iv) proportion of the fly's energy available for flight [47]. Host-finding is likely to be influenced also by parameters other than those governing displacement. For example, changes in vegetation affect the length and structure of odor plumes [50], [51]. Nonetheless, the above four parameters, among which habitat geometry seems very important, could go far towards rationalizing much of the apparent variety of tsetse behavior. Empirical support for a unifying framework is provided by results from three sources.
First, some of the most comprehensive data for savannah tsetse come from Rekomitjie, Zimbabwe. The biggest fly present, female G. pallidipes, is twice the size of the smallest, male G. m. morsitans. In accord with expectation, the larger flies are the most mobile [10], the most available to stationary odor baits, the most repelled by humans [2], the least available to tiny, as against large, targets [19], the least persistent and the least likely to alight [2].
A second source of support is provided by several studies of tsetse that occupy habitats atypical of their group. Thus G. longipennis, of the forest group, occupies savannah and in keeping with its large size and habitat, is as mobile as G. pallidipes [52], is repelled by humans and readily available to host odor [53]. In expected contrast, G. brevipalpis, a large forest species which has remained in forest, is less available to odor [54]. The smallest tsetse, G. austeni, is a savannah-group fly found in coastal thickets. In accord with its small size and dense habitat, its availability to odor is much less than for other savannah species [54]. The riverine fly, G. tachinoides, lives in relatively open habitats and is relatively responsive to odor [55], albeit not as much as other tsetse living in savannah – as predicted since it is smaller than such tsetse.
Third, and perhaps the most telling, studies of the riverine tsetse, G. fuscipes fuscipes, near Lake Victoria in Kenya, showed that adding odor to traps was ineffective in narrow (5–10 m wide) forest habitats but doubled catches in a larger block of forest covering 1.4 km2 [56]. Presumably, the closeness of the habitats ensured that they contained flies with the same innate responsiveness.
While the outputs of the model and the predictions of the unifying framework fit well with existing field data, there is a need for new field experiments specifically aimed at confirming and extending present indications. For example, it would be particularly informative to elucidate the response of riverine species of tsetse to visual and olfactory stimuli under circumstances not expected to limit the expression of such responsiveness. One approach would be to study further the behavior of riverine tsetse in large blocks of woodland [56]. Another approach is suggested by the expectation that the catches in the first few minutes of the exposure of a bait depend primarily on the responsiveness of flies already in the ambit of the bait's stimuli, whereas the later catches are governed by the way that habitat size and shape govern the rate at which tsetse diffuse into that ambit from far away. Hence, to highlight the basic responsiveness to bait stimuli in habitats that reduce fly diffusion, it would be pertinent to accumulate the catches of a bait that appears for brief periods interspersed with longer periods in which the baits are hidden while flies move in to re-populate the vicinity [20]. The time needed to produce such re-population would itself be of interest in indicating the rates of fly movement [10]. A further approach would be to use a bait that moves to a succession of stations a short distance apart, stopping at each just long enough to recruit flies from the area covered by the odor plume. Indeed, such minor movement and stopping would come closer than any research yet done to duplicate the common behavior of natural hosts.
The simulations offer support for using tiny odorless targets to control riverine tsetse in restricted habitats [18] but warn that in broader habitats such as those that can occur in mangrove ecosystems, a larger target with odor might be more cost-effective. Our results confirm that relatively high densities of targets are needed per unit area of habitat to control riverine tsetse, but these high densities are offset by the fact that such habitats cover a small proportion of the land surface. Thus, in places where people and livestock need to be protected against disease during visits to infested localities, the target density required per total land surface tends to be small, at around 7/km2 (Torr and Lehane, unpublished).
While aversion to humans seems to be the main reason why savannah tsetse are minor vectors of sleeping sickness today, they might become more important if climatic or anthropogenic change restricts tsetse habitat.
The relationship between habitat and host-finding in tsetse is likely to apply to other blood-sucking insects. While data are less extensive for other insects, there are indications that differences are consistent with expectations. For instance, horse flies, stable flies, and blackfly living in extensive woodlands [48] are highly responsive to host odors whereas in riverine habitats near Lake Victoria these species show the same type of pattern as for tsetse in riverine [56]. Malaria mosquitoes inhabiting savannah woodland (Anopheles arabiensis, [40] and extensive wetlands (Anopheles melas, [57], [58]) are also highly responsive. On the other hand, bird-biting species of Culex [59], and Aedes aegypti (the vector of dengue virus) in urban settings [60], seem much less responsive. We suggest that the restricted and heterogeneous habitats of tree canopies and urban environments reduces mobility in much the same way that riverine habitats affect tsetse. Field studies to explore this hypothesis could provide important new insights into the transmission dynamics and control of West Nile and dengue viruses transmitted by Culex pipiens and Aedes aegypti, respectively.
|
10.1371/journal.pcbi.1004202 | Deciphering Signaling Pathway Networks to Understand the Molecular Mechanisms of Metformin Action | A drug exerts its effects typically through a signal transduction cascade, which is non-linear and involves intertwined networks of multiple signaling pathways. Construction of such a signaling pathway network (SPNetwork) can enable identification of novel drug targets and deep understanding of drug action. However, it is challenging to synopsize critical components of these interwoven pathways into one network. To tackle this issue, we developed a novel computational framework, the Drug-specific Signaling Pathway Network (DSPathNet). The DSPathNet amalgamates the prior drug knowledge and drug-induced gene expression via random walk algorithms. Using the drug metformin, we illustrated this framework and obtained one metformin-specific SPNetwork containing 477 nodes and 1,366 edges. To evaluate this network, we performed the gene set enrichment analysis using the disease genes of type 2 diabetes (T2D) and cancer, one T2D genome-wide association study (GWAS) dataset, three cancer GWAS datasets, and one GWAS dataset of cancer patients with T2D on metformin. The results showed that the metformin network was significantly enriched with disease genes for both T2D and cancer, and that the network also included genes that may be associated with metformin-associated cancer survival. Furthermore, from the metformin SPNetwork and common genes to T2D and cancer, we generated a subnetwork to highlight the molecule crosstalk between T2D and cancer. The follow-up network analyses and literature mining revealed that seven genes (CDKN1A, ESR1, MAX, MYC, PPARGC1A, SP1, and STK11) and one novel MYC-centered pathway with CDKN1A, SP1, and STK11 might play important roles in metformin’s antidiabetic and anticancer effects. Some results are supported by previous studies. In summary, our study 1) develops a novel framework to construct drug-specific signal transduction networks; 2) provides insights into the molecular mode of metformin; 3) serves a model for exploring signaling pathways to facilitate understanding of drug action, disease pathogenesis, and identification of drug targets.
| A deep understanding of a drug’s mechanisms of actions is essential not only in the discovery of new treatments but also in minimizing adverse effects. Here, we develop a computational framework, the Drug-specific Signaling Pathway Network (DSPathNet), to reconstruct a comprehensive signaling pathway network (SPNetwork) impacted by a particular drug. To illustrate this computational approach, we used metformin, an anti-diabetic drug, as an example. Starting from collecting the metformin-related upstream genes and inferring the metformin-related downstream genes, we built one metformin-specific SPNetwork via random walk based algorithms. Our evaluation of the metformin-specific SPNetwork by using disease genes and genotyping data from genome-wide association studies showed that our DSPathNet approach was efficient to synopsize drug’s key components and their relationship involved in the type 2 diabetes and cancer, even the metformin anticancer activity. This work presents a novel computational framework for constructing individual drug-specific signal transduction networks. Furthermore, its successful application to the drug metformin provides some valuable insights into the mode of metformin action, which will facilitate our understanding of the molecular mechanisms underlying drug treatments, disease pathogenesis, and identification of novel drug targets and repurposed drugs.
| Most drugs exert their therapeutic actions through interactions with specific protein targets. These target proteins are dominated by two categories: enzymes that catalyze reactions essential for the functioning of organisms, and receptors that transmit signals by interacting with messenger molecules [1,2]. The interactions of drugs and their targets initiate the signal transduction cascade that is usually propagated by the involved proteins and multiple pathways. These proteins and pathways act in the mode of crosstalk networks [3]. The process of such signaling transduction converts the chemical signals to a specific cellular response such as gene expression, cell division, and inhibition of cell death and apoptosis [4]. The signaling cascade usually ends at the recipients of chemical signals such as transcription factors (TFs), which have specific binding sites on DNA and play critical roles in the gene expression regulation [5]. In complex diseases such as cancer [6,7], neuropsychiatric disorders [8], and diabetes [9], these molecules involved in the signal transduction cascade that are altered and, thus, become attractive targets for disease treatment [10,11]. Therefore, targeting signaling pathways has become an important approach to discovering new drugs through traditional experimental methods [12,13] and to predicting drug repositioning through systematic approaches [14]. However, the primary challenge for utilizing signal transduction pathways for drug discovery is to synopsize the drug signaling pathways into one comprehensive system, including the major causal genetic factors for pathology of the complex disease and the most elemental components in the drug action.
Recent high-throughput technologies such as array-based mRNA and microRNA expression, genome-wide association studies (GWAS), and next-generation sequencing (NGS) have provided massive amounts of data, enabling investigation of drug effect through pharmacogenomic network approaches. For example, the Connectivity Map (CMap, build 02) studied the effect of 1,309 small chemicals on gene expression in four cultured human cells [15]. Furthermore, multiple reliable drug-centered databases such as DrugBank [16], KEGG (Kyoto Encyclopedia of Genes and Genomes) DRUG [17], PharmGKB (The Pharmacogenomics Knowledge Base) [18], and STITCH (Search Tool for Interactions Chemicals) [19], provide comprehensive and detailed drug information for computational discovery and/or drug design. Therefore, it is possible to integrate known drug targets, genes involved in drug pharmacokinetics (PK) and pharmacodynamics (PD) processes, drug-induced gene expression data, and disease-gene associations. Additionally, network-assisted approaches have become powerful tools to explore disease-gene, gene-gene, as well as drug-target associations in pharmacology and human disease [20–23]. Therefore, we hypothesized that the construction of a signaling pathway network to connect the upstream components and downstream signal recipients for an individual drug would increase power to identify genes that play critical roles in drug action or disease development.
In this study, we develop a computational framework, called DSPathNet, to construct one signaling pathway network (SPNetwork) for a particular drug via amalgamating drug knowledge with drug-induced gene expression data. The main purposes are to capture the principal components in the drug signal transduction process and to provide an alternative approach to identifying critical elements and modules (subnetworks) relevant to drug action. We illustrate the utility of DSPathNet using the metformin, one of the most widely prescribed anti-diabetic drugs in the world which has been recently shown to be useful for cancer treatment and prevention in people at higher risk [24–26]. We started with the collection of known drug-related genes and inference of TFs from metformin-induced gene expression data. Considering that most of the known drug-related genes participate in PK and PD processes and are located in the upstream of the signaling cascade based on their function, we defined them as “metformin upstream genes.” Likewise, we defined the TFs that receive and transmit the chemical signals at the end of the signaling cascade as “metformin downstream genes.” After overlaying the two sets of genes onto human SPNetwork, we employed random walk algorithms to construct a metformin-specific SPNetwork. The random walk-based methodology aims to identify the pathways that are closet to the known disease genes compared to other methods [27] and offers the best predictive performance [28]. The network is expected to enrich with signaling genes involved in metformin signal transduction. We performed the comprehensive gene enrichment analyses of the network using the disease genes of type 2 diabetes (T2D) from GWAS catalog [29], cancer genes from Cancer Gene Census [30], one T2D GWAS [31], three cancer GWAS [32,33], and one novel GWAS of cancer patients with T2D using metformin from BioVU [34]. The enrichment analysis results showed that the network contained a significant number of T2D and cancer disease genes and genes related to metformin action, indicating that the framework is promising as a method to identify critical genes involved in disease pathology and drug action. Additionally, the metformin-specific SPNetwork generated here provides potential metformin targets and molecular insights for further delineating the mechanism of metformin action.
In this study, we develop a novel computational framework to build a Drug-specific Signaling Pathway Network, namely DSPathNet, for constructing a signaling pathway network (SPNetwork) for an individual drug of interest. The drug-specific SPNetwork is expected to contain critical components in the drug’s signal transduction cascade. These components are genes that harbor genetic variations contributing to the pathology of the drug indication or drug response. Thus, the drug-specific SPNetwork would facilitate our understanding of the molecular mechanisms of drug action, disease pathogenesis, and identification of novel drug targets. To prove the principle, we utilized the drug metformin as an example to evaluate the framework.
Fig 1 outlines the framework to build the metformin-specific SPNetwork and S1 Table summarizes the data sources, software and evaluation data used in the study. Briefly, we first collected metformin upstream genes from multiple sources and inferred metformin downstream genes from metformin-induced gene expression data. We compiled a human SPNetwork from two databases, Pathway Commons [35] and TRANSFAC [36], as a background pathway system for all signal transduction processes in humans. To weight the association of each node with metformin action, we assigned a functional similarity score to each node based on their Gene Ontology (GO) annotations and metformin upstream genes. The human SPNetwork included 37,881 edges and 4,367 nodes. Then, we utilized the metformin upstream and downstream genes as seeds to produce the metformin-specific SPNetwork from the human SPNetwork via random walk approaches. In this process, we applied a crossing network strategy to generate the drug-specific SPNetwork from background human SPNetwork by longitudinal and lateral movements. Finally, we computationally evaluated the metformin-specific SPNetwork by examining the enrichment of genes in the network using two types of data. The first includes the disease genes of type 2 diabetes (T2D) and cancer, the two diseases in which metformin has been actively studied. The second contains the individual genotyping data from five GWAS datasets: one T2D GWAS dataset, three cancer GWAS datasets, and one GWAS dataset of cancer patients with T2D treated by metformin. Our evaluation results indicated that the metformin-specific SPNetwork was significantly enriched with genes with mutations that could contribute to the pathology of T2D and cancer, and genes that may be associated with metformin-associated cancer survival (Table 1). To further investigate the molecular mechanisms underlying metformin action, we built a crosstalk subnetwork based on common genes to T2D and cancer, network topology, and functional analyses. We revealed several critical components, modules, and pathways that might be involved in metformin action.
In order to generate a complete and reliable SPNetwork, we extensively collected the metformin related genes, rigorously selected the expressed genes induced by metformin, and comprehensively compared the performance using T2D GWAS data after the SPNetwork generation. For each step, we provide the detailed information as below.
The final metformin-specific SPNetwork generated above comprised 477 nodes and 1,366 edges (Fig 3B, S5 Table). Among the 477 nodes, 215 belonged to metformin upstream network, while 303 belonged to metformin downstream network. There were 41 bridge nodes between them. Thus, 174 genes were unique to the metformin upstream network, and 262 genes were unique to the metformin downstream network. From here, we refer to the three gene sets as upstream genes (number of genes: 174), downstream genes (262), and bridge genes (41) for follow-up network topological and functional analyses.
To explore the topological properties of this SPNetwork, we calculated node degrees (connectivity) and their distribution [41]. In this network, degree values of nodes ranged from 1 to 79 and the average degree was 5.73. The degree distribution was strongly right-skewed, indicating that most nodes had a low degree and only a small portion of the nodes had a high degree (Fig 3C). The nodes with a high degree act as hubs in the network and hold the whole network together [41]. In biological networks, hubs are more likely to be essential genes [42] and disease genes [43–45]. Using the hub defining method proposed by Yu et al. [46], we determined 38 hubs whose degrees were larger than 14. Among them, one gene (PPARG) belonged to both metformin upstream and downstream gene sets, two genes (TP53 and SREBF1) were metformin upstream genes, 13 belonged to the metformin downstream gene set only, and 22 were novel genes. After extracting these hubs from metformin-specific SPNetwork, we generated a hub-centered subnetwork (Fig 3D). Among the 38 hubs, 19 (50.00%) are included in ‘pathway in cancer’ and 9 (23.68%) in ‘MAPK signaling pathway’ according to KEGG pathway annotation. The MAPK signaling pathway plays important roles in the pathology of both cancer [47] and diabetes [48]. Thus, the 477 genes had two genes belonging to metformin upstream and downstream genes, 33 to the metformin upstream genes, 58 to the metformin downstream genes, and 384 novel genes (S6 Table). The novel genes may provide a valuable resource for further investigation of the pathology of T2D and cancer, and the metformin action.
We further examined pathway enrichment in these 477 nodes based on KEGG pathway annotation using the online tool WebGestalt [49]. We identified 69 significant pathways (adjusted P-value < 1.00 × 10–4) (S7 Table). According to the KEGG pathway first-level category annotation (Materials and Methods), 12 pathways belonged to ‘environmental information processes,’ nine to ‘cellular processes,’ 18 to ‘organismal systems,’ and 29 to ‘human disease.’ Among these 12 environmental information processes pathways, eight were signal transduction pathways, of which the top three pathways were ‘MAPK signaling pathway’ (32genes, adjusted P-value: 3.39 × 10–22), ‘mTOR signaling pathway’ (13 genes, adjusted P-value: 6.39 × 10–14) and ‘ErbB signaling pathway’ (15 genes, adjusted P-value: 1.89 × 10–13). Among the 18 pathways related to organismal systems, five belonged to the endocrine system, of which the top three pathways were ‘adipocytokine signaling pathway’ (22 genes, adjusted P-value: 3.19 × 10–25), ‘PPAR signaling pathway’ (22 genes, adjusted P-value: 5.36 × 10–25), and ‘insulin signaling pathway’ (23 genes, adjusted P-value: 1.91 × 10–19). Among the 29 pathways related to human disease, 15 were directly related to cancer. Importantly, the pathway ‘type II diabetes’ (10 genes, adjusted P-value: 1.12 × 10–10) and the ‘maturity onset diabetes of the young’ (8 genes, adjusted P-value: 1.94 × 10–10) were among the enriched pathways. Together, the evidence indicates that the metformin-specific SPNetwork involves both diabetes and cancer at the pathway level.
In the metformin-specific SPNetwork, there were 41 genes (bridge genes) common to both the metformin upstream and downstream networks. As mentioned above, most of them (31, 75.6%) were novel linkers (S6 Table). To interrogate their roles, we compared them with upstream genes (174) and downstream genes (262) via network topological and functional analyses, as described below.
Since metformin is a well-studied drug for T2D treatment, the metformin-specific SPNetwork was expected to contain genes that have genetic association with T2D. To examine this expectation, we comprehensively performed enrichment analysis using two sets of genes. The first one contained 131 genes collected from 66 T2D GWAS studies curated by the NHGRI GWAS Catalog database (April 1, 2014) [29]. Those genes have been reported to be significantly associated with T2D based on GWA studies. Here, we selected these genes having at least one SNP with P-value less than 1.0 × 10–8 as T2D associated genes. The second set included the T2D-related genes from the WTCCC T2D study [31] as mentioned above.
Among the 477 nodes in the metformin-specific SPNetwork, 11 genes were found in the first set of 131 genes. Compared to the human protein-coding genes (20,716), the network was significantly enriched for T2D associated genes (Hypergeometric test, P-value: 1.36 × 10–4). Similarly, among the 131 T2D disease genes, 43 existed in the human SPNetwork (4,367). Thus, compared to all nodes in the human SPNetwork, the metformin-specific SPNetwork was significantly enriched for T2D associated genes too (P-value: 3.62 ×10–3). These 11 genes were CDKN2B, HNF1A, HNF4A, IRS1, ITGB6, KCNJ11, LEP, PPARD, PPARG, SND1, and TCF7L2. Among them, KCNJ11, PPARG, and TCF7L2 have the strongest genetic association among genes that appear in the T2D GWAS studies based on a comprehensive review [59].
Among the 477 genes in metformin-specific SPNetwork, 445 had genotyping data from WTCCC T2D GWAS dataset. Among them, 169 genes belonged to T2D-related genes. Compared with all genes with genotyping data in the GWAS, the network was significantly enriched with T2D-related genes (Hypergeometric test P-value: 3.08 ×10–5). We further compared the 169 genes with the genes having genotyping data in the human SPNetwork. Among the 4,367 nodes in the human SPNetwork, 3,446 genes had genotyping data, in which 1,048 genes were T2D-related genes. Thus, the metformin-specific SPNetwork was significantly enriched for the T2D-related genes as compared to the whole human SPNetwork in this study (P-value: 7.47 ×10–3). Fig 5A shows the comparison of P-value distributions of genes in whole GWAS data (T2D GWAS), human SPNetwork, and metformin-specific SPNetwork. These comparisons indicate that the network is enriched with genes that might be involved in the pathology of T2D.
We further generated a subnetwork for 169 nominally significant genes with T2D (Fig 5B) by their direct links. Among the 169 genes, 50 genes had SNPs whose P-values were less than 0.01 in the WTCCC T2D GWAS. In addition, there were six genes observed in both the 131 GWAS Catalog genes and the 169 genes; they are CDKN2B, ITGB6, KCNJ11, PPARD, PPARG, and TCF7L2. Among them, the SNP rs4506565 in gene TCF7L2 has the strongest significance (P = 5.68 ×10–13). TCF7L2 encodes a transcription factor that regulates the transcription of several genes. It is a key element in the WNT signaling pathway, which has been reported to contribute to T2D risk significantly [59].
Above pathway analysis indicated that the metformin-specific SPNetwork was significantly associated with cancer-related pathways. Here, we further examined if the SPNetwork is enriched with cancer genes from four data sets. The first one included 509 cancer genes downloaded from the Cancer Gene Census (December 11, 2013, http://cancer.sanger.ac.uk/cosmic/census). Among them, 64 genes were included in the metformin-specific SPNetwork. Compared to all human genes or the protein-coding genes in the human SPNetwork, the network was significantly enriched with cancer genes (Hypergeometric test, P-value: 1.64 × 10–29 and 6.48 × 10–8, respectively). Interestingly, 3 of the 64 genes (HNF1A, PPARG, and TCF7L2) were in the T2D GWAS Catalog gene list, and 21 genes belonged to 169 T2D-related genes (see above). This observation strongly indicates that metformin may affect the shared genetic risk factors between T2D and cancer. Such information provides clues for how metformin acts in T2D and cancer treatments. This observation also provides evidence for epidemiological studies of metformin in both T2D and cancer [50].
Additionally, we performed the GSEA of the metformin-specific SPNetwork using three cancer GWAS datasets from the Cancer Genetic Markers of Susceptibility (CGEMS) projects (breast cancer [32], pancreatic cancer [33], and prostate cancer [32]). Table 1 summarizes the corresponding gene numbers in each GWAS dataset. Compared with all genes with genotyping in each GWAS dataset, the metformin SPNetwork was slightly significantly enriched in nominally significantly associated genes (Hypergeometric test P-values: 0.0144, 0.0120, and 0.0053 for breast, pancreatic, and prostate cancer, respectively). Though the results of these statistical tests are not as robust as that of the genotyping data from the T2D GWAS study, the results confirm that the metformin-specific SPNetwork was enriched with genetic factors associated with cancer development.
From above analyses, the metformin-specific SPNetwork is enriched with genes associated with T2D and cancer. Several studies over the last few years have demonstrated that patients using metformin have reduced cancer risk and improved cancer survival in T2D patients [24,26,60,61]. Thus, we evaluated whether metformin-specific network enrich genes associated with cancer survival among cancer patients with T2D using metformin. In this study, we took advantage of GWAS data of cancer subjects with T2D treated with metformin from BioVU [34,62] (Materials and Methods). Hereafter, this dataset is referred as “metformin GWAS.” Among the 477 nodes in the metformin-specific SPNetwork, 458 genes had genotyping and 177 genes were nominally significantly (P-value < 0.05) associated with T2D with better survival. Compared with all genes with genotyping data in the metformin GWAS data, the metformin-specific SPNetwork was enriched with nominally significant genes too (Hypergeometric test, P-value: 0.0181). We further compared the P-value distribution of metformin GWAS data for three gene sets: the metformin-specific SPNetwork, human SPNetwork, and all genes in metformin GWAS data set (S6 Fig). The genes in the metformin SPNetwork had the highest proportion of P-values (P-value < 0.05) in metformin GWAS data at the gene level.
Among the 177 genes, 81 genes were included in the 169 genes whose smallest P-values were less than 0.05 in T2D GWAS data. While most of them did not link to each other (S7 Fig), these 81 genes directly linked to other 175 genes to form a subnetwork that included 256 nodes and 910 edges. This feature indicated that the 81 genes and their direct interactors dominated the metformin-specific SPNetwork. For example, the 256 nodes accounted for 53.7% of all nodes and the 910 edges accounted for 66.6% of all edges in the metformin-specific SPNetwork. Additionally, among the 81 genes, 17 belonged to ‘pathway in cancer’: COL4A1, COL4A2, ERBB2, GLI3, ITGB1, MECOM, MMP1, PLD1, PRKCA, RARB, RXRG, SMAD3, TCF7L1, TCF7L2, TGFA, TGFB2, and ZBTB16. Collectively, the above observations indicate that the network was enriched in genes that might contribute to overall survival among cancer patients with metformin therapy.
From above analyses, we observed that the metformin-specific SPNetwork was enriched with genes associated with T2D and cancer, and genes associated with metformin-associated cancer survival. To gain more insights into how metformin act in T2D and cancer treatment, we generated a subnetwork to synopsis the crosstalk between T2D and cancer based on the common genes with nominal significance (P-value < 0.05) among the four GWAS data sets (T2D, CGEMS breast cancer, pancreatic cancer, and prostate cancer). There were 25 genes common to all the four gene sets (Fig 6A), and there were only five edges in the metformin-specific SPNetwork (S8 Fig). By further examining degree distributions of the common 25 genes and their direct interactors (71 genes), we found that their interactors had significantly more interactions than the 25 genes as well as all the genes in the metformin-specific SPNetwork (Wilcoxon’s test P-value: 2.1 × 10–4 and 2.4 × 10–9, respectively) (Fig 6B). The 25 genes included one hub (PPARG) while the 71 genes included 21 of the 38 hub nodes in the metformin-specific SPNetwork. Similarly, the 25 genes contained three bridge nodes while the 71 genes contained 15 of the 41 bridge nodes between metformin upstream and downstream network. These observations indicate that the interactors of the 25 common nodes were more likely to play important roles for signal transduction.
Starting with the 25 genes and their 71 interactors, we assembled a subnetwork by their direct links among 96 nodes. The subnetwork comprised 96 nodes and 269 edges (S9 Fig). To further explore the metformin treatment mechanisms in T2D and cancer through the protein modules, we utilized software CFinder to perform network cluster and community analysis [63]. We required each node in one module participate at least one 3-vertex clique. Accordingly, we obtained three modules, which contained 6, 9, and 51 genes, respectively (S10 Fig). We found no gene shared between the first and second modules, but one gene (STK11) common to the first and third modules, or five genes (EIF4E, PPARGC1A, PRKCA, RPS6KB1, and SREBF1) common to the second and third modules. All the genes of the first and second modules belonged to metformin upstream network while most of the genes in the third module belonged to metformin downstream network. We merged them to form a network, which included 60 nodes and 210 edges (Fig 6C). Since this subnetwork was generated from common genes to T2D and cancer genotyping data, we defined it as a crosstalk subnetwork of metformin action in T2D and cancer.
We realized that, if we removed the nodes (CDKN1A, ESR1, MAX, MYC, PPARGC1A, STK11, and SP1), the connections among three modules would be lost (S11 Fig). Among them, three (MAX, MYC, and SP1) were both the bridge nodes and hub nodes. Therefore, these seven nodes might be functionally critical in the metformin signal transduction cascade. To further explore how the three modules and the seven key nodes might be related to metformin treatment in term of biological function meaning, we performed the KEGG pathway enrichment analysis on each module. Table 3 summarizes the enriched pathways for each module (adjusted P-value < 1.0 × 10–4). We labeled the enriched KEGG pathways (adjusted P-value < 1.0 × 10–9) for each module in Fig 6C.
In the first and second modules, there were two common pathways: adipocytokine signaling pathway and insulin signaling pathway. Adipocytokine signaling pathway was the top pathway in the first module (adjusted P-value: 2.01 × 10–13). The adipocytokine is a group of cytokines secreted by adipose tissue, which contributes to the development of insulin resistance, T2D, and cardiovascular disease [64,65]. The insulin signaling pathway, the top pathway in the second module, plays important roles in many complex diseases such as diabetes, obesity [66], and neurological disorders [67]. In addition, the mTOR signaling pathway and ErbB signaling pathway were also enriched in the second module. There were 28 pathways enriched in the third community. According to KEGG pathway annotation at the second level, 15 of these 28 pathways belonged to human disease, six to signal transduction, and three belonged to the endocrine system, one to cell communication, one to cell growth and death, one to development, and one to environmental adaptation. Among the 15 human disease related pathways, 11 were for specific types of cancer. Therefore, the three modules reflected different biological processes involved in T2D and cancer. Additionally, the pathway analyses highlighted the seven nodes that are not only topological linkers but also functional linkers in the crosstalk SPNetwork of metformin action in T2D and cancer.
Starting from above crosstalk subnetwork and the seven key nodes, we manually checked their publications and integrated the experimental evidence for further understanding their roles in the metformin actions. Through careful review, we summarized their function and action together and found that a novel MYC-centered pathway was hidden under the crosstalk subnetwork, which may play important roles in metformin action in T2D and cancer (Fig 7). The Myc-centered pathway included AMPK, STK11, MYC, SP1, and CDKN1A, which formed two small motifs: AMPK-STK11-MYC and MYC-SP1-CDKN1A.
It is well known that metformin exerts anti-diabetes and anti-cancer effects via mitochondrial complex I inhibition [68,69]. Mitochondrial complex I inhibition increases AMP/ATP ratio, which activates AMP-activated protein kinases (AMPKs) [70] to cause human disease [71]. In the crosstalk subnetwork, the first module contained core members of AMPK signaling pathways (PRKAA2, PRKAB2, and PRKAG2), which is linked to the second and third modules through the STK11-MYC interaction. The gene LKB1encodes a key upstream activator of AMPK [51] and is known to be inactivated through mutations during lung carcinogenesis [72]. Furthermore, the metformin induces activation of LKB1 [73]. For the MYC and LKB1, several lines of evidence show they are in opposite action in tumor. For example, LKB1 is overexpressed partly by degradation of MYC protein to inhibit lung carcinoma cell proliferation [74]. Nevertheless, their direct relationship is not clear. Recent studies have shown that metformin has an ability to reduce MYC protein level in vivo and in vitro in several types of cancer, including lung cancer [75] and prostate cancer [76]. Based on the integrative network and function analyses with experimental evidence, we suggested a feed-forward loop (AMPK-STK11-MYC) exists in metformin action. This network motif may act cohesively to strengthen the inhibition of MYC expression.
In addition, in the crosstalk subnetwork, three nodes (CDKN1A, MYC, and SP1) formed a 3-node clique. The network small motif bridges the three modules together. The SP1 is a TF that binds to the GC-rich motif of numerous genes’ promoters and is involved in many cellular processes, including cell differentiation, cell growth, apoptosis, immune responses, response to DNA damage, and chromatin remodeling. It has been reported that SP1 could cooperate with MYC to activate transcription of the human telomerase reverse transcriptase gene (TERT), which is responsible for maintenance of the length of telomeres and its defects may lead to diseases including cancer [77]. During the process of carcinogenesis, expression of MYC and SP1 is known to be up-regulated [78]. It has been reported that metformin has an ability to down-regulate MYC [75,76] and SP1 [61]. Additionally, MYC [79,80] and SP1 [81,82] are also the key transcription factors involved in the regulation of insulin and insulin regulated gene transcription. MYC could directly induce both impaired insulin secretion and loss of β-cell mass [83]. SP1 could regulate the upstream target STK11 expression [84,85]. MYC could activate AMPK in multiple cell lines [86]. AMPK activation could reduce SP1 translocate from cytoplasm to nucleus [87]. The CDKN1A, a cyclin-dependent kinase inhibitor p21, inhibits proliferation both in vitro and in vivo. After metformin treatment, the expression of CDKN1A is upregulated in hepatocellular carcinoma [88] and bladder cancer cells [89]. Additionally, multiple lines of evidence have demonstrated that MYC can suppress the expression of CDKN1A in cancer like colorectal cancer [90]. Therefore, taken all evidence together with the crosstalk network, we propose a new biological pathway for metformin action focused on four key nodes (CKDN1A, MYC, SP1, and STK11) (Fig 7). The pathway highlights several new questions, which may have been missed by previous studies. Specifically, we speculate that MYC and its networks are the key downstream targets of metformin. Further investigations are needed to illustrate this mechanism.
In this study, we developed a computational framework (DSPathNet) to construct a signaling pathway network for a given drug, specifically, metformin. The framework first collected metformin upstream genes from different data sources and inferred chemical signaling receptor TFs based on metformin-induced gene expression data. Then, a metformin-specific SPNetwork was produced using the random walk-based algorithms by applying longitudinal and lateral movements starting from metformin upstream genes and downstream TFs. By examining the enrichment of disease genes in the network, the metformin-specific SPNetwork proved to be enriched with genes that could contribute to the pathology of T2D and cancer, or reducing cancer risk in T2D patients undergoing metformin treatment. Starting from the genes common to T2D and cancer GWAS data, we further produced a crosstalk subnetwork of metformin action in T2D and cancer. Through comprehensive network and functional analyses and literature mining, we identified seven critical genes (CDKN1A, ESR1, MAX, MYC, PPARGC1A, STK11, and SP1), some of which have been implicated in previous studies. Furthermore, the MYC and its motifs were suggested to play important roles in metformin action. In summary, this study has the following major results: 1) we developed a computational framework for building drug-specific signaling pathway networks; 2) we generated a metformin-specific signaling pathway network that is significantly enriched with genes associated with T2D, cancer, or metformin-associated cancer survival, and 3) we pinpointed the MYC-centered pathway that may play important roles in metformin action. These results demonstrate that the computational framework effectively integrates various types of data, such as prior drug knowledge and drug-induced gene expression to identify critical genetic factors responsible for drug indications and drug response. This framework is a novel approach that provided a broader and deeper understanding of metformin actions in both T2D and cancer. This computational approach can be applied to other drugs as well.
This framework applies a new network generation strategy that focused on a drug of interest. In our framework, we utilized the gene expression data to infer the drug related gene expression regulators TFs, which is different from the methods that have been developed to infer signaling pathway networks directly from gene expression data [91]. As we know, the gene expression represents the transcriptional changes in the downstream genes of a pathway and provides an indirect view of pathway structure and gene activity after modulation of the system. Thus, the gene expression cannot directly represent the activity state of many signaling components that mediated the cellular response [92]. It is well known that the signal transduction network is not linear; rather it is quite complex [3]. During the development of this framework, we observed only two genes overlapped between metformin upstream genes and downstream genes. This small overlap presents us with a big challenge: how to fill the gap to rebuild a complete cascade for drug action? To tackle this challenge, we proposed a novel strategy from background human SPNetwork through both longitudinal and lateral movements. For the longitudinal movement, we employed the software NetWalker that implemented the random walk with a starting probability. For the lateral movement, we took advantage of K-Walk algorithm that simulates random walks in the network using a Markov Chain to build the most relevant subnetwork. In this study, we combined them together to achieve our goal. Table 2 summarizes the number of genes in each step and the hypergeometric tests based on the number of genes with smallest P-value less than 0.05 in the corresponding network compared to all genotyping data in T2D GWAS data. The evaluation results indicated that the process is promising since it recruited more informative genes; the significance of the association between the network and disease-related mutation signals became stronger.
However, the major concern regarding the framework is to rebuild a complete and reliable human SPNetwork and to control false positives from both public data and prediction results caused by the computational tools. To balance these two factors, we rigorously compiled the information involved in the signaling pathways, extensively collected the drug related data from multiple data sources, applied rigorous parameters during the use of computational approaches, and performed comprehensive evaluations for metformin-specific SPNetwork. To increase the accuracy of results, we only included the protein-protein pairs with experimental evidence and excluded the pairs only involved in the protein complexes. Thus, the coverage of the human SPNetwork was lower than a typical protein-protein interaction network; it contained only 37,881 edges and 4,367 proteins. With the rapid development of human experimental technologies, we believe more data with higher coverage and accuracy will become available, which will enable the construction of a more comprehensive signaling pathway network with high quality. To collect as many metformin-related genes as possible, in addition to the public databases DrugBank and PharmGBK, we further performed literature mining from PubMed abstracts, which provided an additional 19 genes. To ensure the accuracy of TF inference, we only utilized gene expression data from the four treatments of metformin that showed significant consistency with each other. To comprehensively evaluate if the metformin-specific SPNetwork was enriched with mutation signals of T2D and cancer, we not only took advantage of the well-studied disease genes but also individual genotyping data from GWAS data sets. Thus, our framework has the ability to recruit more key components in the drug signal transduction process. It could be potentially applied to other drugs for the purpose of deciphering their signaling pathway networks and identifying critical genes. Another limitation of this framework is the absence of a control network representing the normal state. The signaling network at the normal state may provide additional insights into drug action. However, it is very difficult and challenging to construct a normal-state signaling transduction network for drug action. Though some pathway data sources such as KEGG provide the relevant signaling networks in the normal state, most of them only provide a limited view by focusing on one or two related pathways. Compared to these individual pathway networks, the metformin-specific SPNetwork provides a comprehensive view by including many well-known metformin-related pathways, T2D-related pathways, and cancer-related pathways (Results).
This computational framework is strongly dependent on the available literature about the investigated drugs. Thus, it is not suitable for these drugs or chemicals that do not have many basic research reports. However, it is known that, during the drug development, most of them cannot be approved by FDA even after entering the clinical trials [93]. Furthermore, as the time and costs for developing novel drugs dramatically increased recently, many drug developers prefer to find new uses for existing drugs including the approved and non-approved drugs. As more large-scale data become publicly available, researchers could utilize the framework to build a SPNetwork for each drug of interest, and then examine the relationship between the network and disease genes, or calculate network similarities with the known drugs for a certain indication. These relationship or network similarities may provide more clues for drug repurposing at the network level. Therefore, the framework will be promising for identification of drugs that may be used to treat secondary indications by constructing and comparing the drug-specific SPNetworks. Moreover, since the drug-specific SPNetwork contains comprehensive information regarding the drug action of the components, we speculated that some off-targets might be included in the network. Thus, our network approach can be extended to evaluate the association between drugs and their potential side effects. However, it is challenging to identify large-scale side effect data associated with genes or their proteins. So far, several studies have used the available biochemical data to determine candidate targets for specific side effects [94–96]. Such data is limited and likely with a high false positive rate. When more relevant data becomes available in future, our approach will be applied to assess drugs’ side effects.
An important output of this study is the metformin-specific SPNetwork consisted of metformin related genes, metformin related TFs, and many novel genes. The network provides a valuable gene pool for further investigation of metformin action. Metformin has been used to treat diabetic disorders for many years because of its ability to lower glucose levels and improve insulin sensitivity [97]. Recently, several findings from epidemiological studies have shown that metformin can reduce cancer risk and improve cancer survival in the T2D patients [60,98,99], including a recent electronic health record (EHR) study we participated in that demonstrated the effect was seen for many cancer types [26]. However, the molecular mechanisms underlying metformin action are complex and remain unclear, especially for its ability of decreased cancer risk [100,101]. In this study, we first constructed a complex metformin-specific SPNetwork and then produced a crosstalk subnetwork from the metformin-specific SPNetwork. This subnetwork contained three modules highlighting different pathways (Fig 6C). The first and second modules were enriched with genes from the insulin signaling pathway and adipocytokine signaling pathway, and the third module was enriched with genes involved in cancer related pathways. The adipocytokine signaling pathway contains the major components of AMPK signaling pathway according to KEGG annotation. Through seven nodes, the first and second modules were linked to the third module. These observations suggest that the metformin possibly affects the AMPK signaling pathway and the insulin signaling pathway directly, which subsequently decrease the chance of cancer development. This outlook is consistent with a previous review [102].
The seven nodes act as bridges linking the first and second modules to the third module. We predicted they might play critical roles in the metformin signaling transduction process (Fig 6C). Among them, two genes (PPARGC1A and STK11) belonged to metformin upstream genes; one (ESR1) to metformin downstream genes; four genes (CDKN1A, MAX, MYC, and SP1) were both hubs and bridge nodes. It is well known that gene STK11, also known as LKB1, encodes a member of the serine/threonine kinase family that regulates cell polarity and functions as a tumor suppressor [103]. Additionally, previous studies have shown that mutations in the STK11 gene influence insulin sensitivity and metformin efficacy [104,105]. The MYC gene encodes a protein that plays a role in cell cycle progression, apoptosis, and cellular transformation [106]. It has been shown that MYC gene plays important roles in the anticancer metabolic effects of metformin [75,76]. The PPARGC1A gene encodes a transcriptional coactivator that regulates the genes involved in energy metabolism. Its variant rs2970852 has been reported to modify the effects of metformin on triacylglycerol levels [107]. Recent studies have shown that gene regulation induced by metformin involves the transcription factor SP1 in cancers [61,108]. Moreover, the expression of CDKN1A (also known as P21) is upregulated in hepatocellular carcinoma [88] and bladder cancer cells [89] after metformin treatment. The evidence from these studies suggests that our approach is effective for identifying the key components in the signaling pathway. To further investigate detailed information for these genes, more experimental validations are needed. To our knowledge, there is no any positive evidence for the association of the genes ESR1 and MAX of the seven critical genes with metformin action. Thus, they are two novel genes for further experimental validation.
In addition to the DSPathNet framework to effectively recruit critical components in the mode of drug action, there are other ways to expand this approach. First, integrating multiple layers of data involving the signal cascade beyond gene expression data into a comprehensive method might improve our ability to identify the association between the genetic changes and their response to drugs. Second, although we have shown the utility of two sources for compiling the human SPNetwork, there are other data worth exploring such as those involved in the metabolism, protein phosphorylation, and protein kinase and phosphatase interactions. While this study focused on one medication, metformin, the computational framework is broadly applicable to any drug for which induced gene expression data is available. Moreover, several experimental data sources are available for further data integration and mining such as the Connectivity Map project [15], Genomics of Drug Sensitivity in Cancer [109], Cancer Cell Line Encyclopedia (CCLE) [110], and anticancer compounds in breast cancer [111]. Finally, analyzing the crosstalk among different types of diseases in the context of networks will offer an intriguing opportunity to explore the underlying molecular mechanisms of drug action, which will provide an alternative approach for drug repurposing.
Before generating the metformin-specific SPNetwork, we need a global signal transduction network for humans as the background network. We therefore integrated signaling transduction related associations from Pathway Commons with experimental evidence [25], and TF-TF/target pairs from TRANSFAC [26]. The Pathway Commons database collected publicly available pathways from multiple organisms with over 1,400 pathways and 687,000 interactions. We first downloaded the edge data specific for humans from the Pathway Commons (release 2011.10). Since the interactions that occur within the protein complexes do not reveal the flow of signaling information [3], we excluded the edges that came from the same complex. This process resulted in 33,614 pairs among 3,502 proteins. Additionally, we obtained 1,325 pairs among 487 TFs, and 2,723 pairs between 428 TFs and 1,315 targets downloaded from TRANSFAC database (release 2011.4). The TRANSFAC database manually collects eukaryotic TFs, their genomic binding sites, and DNA binding profiles with experimental evidence [112]. After merging the two data sets and removing the redundancies, we obtained a network with 37,881 edges and 4,367 nodes. This network was used to represent global signaling pathways in humans.
To further weight the association of each node in human SPNetwork with metformin action, we assigned a functional similarity score by calculating its functional similarity to the metformin upstream genes using the R package GoSemSim based on GO annotations [113]. GO annotations have three functional domains (k): molecular function (MF), biological process (BP), and cellular component (CC). First, for a given node i in each domain (k), we calculated its score as Scorei = ∑j = 1nScorei,j/n, where n is the number of existing scores between node i and metformin upstream gene j. Second, for the given node i in all domains, we calculated a final score as S^ = ∑k = 1NScorek/N, where N is the number of the domains having scores for the node.
Gene expression profiles of cancer cells following drug treatment are useful for better understanding cellular changes reflective of drug treatment [114]. In this study, we integrated the known TF-target association and drug-induced gene expression data to infer the metformin downstreams. We first comprehensively collected the TF-target associations, then calculated the up- or down-regulated genes from drug-induced gene expression data, and finally performed the hypergeometric test to evaluate the over-representation of the up- or down-regulated genes in multiple TF target gene datasets.
To compile a target gene set for each TF comprehensively, we downloaded data from two sources: TRANSFAC Professional (release 2011.4) and MSigDB database [38]. From the TRANSFAC database, we extracted known TFs and their targets in human. From the MSigDB, we downloaded the gene sets that share one TF binding site. The gene sets were derived from a comparative analysis of human, mouse, rat, and dog genomes and were organized by TF binding motifs. Genes associated with different binding motifs that correspond to a common transcription factor were combined into one gene set. After merging the two data sets, we obtained 666 human TFs and 8,502 human targets.
To calculate the potential differentially expressed genes induced by metformin, we downloaded ten gene expression datasets from Connectivity Map website (version 2.0). The gene expression datasets were generated from metformin treated cell lines. We calculated the ranked probes by using the method described in Lamb et al. [15] and selected the top 100 and bottom 100 probes in each treatment to represent the differentially expressed probes [115]. We examined the expression consistency among them using the software GSEA. We noticed that, among ten metformin treatment data sets, four had the highest consistency based on GSEA analysis [38]. Therefore, we utilized these four treatment gene expression data to perform the GSEA leading edge analysis to detect differentially expressed probes. Then, by mapping the differently expressed probes to genes using Ingenuity Pathway Analysis Tool (http://www.ingenuity.com/), we obtained the up-regulated genes and down-regulated genes.
Finally, we performed the hypergeometric test to evaluate the over-representation of the up- or down-regulated genes in the different TF gene sets. The TFs with P-value less than 0.05 were identified as significant TFs related to metformin action and their genes as metformin downstream genes.
Considering that the signal transduction cascade is not linear, we adopted a two-step strategy to construct the metformin-specific SPNetwork from the human SPNetwork. More specifically, in the first step, we utilized the software NetWalker to expand metformin upstream genes and downstream genes for longitudinal conduction [116]. The NetWalker implements the random walk with a starting probability. In this study, we gave equal starting probability of 0.5 to each gene in the metformin upstream genes and downstream genes and required those nodes with both local P-value < 0.05 and global P-value < 0.05 as the expanded genes. In the second step, we expanded the nodes from in the first step by lateral movement by applying the K-Walk method implemented in the Python package GenRev [117]. The K-Walk algorithm simulates random walks in the network using a Markov Chain to build the most relevant subnetwork, connecting seed nodes by walk a fixed length L or up to a maximal length Lmax in a large network. A subnetwork is obtained by keeping only edges that are above a minimal relevance threshold. The threshold is automatically fixed after the subnetwork has the maximum score. As such, the limited K-Walk algorithm computes edge and node relevance from random walks connecting the seed nodes [118].
We used one T2D GWAS data set, three cancer GWAS data sets, and one GWAS data set for T2D patients with metformin treatment. The T2D GWAS data was individual-level genotype data generated from the WTCCC [31]. The three cancer GWAS datasets were generated by the Cancer Genetic Markers of Susceptibility (CGEMS) project: breast cancer [32], pancreatic cancer [33], and prostate cancer [32]. We downloaded the genotype data from the National Center for Biotechnology Information (NCBI) dbGaP with approved access for the CGEMS project. For these four GWAS datasets, we first removed individuals with genotyping rate < 95% and SNPs with missing rate >5%. A single SNP associated test was conducted using the Armitage trend test for SNPs with a minor allele frequency (MAF) > 0.05. S10 Table summarizes the data.
T2D cancer patients from Vanderbilt University Medical Center (VUMC) were identified using the Synthetic Derivative (SD), a de-identified copy of the electronic health records from VUMC. Eligible subjects were individuals who 1) had a cancer diagnosis (excluding non-melanoma skin cancers) between January 1, 1995 and December 31, 2010 identified through the Vanderbilt tumor registry, and 2) were older than 18 years at the time of cancer diagnosis. Using a previously developed algorithm [119,120], we identified T2D subjects having at least two pieces of clinical information in their medical record: 1) ICD9 code for type 2 diabetes, 2) medications for type 2 diabetes, or 3) clinical labs suggestive of T2D (random glucose >200 mg/dl or hemoglobin A1c > 6.5%). Individuals without at least two of the above types of information were excluded. At least two mentions of metformin use (mono-therapy or combined therapeutic) and one mention of metformin use within 5 years after cancer diagnosis were required for study inclusion. Individuals on other T2D medications were excluded from analysis. Subjects were followed for overall mortality that was determined through linkage with the Vanderbilt tumor registry. Physician-reported European descent individuals with an available DNA sample in the Vanderbilt biobank (BioVU) [121] were genotyped on either the Illumina HumanOmni1-Quad or the Illumina HumanOmni5-Quad. Only the consensus single nucleotide polymorphisms (SNPs) between the two genotyping platforms were used. Standard quality control (QC) procedures were applied to remove individuals and autosomal SNPs not meeting standard QC criteria (i.e. related individuals, discordant sex, sample efficiency < 98%, genotyping efficiency < 98%, deviations from Hardy-Weinberg equilibrium (p < 1×10–6), and MAF < 5%). Palindromic SNPs were also removed. After QC, 461 individuals and 551,745 SNPs remained. Principal components were estimated using EIGENSTRAT [122]. The association between each SNP, assuming an additive genetic model, and overall survival was examined using Cox proportional hazards models, adjusted for age, sex and one principal component, using the GenABLE package of R [123]. The GWAS analysis of this set is ongoing and will be reported in a separate publication.
In this study, we defined the genes having at least one SNP with nominal P-value less than 0.05 as disease or drug related genes. The SNP is located in the gene’s region or its 20kb up- or down-stream sequence based on the gene annotation and human reference genome build 36 for T2D GWAS study and cancer GWAS studies and build 37 for metformin GWAS study.
To identify pathways overrepresented in gene sets, we performed KEGG pathway enrichment analyses using WebGestalt [49] (version 1/30/2013). Given a list of genes, a hypergeometric test was performed for the enrichment of these genes, which was implemented in the WebGestalt tool. To control the error rate in the analysis results, WebGestalt also provides a corrected P-value based on the Benjamini-Hochberg method [124]. To summarize the enriched pathways, we took advantage of KEGG pathway category annotation, which included the two-level categories and represent the relative abundance of the pathways [125]. These pathways are grouped into seven categories at the first level of KEGG annotation and 43 categories at the second level of KEGG annotation. At the second-level category, we further calculated a Z-score for each category to represent the KEGG pathway relative abundance: Z-score = x-uσ, where x is the number of pathways in one category in the first or second level, u is the mean of the pathway number in the first or second category, σ is the standard deviation of the pathway number in the first or second category. The pathway categories were selected for further analysis if their Z-scores were higher than zero.
In this study, we adopted the statistical design for gene set enrichment analysis [126] to compare a gene set (A) in the drug-specific network to a reference gene set (B). The design has been commonly used to conduct the gene annotation enrichment analysis [127]. Suppose that the gene set (A) has n genes, of which most genes (n’) belong to the reference gene set (m). Among n’ gene, k genes belong to a given category (C). And the reference gene set has j genes belong to the same category (C). Based on the definition of the hypergeometric test, we performed the hypergeometric test to get a P-value to evaluate the significance of enrichment for category C in the gene set A.
For network property analysis, we calculated degree of each node and degree distribution of all nodes, which are the most basic measures of biological networks [41]. The node degree (connectivity) is the number of links of a node in the network. If degree distribution of one network follows a power law, the network would have only a small portion of nodes with a large number of links (i.e., hubs) [41]. To determine the hubs in metformin-specific SPNetwork, we adopted the method utilized by Yu et al. [46], as we did in a previous study. We first drew a degree distribution for the whole network to define a specific degree value as a cut-off point (S12 Fig). If a node has the degree greater than the cut-off value, then the node is a hub. To identify the modules, we performed the cluster and community analysis using the software CFinder (version 2.0.5) [63]. CFinder is a fast program to locate and visualize overlapping, densely interconnected groups of nodes in undirected network. We required each node in the module being involved in at least one 3-vertex clique. We visualized the networks using Cytoscape (version 3.2) [128].
|
10.1371/journal.pntd.0007053 | A longitudinal systems immunologic investigation of acute Zika virus infection in an individual infected while traveling to Caracas, Venezuela | Zika virus (ZIKV) is an emerging mosquito-borne flavivirus linked to devastating neurologic diseases. Immune responses to flaviviruses may be pathogenic or protective. Our understanding of human immune responses to ZIKV in vivo remains limited. Therefore, we performed a longitudinal molecular and phenotypic characterization of innate and adaptive immune responses during an acute ZIKV infection. We found that innate immune transcriptional and genomic responses were both cell type- and time-dependent. While interferon stimulated gene induction was common to all innate immune cells, the upregulation of important inflammatory cytokine genes was primarily limited to monocyte subsets. Additionally, genomic analysis revealed substantial chromatin remodeling at sites containing cell-type specific transcription factor binding motifs that may explain the observed changes in gene expression. In this dengue virus-experienced individual, adaptive immune responses were rapidly mobilized with T cell transcriptional activity and ZIKV neutralizing antibody responses peaking 6 days after the onset of symptoms. Collectively this study characterizes the development and resolution of an in vivo human immune response to acute ZIKV infection in an individual with pre-existing flavivirus immunity.
| Zika virus (ZIKV) is an emerging flaviviral infection that causes significant clinical disease. It is estimated that approximately one half of the world’s population is at risk for ZIKV infection. There are only a limited number of studies describing the human immune response to ZIKV infection. Carlin et al. combined conventional and genomic approaches to longitudinally analyze the innate and adaptive immune responses to acute ZIKV infection and its resolution in a person who was infected while traveling in Venezuela during the 2016 ZIKV epidemic year. Genome-wide sequencing in individual cell types revealed that although many populations respond to interferon stimulation, only specific cell populations within peripheral blood mononuclear cells upregulate important inflammatory cytokine gene expression. Additionally, analysis of open chromatin using ATAC-seq suggests that chromatin remodeling at sites containing cell-type specific transcription factor binding motifs may help us understand changes in gene expression. Consistent with previous reports, this individual with prior exposure to dengue virus (DENV), rapidly developed neutralizing anti-ZIKV responses that were cross-reactive with multiple DENV serotypes. Collectively this study combines traditional and genomic approaches to characterize the cell-type specific development of an in vivo human immune response to acute ZIKV infection.
| Zika virus (ZIKV) is an emerging arthropod-borne flavivirus. It is primarily transmitted by Aedes sp. mosquitos but can also be transmitted person to person vertically from mother to child, sexually and in blood during transfusions [1]. Clinical manifestations occur in approximately 20% of infections and can include an acute onset low grade fever, pruritic erythematous macular papular rash, arthralgias and conjunctivitis [2]. Clinically these symptoms can be confused with dengue virus (DENV) or chikungunya virus (CHIKV) infections that are transmitted by the same mosquito vectors and can co-circulate with ZIKV [3]. During pregnancy, ZIKV can cause congenital Zika syndrome and other severe birth defects in fetuses [2]. In adults, ZIKV is associated with life-threatening Guillain-Barré Syndrome (GBS) [4, 5]. The details of how ZIKV bypasses immune restriction to cause disease are still under investigation.
The relationship between flaviviruses and the immune system is complex [6]. On one hand, the immune system can exacerbate viral pathogenesis. For example, ZIKV, like DENV and West Nile virus (WNV), infect innate immune white blood cells early in infection [7–11]. Studies in ZIKV infected children identified monocytes, in particular CD14+CD16+ intermediate monocytes, and myeloid dendritic cells as the main targets of ZIKV infection in peripheral blood mononuclear cells (PBMCs) [9]. These infected cells may act like a “Trojan horse” to increase spread of the virus to different tissue compartments. Antibody (Ab) responses to flaviviruses are often cross-reactive and have the potential to mediate antibody-dependent enhancement (ADE). While there is no evidence that ADE alters ZIKV pathogenesis in humans, in a mouse model of ZIKV infection, administration of DENV or WNV convalescent plasma increased ZIKV morbidity and mortality through ADE [12]. On the other hand, the development of protective adaptive immune responses is thought to be critical to clear ZIKV infection [6]. Therefore, increasing our understanding of human immune responses to ZIKV infection can lead to better understanding of ZIKV clinical manifestations and pathogenesis and inform the development of vaccines.
Only a small number of studies have examined human responses to ZIKV infection in vivo. Analysis of serum inflammatory markers during acute ZIKV infection identified some potential biomarkers associated with neurologic complications [13] and viremia plus moderate symptoms [14]. Monoclonal Abs isolated from four donors infected with ZIKV demonstrated that neutralizing Abs primarily recognized the envelope protein domain III of ZIKV and that Abs recognizing different ZIKV epitopes could alternatively protect against ZIKV challenge or enhance subsequent DENV infection in mice [15]. Another study tracking the development of Ab responses to ZIKV in three DENV-experienced and one DENV-naïve individual found that acute-phase Abs developing during ZIKV infection in DENV-experienced individuals were highly cross-reactive but poorly neutralizing [16]. In a single flavivirus naïve individual, anti-ZIKV B-cell plasma neutralization activity and T-cell responses peaked later between day 15 and day 21 [17]. A large study examining T cell responses to ZIKV in DENV-naïve and DENV-immune patients revealed that DENV exposure prior to ZIKV infection influences the timing, magnitude, and quality of the T cell response [18]. In another study that examined both innate and adaptive immune responses in 5 individuals infected with ZIKV, Lai et. al. observed that flavivirus-experienced individuals developed rapid cross-reactive antibody responses against both DENV and ZIKV as well as activated CD8+ T cell responses, albeit few ZIKV-specific CD8+ T cells were identified [19].
These studies provide insight into human ZIKV infection, but our understanding remains limited due to the small number of reported cases. Additionally, published reports have utilized conventional approaches to study the in vivo immune responses to ZIKV. Combining these approaches with genome-wide next-generation sequencing (NGS) analyses could bring new insight into human ZIKV responses and inform direction and design of future studies of immune responses during infection in larger cohorts. As a step towards improving our understanding of human immune responses to acute ZIKV infection through new approaches, we present a detailed immunologic characterization of the innate and adaptive temporal and cell type-specific responses to an acute ZIKV infection in a DENV-experienced patient.
This research study was approved by the UCSD IRB with Human Research Protections Program # 161060. Written informed consent was obtained from the adult human subject described in this report.
After obtaining written informed consent, blood was collected on five occasions d3, d6, d17, d48, and d240 post-onset of symptoms (POS). Urine was collected on d3 and d6 only. Serum was isolated by collecting blood into a plain tube containing no anticoagulant, allowed to clot at room temperature for 20 minutes followed by centrifugation at 1500xg for 10 minutes in a refrigerated centrifuge. Serum was frozen in single use aliquots at -80°C. Peripheral blood mononuclear cells (PBMCs) were isolated from heparinized blood using Histopaque-1077 per manufacturer's instructions and subjected to flow activated cell sorting (FACS) or cryopreserved in 5 million cell aliquots in 90% FBS + 10% DMSO (Hybri-max Sigma) using a Nalgene Mr. Frosty at -80°C for 24 hours before transfer to liquid nitrogen. Cryopreserved cells were thawed rapidly to 37°C and slowly diluted with pre-warmed growth media, followed by gentle pelleting and resuspension in cold FACS staining buffer.
Five microliters of d3 POS serum or blood was inoculated into a T25 flask of C6/36 mosquito (Aedes albopictus) cells. Supernatants (5 mL) were harvested seven days after culture and titrated via BHK-21 cell-based focus forming assay (FFA) and anti-Flavivirus envelope (E) protein antibody clone 4G2. The urine culture supernatant had a titer of 2.0 x 104 focus forming units (FFU)/mL. Infectious virus in the serum culture supernatant was undetectable.
Viral RNA from 0.2ml of C6/36 supernatant that was inoculated with d3 POS urine was extracted using the Roche High Pure Viral RNA Kit (Roche) and reversed transcribed using a primer specific method for ZikaBr (Forward primer AGTGGAGACGATTGYTGTNGT, Reverse primer AACATGTCTTCTGTGGTCATCCA) (SuperScript III First-Strand Synthesis System for RT-PCR, Invitrogen). cDNA was amplified using Taq polymerase (Roche), cleaned using QIAquick PCR Purification Kit (Qiagen) and sequenced using BDT v3.1 on the ABI 3130xl Genetic Analyzer. Forward and reverse sequences were used to make a contig and manually edited using Bioedit [ref http://www.mbio.ncsu.edu/BioEdit/bioedit.html]. The Basic Alignment Search Tool (BLAST) [ref:] was then used with the resultant sequence [ref: https://blast.ncbi.nlm.nih.gov/Blast.cgi?PROGRAM=blastn&PAGE_TYPE=BlastSearch&LINK_LOC=blasthome] which most closely aligned with other ZIKV NS5 sequences. For phylogenetic analyses, RNA from ZIKV SD001 infected primary human macrophages were aligned to the human hg19 genome using STAR [PMID: 23104886]. Any unmapped reads were used as input for strand-specific de novo transcriptome assembly with Trinity [PMID: 21572440]. The longest assembled transcripts were approximately 9 kb, and corresponded to near full-length viral genomes. The resulting alignment from ZIKV SD001 and 435 publicly available ZIKV sequences from NCBI viral genomes resource [20] were used to perform an approximate maximum likelihood phylogenetic tree with PhyML [21]. The tree was rooted with ZIKV (GenBank accession number KY241712) isolated in Asia.
For innate immune cell sorting ten million PBMCs were stained with antibodies against CD3 PE-Cy7, CD19 PE-Cy7 CD20 PE-Cy7, HLADR BV421, CD11c AF700, CD123 PE, CD14 AF488, CD16 APC, CD56 APC-Cy7, and Zombie Aqua Fixable viability dye and separated as shown. For T cell sorting, five million cryopreserved PBMCs were stained with CD16 BV510, CD56 BV510, CD4 APC-eFluor780, CD3 AF700, CD8 BV785, CD45RA BV570, CCR7 PE-Cy7, CXCR5 BV421, CXCR3 BV605, TCR V_24-J_18 BV711, CD226 BB515, CCR6 PerCP-Cy5.5, CCR4 PE, CD25 PE-Dazzle 594, and CD127 AF647 and sorted into CD3+ T cell CD4+ and CD8+ populations. T cells were further analyzed for effector or memory phenotypes, CD4 T helper (Th) subsets based on the expression of chemokine receptors (Th1: CCR6-CCR4-CXCR3+; Th2: CCR6-CCR4+CXCR3-; Th1/17: CCR6+CCR4-CXCR3+; and Th17: CCR6+CCR4+CXCR3-) as well as the cytotoxicity marker CD226. Stained PBMCs were sorted in the La Jolla Institute (LJI) Flow Cytometry Core Facility on a FACSAria Fusion sorter.
Sequencing libraries were prepared using a low input RNA-seq prepared according to the Smart-seq2 method [22] with some modifications. 5000–15,000 PBMCs (pre-sort) or FACS isolated cell populations were lysed in TRIzol and RNA extracted using Direct-zol RNA Microprep (Zymo) with on-column DNAseI treatment. 10 μL purified RNA was mixed with 5.5 μL of SMARTScribe 5X First-Strand Buffer (Clontech), 1 μL polyT-RT primer (2.5 μM, 5’-AAGCAGTGGTATCAACGCAGAGTAC(T30)VN, 0.5 μL SUPERase-IN (Ambion), 4 μL dNTP mix (10 mM, Invitrogen), 0.5 μL DTT (20 mM, Clontech) and 2 μL Betaine solution (5 M, Sigma), incubated 50°C 3 min. 3.9 μL of first strand mix, containing 0.2 μL 1% Tween-20, 0.32 μL MgCl2 (500 mM), 0.88 μL Betaine solution (5 M, Sigma), 0.5 μL (5 M, Sigma) SUPERase-IN (Ambion) and 2 μL SMARTScribe Reverse Transcriptase (100 U/μL Clontech) was added and incubated one cycle 25°C 3 min., 42°C 60 min. 1.62 μL template switch (TS) reaction mix containing 0.8 μL biotin-TS oligo (10 μM, Biotin-5’-AAGCAGTGGTATCAACGCAGAGTACATrGrG+G-3’), 0.5 μL SMARTScribe Reverse Transcriptase (100 U/μL Clontech) and 0.32 μL SMARTScribe 5X First-Strand Buffer (Clontech) was added, then incubated at 50°C 2 min., 42°C 80 min., 70°C 10 min. 14.8 μL second strand synthesis, pre-amplification mix containing 1 μL pre-amp oligo (10 μM, 5’AAGCAGTGGTATCAACGCAGAGT-3’), 8.8 μL KAPA HiFi Fidelity Buffer (5X, KAPA Biosystems), 3.5 μL dNTP mix (10 mM, Invitrogen) and 1.5 μL KAPA HiFi HotStart DNA Polymerase (1U/μL, KAPA Biosystems), was added, then amplified by PCR: 95°C 3 min., 5 cycles 98°C 20 sec, 67°C 15 sec and 72°C 6 min, final extension 72°C 5 min. The synthesized dsDNA was purified using Sera-Mag Speedbeads (Thermo Fisher Scientific) with final 8.4% PEG8000, 1.1M NaCl, then eluted with 13 μL UltraPure water (Invitrogen). The product was quantified by Qubit dsDNA High Sensitivity Assay Kit (Invitrogen) and libraries were prepared using the Nextera DNA Sample Preparation kit (Illumina). Tagmentation mix containing 11 μL 2X Tagment DNA Buffer and 1 μL Tagment DNA Enzyme was added to 10 μL purified DNA, then incubated at 55°C 15 min. 6 μL Nextera Resuspension Buffer (Illumina) was added and incubated at room temperature for 5 min. Tagmented DNA was purified using Sera-Mag Speedbeads (Thermo Fisher Scientific) with final 7.8% PEG8000, 0.98M NaCl, then eluted with 25 µL UltraPure water (Invitrogen). Final enrichment amplification was performed with Nextera primers, adding 1 μL Index 1 primers (100 μM, N7xx), 1 μL Index 2 primers (100 μM, N5xx) and 27 μL NEBNext High-Fidelity 2X PCR Master Mix (New England BioLabs), then amplified by PCR: 72°C 5 min., 98°C 30 sec., 6–12 cycles 98°C 10 seconds, 63°C 30 sec., and 72°C 1 min. Libraries were size selected, quantified using the Qubit dsDNA HS Assay Kit (Thermo Fisher Scientific), pooled and sequenced on a Hi-Seq 2000 sequencer using single-end 50bp reads at a depth of 25 to 30 million single end reads per sample.
50,000 FACS isolated classical monocytes or NK cells were lysed in 50 μl lysis buffer (10 mM Tris-HCl ph 7.5, 10 mM NaCl, 3 mM MgCl2, 0.1% IGEPAL, CA-630, in water) on ice and nuclei were pelleted by centrifugation at 500 RCF for 10 min. Nuclei were then resuspended in 50 μl transposase reaction mix (1x Tagment DNA buffer (Illumina 15027866), 2.5 μl Tagment DNA enzyme I (Illumina 15027865), in water) and incubated at 37°C for 30 min on a PCR cycler. DNA was then purified with Zymo ChIP DNA concentrator columns (Zymo Research D5205) and eluted with 10 μl of elution buffer. DNA was then amplified with PCR mix (1.25 μM Nextera primer 1, 1.25 μM Nextera index primer 2-bar code, 0.6x SYBR Green I (Life Technologies, S7563), 1x NEBNext High-Fidelity 2x PCR MasterMix, (NEBM0541)) for 8–12 cycles, size selected for fragments (160–290 bp) by gel extraction (10% TBE gels, Life Technologies EC62752BOX) and single-end sequenced for 51 cycles on a HiSeq 4000 or NextSeq 500.
RNA-seq reads were aligned to the GRCh38/hg38 assembly of the human genome using STAR (version 2.5.2a) using default parameters [23]. Gene expression values were calculated as fragments per kilobase per million mapped reads (FPKM) across GENCODE transcript exons (release 24) [24] using HOMER [25]. To remove possible contamination from genomic DNA in the RNA-seq samples, FPKM measurements were calculated for long introns (>10 kb) and the median intron FPKM per experiment was subtracted from each exon FPKM values to remove background signal. Gene expression FPKM values across all samples set to a minimum of zero and then quantile normalized. Only GENCODE transcripts with length greater than 300 bp were considered. Log2 fold change ratios were calculated using a pseudo count by adding a FPKM of 4 to both numerator (i.e. day 3, 6, 17) and denominator (i.e. day 48/convalescent) to reduce the impact of low expression noise and contamination on the lists of regulated genes. Functional enrichment was performed using HOMER using pathway definitions from Gene Ontology and HALLMARK pathways from MSigDB [26]. Promoter known motif enrichment was calculated using HOMER using sequence from -300 bp to +50 bp relative to annotated transcription start sites. Hierarchical clustering of correlated gene expression profiles, motif enrichment, and GO/pathway function enrichment values were performed using Cluster 3.0 [27] and visualized using Java TreeView [28]. For ATAC-seq, fastq files were trimmed and aligned to hg38 using bowtie2. Reads mapping to Mitochondrial DNA were removed and PCR duplicates were removed. Peaks were called using a standardized peak size using HOMER (300 bp). To compare classical monocytes and NK cells the appropriate peak files were merged and differential peaks identified using getDifferentialPeaks command (HOMER) with threshold of fold change >3 and P-value < 0.001. Motif analysis was performed on differential peak files using findMotifsGenome.pl (HOMER). All human RNA-seq and ATAC-seq data described in this manuscript are available at the National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO) accession number GSE123541.
Affymetrix gene expression microarray CEL files were downloaded from NCBI GEO for longitudinal DENV infection in humans (GSE43777) and ZIKV infection in Rhesus macaques (GSE93861) and processed into gene expression values using R/Bioconductor using GCRMA with default options. For the human DENV infection data, only samples performed on whole genome HG-U133plus2 microarrays were used for the comparison. Samples for the human DENV study were identified based on their annotated number of days since initial fever (G1, G2, etc.) and averaged to generate per day expression values. Rhesus macaque ZIKV infection gene expression values were averaged based on the day post infection, and human orthologues were assigned using one-to-one orthologues defined by ENSEMBL BIOMART (https://www.ensembl.org/biomart). For each study, log2 activation ratios were calculated using the average expression for each day compared to the average of the convalescent samples (human) or pre-infection samples (Rhesus). Microarray and RNA-seq activation ratios were compared by linking the datasets using gene symbols, using data from the highest expressed isoform in the cases where multiple isoforms exist per gene.
Flow cytometry-based neutralization assay was used to evaluate SD001 serum neutralization of ZIKV (strains FSS13025 and SD001 [29]) and DENV (DENV1 strain West pacific 74 and DENV4 strain TVP-360) in vitro. 2×104 FFU DENV or ZIKV were incubated with or without serial 3-fold dilutions (starting at 1:10) of heat-inactivated SD001 serum in 96-well round bottom plates for 1-hour at 37°C. U937 cells stably expressing DC-SIGN (1x105) were seeded in each well and incubated for 2 h at 37°C with occasional rocking. After incubation, the plates were centrifuged for 5 minutes at 1500 rpm, supernatants aspirated and fresh medium added followed by incubation for 16 h at 37°C. U937 cells were then fixed, permeabilized, stained with anti-CD209 PE and 4G2 FITC (to detect ZIKV) or 2H2 FITC (to detect DENV) and analyzed using an LSRII. Percent inhibition was calculated by determining the relative infection in virus incubated with serial diluted patient serum (tests) versus no serum (control). Best fit curves and neutralizing titer 50 (NT50) were determined using Prism 7.0 (GraphPad).
Serum from 8 months prior to infection (pre-infection) as well as d3, d6, d17, and d48 POS were prepared in duplicate using the Bio-Plex Pro Human Cytokine 27-plex Assay (Bio-rad #M500KCAF0Y) per manufacturers protocol and read using a Luminex machine. Cytokine concentrations were calculated from standard curves generated using references included in the kit. The following cytokines were measured FGF basic, Eotaxin, G-CSF, GM-CSF, IFN-γ, IL-1β, IL-1ra, IL-2, IL-4, IL-5, IL-6, IL-7, IL-8, IL-9, IL-10, IL-12 (p70), IL-13, IL-15, IL-17, IP-10, MCP-1 (MCAF), MIP-1α, MIP-1β, PDGF-BB, RANTES, TNF-α, and VEGF.
A middle-aged, previously healthy, dengue virus (DENV)-experienced woman developed fatigue, an erythematous pruritic macular rash, and arthralgias six days after traveling to Caracas, Venezuela in March 2016 (Fig 1A and 1B). She presented on day 3 (d3) post-onset of symptoms (POS). A comprehensive metabolic panel and complete blood count were within normal limits except for slight elevations in ALT (50 U/L, normal range 0–41 U/L) and AST (44 U/L, normal range 0–40 U/L). Serologic testing was consistent with acute flaviviral infection but did not differentiate between DENV and ZIKV infection (Table 1) [30]. A research-use nucleic acid amplification test (NAAT) (Hologic) was positive for ZIKV infection in d3 POS blood and urine samples (Table 1). Blood and urine on d3 and d6 POS were negative for DENV, as determined via qRT-PCR. Urine, from d3 and d6 POS, inoculated onto C6/36 cells produced infectious virus as measured by focus forming assay (FFA). Sequence analysis of C6/36 amplified virus was confirmed to be ZIKV using a validated population-based sequencing protocol for ZikaBr targeting ZIKV NS5. Phylogenetic analysis of the near complete viral genome (>9kb) showed the ZIKV San Diego isolate (ZIKV SD001 [29]) was most closely related to other Latin American ZIKV isolates downloaded from Genbank (Fig 1C) [31].
To characterize the systemic immune response to ZIKV infection, we first measured circulating serum cytokine levels. Serum was collected on d3, d6, d17 and d48 POS. These samples were compared to baseline pre-infection serum collected from this individual 8 months prior to infection. We found that only a small number of cytokines, including IP-10, MCP-1 and IL-1RA, showed dramatic increases during early infection (Fig 2A). Each of these cytokines peaked on d3 POS before returning toward baseline. The levels of many inflammatory cytokines, including IFNγ and TNFα, did not change or minimally changed throughout infection (S1 Fig).
To evaluate the cellular response to infection, we first performed RNA sequencing (RNA-seq) on PBMCs. To identify induced and repressed genes during infection we compared transcriptomes at d3, d6 and d17 POS with d48 (convalescent) (Fig 2B). Hierarchical clustering of normalized PBMC transcriptional profiles showed dynamic induction patterns with strong d3 up-regulation of many interferon-stimulated genes (ISGs), which steadily declined at d6 and d17. Published human PBMC studies during acute DENV infections demonstrated sequential waves of gene expression with early induction of ISGs and inflammatory chemokines followed by a switch to induction of genes involved in cell proliferation [34]. During ZIKV infection in this individual, there was similar strong induction of type I ISGs, exemplified by MX1, OAS3, RSAD2, and IFI27 genes (Cluster 2), but minimal coincident induction of chemokines involved in leukocyte chemotaxis (Cluster 1) (Fig 2B). This includes CXCL10 and CCL2, that encode the chemokines IP-10 and MCP-1, that were elevated at the protein level d3 POS. Genes associated with cell differentiation and proliferation, such as BUB1, DLGAP5, PBK and CEP55 (Cluster 3) were upregulated during DENV infection but not ZIKV, while EGR1, HBEGF and MAFB (Cluster 4), were up-regulated at d17 POS during ZIKV infection (Fig 2B).
To better understand if low-level cytokine gene induction in PBMCs was characteristic of ZIKV infection, we analyzed a published study where temporal gene expression profiles were measured in rhesus macaques following ZIKV infection [33]. PBMCs from our patient and rhesus macaques showed similar early transcriptional upregulation of ISGs but minimal chemokine gene induction with the possible exception of CXCL10 in monkeys (Fig 2C).
Analyzing PBMC transcription and serum cytokines provides important information about global immune responses but lacks cell population-level resolution. To better understand how individual cell populations responds to ZIKV infection, we isolated three monocyte subsets; classical, intermediate, and non-classical; natural killer (NK) cells; two dendritic cell (DC) subsets; myeloid DCs (mDCs) and plasmacytoid DCs (pDCs); as well as CD4+ and CD8+ T cells at d3, d6, d17 and d48 POS using Fluorescence-activated cell sorting (FACS) (S2 Fig). RNA-seq transcriptional analysis of individual cell types and PBMCs together identified 1,147 genes induced at least 2-fold at d3, d6, or d17 when compared to d48 (Fig 3A). A similar analysis of PBMCs alone identified only 452 induced genes (Fig 3A). Innate immune cells (monocytes and DCs) induced the highest number of genes on d3 POS (Fig 3B). Genes up-regulated in innate immune subsets were most enriched for functional annotations associated with interferon (IFN) and immune responses at d3 and d6 POS (Fig 3C). Additionally, the promoters of genes induced at d3 and d6 POS in innate immune cells were most significantly enriched for ISRE, IRF-composite and STAT1 motifs (Fig 3D). Together, this data is consistent with early activation of type I IFN responses in innate immune populations through activation of interferon regulatory factors (IRFs) and interferon-stimulated gene factor 3 (ISGF3) transcription factors [35, 36]. In contrast to innate immune cells, the peak of T cell gene up-regulation was delayed (Fig 3B). Genes induced in CD8+ T cells were functionally enriched for terms associated with cell cycle progression such as E2F and MYC targets and G2M checkpoint (Fig 3C). Additionally, the promoters of these induced genes were enriched for E2F, NFY and POU binding motifs where transcription factors involved in controlling cell cycle and cell differentiation can bind (Fig 3D). Like innate immune populations, NK cells responded rapidly to infection by inducing IFN pathways (Fig 3C). However, NK cells also activated cell cycle progression pathways like CD8+ T cells but did so earlier, d3 compared to d6 POS, during infection (Fig 3C and 3D). Although PBMC analysis identified fewer induced genes, PBMC functional and promoter motif enrichment analyses captured many core components observed in both individual innate immune and T cell analyses (Fig 3C and 3D).
An unbiased analysis of gene expression profiles using hierarchical clustering of the top induced genes in all individual cell types and PBMCs revealed both temporal and cell type specific patterns of gene expression (Fig 3E). Gene expression at early time points, d3 and d6 POS, generally cluster together and apart from d17 responses (Fig 3E). The exception is T cells, where only d6 POS gene expression cluster in the early group. The genes driving this difference are largely induced in a time dependent manner, with anti-viral genes (Cluster B) being up-regulated early and other immune pathway genes (Cluster D) later in infection (Fig 3E). Additionally, at d3 and d6 POS, transcriptional responses cluster by cell type suggesting early transcriptional responses are in part cell type specific (Fig 3E). Population-specific induction of genes is evident from genes that are up-regulated exclusively in pDCs (Cluster A) or NK and T cells (Cluster C, Fig 3E). In contrast to all other cell types, classical and intermediate monocytes cluster together based on day POS suggesting that gene induction in these two cell types is more dependent on time POS than cell type. Many genes, including AIM2, an ISG involved in inflammasome activation in macrophages, is induced in both time and cell-type specific manners (Fig 3F and 3G). Additionally, CXCL10, CCL2, and IL1RN that encode the cytokines, IP-10, MCP-1 and IL1RA, were upregulated at least 2-fold in certain monocyte populations at d3 POS even though they were not significantly induced in PBMCs as a whole (S3 Fig).
Chromatin accessibility is a major component of genome regulation. Open regions of chromatin are putatively associated with genomic regulatory regions, including both promoters and enhancers. The Assay for Transposase Accessible Chromatin using sequencing (ATAC-seq) can be used to identify transcription factors (TFs) involved in regulating important functions, such as differentiation and gene regulation through the analysis of open chromatin. We did not obtain high quality d3 ATAC-seq data. However, high quality ATAC-seq data were produced using samples from d6 and d17 POS. Comparing ATAC-seq peaks in classical monocytes with NK cells on d6 POS we identified 13,792 and 13,200 peaks unique to classical monocytes and NK cells respectively. De novo motif analysis of these peaks identified PU.1, CEBP and AP-1 in monocytes and ETS1, RUNX and T-box in NK cells as the most enriched TF binding motifs (Fig 4A). Each of these TFs have been identified as important lineage-determining transcription factors (LDTFs) in monocytes and NK cells, respectively. To investigate TFs that may be important during the cellular response to infection we examined dynamic changes in chromatin accessibility over time. We identified 1,493 and 1,261 ATAC-seq peaks that were significantly upregulated at d6 POS compared to d17 POS in classical monocytes or NK cells respectively. In addition to cell type specific LDTFs in monocytes and NK cells, motif analysis of upregulated ATAC-seq peaks at d6 POS demonstrated increased enrichment of PU.1:IRF8 and bZIP TF binding motifs in classical monocytes (Fig 4B). In contrast, ISRE/IRF motifs were equally represented in regulated ATAC-seq peaks in classical monocytes and NK cells (Fig 4B).
To help illustrate how these data characterize individual gene loci, we considered the open chromatin landscape at genes with both common and cell-type specific patterns of regulation. Both classical monocytes and NK cells upregulated the ISGs IFIT2 and IFIT3 early in infection and ATAC-seq peaks were identified at sites containing ISRE motifs (Fig 4C). Although the ISRE associated peaks were common at these loci, the other ATAC-seq peaks were monocyte or NK specific and were associated with TF binding motifs enriched in the corresponding cell-type. This suggests that although both cell types induce IFIT2 and IFIT3 they may utilize cell type specific TFs to help regulate gene expression. Another ISG, APOBEC3A, was induced in monocytes but not in NK cells (Fig 4D). At this gene locus, the ATAC-seq peaks were all monocyte specific and were associated with monocyte-enriched TF binding motifs (Fig 4D). The gene MKI67 encodes the protein Ki-67 and is a marker of proliferation. This gene was induced in NK cells early in infection but was never induced in classical monocytes (Fig 4E). The ATAC-seq peaks associated with this gene are NK-specific and associated with NK enriched TF binding motifs except for one common peak associated with an E2F motif (Fig 4E). These examples help illustrate how open chromatin patterns associated with cell-type specific transcription factors may play a role in defining common and cell-type specific patterns of gene expression (Fig 4D and 4E).
We next evaluated the temporal development of adaptive immune responses. Prior to the acute ZIKV infection, this individual had low but detectable neutralizing Abs to both DENV and ZIKV strains (Fig 5A). Neutralizing Ab titers to ZIKV and DENV rapidly increased after infection, peaking on d6 POS (Fig 5A–5D). The highest neutralizing titer 50 (NT50) developed against the patient’s own virus followed by the related ZIKV FSS13025 (Cambodia, 2010) (Fig 5A and 5B) [37]. The NT50 also increased against both DENV1 and DENV4 but to a lesser degree than either ZIKV strain. These results are consistent with the idea that ZIKV infection can induce cross-reactive neutralizing Ab responses to DENV especially in individuals with prior flavivirus experience with faster kinetics relative to naïve people [17, 19].
Lastly, we assessed the T cell response by flow cytometry. Published studies have shown that the majority of DENV-specific and ZIKV-specific T cells display an effector or memory phenotype based on expression of CD45RA and CCR7 [18, 38, 39]. Moreover, in secondary DENV infections, the T cell response is associated with an expansion of T effector memory RA (TEMRA) and T effector memory (TEM) cells that can be more vigorous than in primary DENV infection [39]. Accordingly, our data on bulk populations of unstimulated T cells showed higher proportions of CD4+ TEMRA cells and lower proportion of naïve CD8+ T cells (TN) at d6 POS as compared to 3 healthy DENV-naïve and 2 DENV-immune control individuals (Fig 6A and 6B). We also examined CD4 T helper (Th) subsets based on the expression of chemokine receptors (Th1: CCR6-CCR4-CXCR3+; Th2: CCR6-CCR4+CXCR3-; Th1/17: CCR6+CCR4-CXCR3+; and Th17: CCR6+CCR4+CXCR3-). No specific Th profile was observed in this individual (S2 Table), consistent with the published observation that the majority of DENV-specific CD4+ T cells are not associated with common Th subsets [39].
Studies of DENV-infected individuals have suggested that expanded CD4+ TEMRA cells can exhibit a virus-specific cytotoxic phenotype that has been associated with protection against severe DENV disease [39, 40]. Cytotoxic CD4 T cells are CD45RA+CCR7- (TEMRA) with increased expression of CD8α, cytotoxic effector molecules such as granzyme B and perforin, and CD226, a co-stimulatory molecule that enhances CD8 effector and cytotoxic functions. A CD4+ T cell population with low level CD8 expression (CD4+CD8dim) was detected in our individual with acute ZIKV patient (Fig 6C). The frequency of this population was between 3.9 and 9.1% of all CD3+ T cells on d3, d6, d17 and d48 POS but decreased to 1.1% by day d240 (Fig 6C and 6D). The CD4+CD8dim population was less than 0.6% in three ZIKV-naïve controls (Fig 6C). In two DENV-immune individuals this population was 0.6% and 2.7% of all T cells (Fig 6C and 6D). At d3 POS 52.4% of CD4+CD8dim cells were also CD45RA+CD226+ and negative for three chemokine receptors (CRs) CCR6, CCR4 and CXCR3 (Fig 6E and 6F). By d240 POS the frequency of CD4+CD8dim cells that were CD45RA+CD226+CR- fell to 0.2%. In the DENV-immune control with a significant CD4+CD8dim population, 14.5% of this population was CD45RA+CD226+CR- (Fig 6E and 6F). Based on these markers, the CD4+CD8dimCD45RA+CD226+CRs- subset is likely to be cytotoxic CD4+ T cells.
Herein, we combine global PBMC and cell type-specific transcriptional and epigenetic analyses to characterize the development and resolution of an in vivo human immune response to an acute viral infection. This is a single patient study and therefore broad conclusions cannot be drawn. However, given the limited number and scope of published ZIKV in vivo response data, we feel that our study presents a unique and detailed perspective of both the innate and adaptive immune responses to ZIKV, and provides important considerations for designing future studies.
Approximately ten years prior to the ZIKV infection reported here, this individual was infected with DENV. She has no known subsequent exposure to DENV or ZIKV and has not lived in an endemic region where exposure is likely during this interval. Studies have demonstrated that prior DENV exposure influences the timing, magnitude, and quality of adaptive immune responses to ZIKV infection [18, 19]. The influence of prior DENV-exposure on innate immune responses are not understood. ZIKV is transmited by the same vector as DENV and circulates in geographical regions where DENV is endemic or hyper-endemic. Morever, ZIKV vaccine candidates have been designed for testing and deployment in DENV-endemic countries. Thus, understanding ZIKV immune responses in individuals with DENV-immunity is highly relevant.
Our analyses of PBMC and cell-specific responses demonstrate that, during acute ZIKV infection, a robust type I IFN transcriptional response was induced at early time points. Based on promoter motif analysis, this IFN response is likely driven by activation of JAK/STAT and IRF transcription factor signaling. Induction of ISG genes broadly are common to all innate immune cells tested, including monocytes (classical, intermediate and non-classical), mDCs, pDCs and NK cells. However, induction of some individual ISGs, such as AIM2, are induced in cell-type specific manners. Transcription analysis of bulk PBMCs is sufficient to capture a significant proportion of the response at both a pathway and gene-specific level but individual cell analysis identifies specific gene regulation and the cell type responsible for those responses during ZIKV infection that is not appreciated in the PBMC analysis. ATAC-seq enables assessment of enhancer elements distal from promoters that play important roles in modulating the immune response. This assay identified common changes in ISRE/IRF motifs in NK and classical monocytes, but also indicated substantial chromatin remodeling at sites containing cell-type specific TF binding motifs that help to explain the observed changes in gene expression. The ATAC-seq analysis was limited to d6 and d17 samples. ATAC-seq analysis at earlier time points or inclusion of other cell types could provide higher resolution of time- and cell-type dependent changes in chromatin accessibility.
Analysis of ZIKV infected cohorts in Brazil and Singapore demonstrated elevations in many serum cytokines, including IFNγ, MCP-1, IL1RA, IL-18, IL-10, IP-10 and TNFα [13, 14]. We found similar elevations in IP-10, MCP-1 and IL1RA at d3 POS but other cytokines tested showed smaller variation that is difficult to interpret. At the transcriptional level, PBMCs showed minimal induction of most chemokine genes. This included the genes CXCL10, CCL2 and IL1RN, that encode for IP-10, MCP-1 and IL1RA. This low-level chemokine gene induction was similar to what was observed during in vivo ZIKV infection of macaques [33]. In contrast to PBMCs, CXCL10, CCL2 and IL1RN were induced at least 2-fold in specific monocyte populations. Thus, specific populations, such as monocytes, or non-PBMCs may be the source of elevated IP-10, MCP-1 and IL1RA.
In our individual, neutralizing Ab titers increased rapidly post infection, peaking at d6 POS. We observed robust increases in neutralizing anti-ZIKV Ab responses with more modest increases in cross-reactive DENV-neutralization titers. This pace of neutralizing response is consistent with previous findings that humoral responses develop rapidly in DENV-immune individuals [19]. Analysis of serum from 8 months prior to infection revealed this individual had pre-existing low level neutralizing Ab titers (1:83) against ZIKV SD001 that did not prevent symptomatic ZIKV infection. Previous studies have demonstrated that individuals with remote exposure to DENV infrequently have cross-neutralizing Abs to ZIKV [41, 42]. In two studies, 0 of 19 [41] and 3 of 17 (18%) convalescent-phase [42] sera from recovered individuals with single DENV infections, had detectable cross-neutralizing Abs against ZIKV. Among persons exposed to repeat DENV infections, 3 (23%) of 13 [41] and 6 of 16 (38%) [42] convalescent-phase sera had ZIKV neutralizing Abs. Most of these individuals who develop cross-neutralizing Abs against ZIKV had relatively low Ab titers (<1:100). During DENV infections, higher levels of cross-reactive pre-infection neutralizing Ab titers in humans correlate with reduced probability of symptomatic secondary DENV infection [43]. In our individual, prior DENV exposure induced low-level ZIKV cross-neutralizing Abs that did not protect against subsequent ZIKV infection.
During our T cell phenotyping, we found a significant CD4+CD8dim T cell subset on d3 through d48 POS that largely resolved by d240 POS. Previous studies have shown that CD4+CD8dim T cell populations can be highly enriched for cells recognizing DENV, HCMV and HIV antigens [39, 44, 45]. In our patient, more than 50% of the CD4+CD8dim T cells during acute ZIKV infection were CD45RA+CD226+CR-. This expression pattern is suggestive of cytotoxic T cells, a population not yet reported during ZIKV infection. Increased frequencies of these cells have been observed after primary and secondary DENV infections, particularly in individuals expressing HLA alleles that are associated with protection against DENV [39]. Our study provides a rationale and framework for investigating the importance of the CD4+CD8dim T cell response in ZIKV immunity.
Collectively, these results detail the global and cell type-specific innate immune responses during an acute ZIKV infection and highlight the rapid development of neutralizing Ab and effector memory T cell responses in a DENV experienced host. These data supports accumulating evidence that prior exposure to DENV accelerates and alters adaptive immune responses likely via the presence of cross-reactive epitopes [16–19, 46]. Measuring time point- and cell type-specific transcriptional signatures of innate vs. adaptive immune cell populations in the blood of individuals with and without a history of flavivirus infection and vaccination can elucidate how prior flavivirus exposure might alter the magnitude, specificity, breadth, phenotype, and functionality of both humoral and cellular immune response to ZIKV. Our findings indicate that information is lost using conventional approaches and that genomic assays have the potential to provide substantial additional mechanistic insight. Combining detailed longitudinal systems biology analysis with classic immunologic techniques in future clinical studies has great potential to improve our understanding of human immune responses to pathogens at a broad level by identifying communication pathways that connect innate and adaptive immunity and regulate the balance between protection and pathogenesis. More urgently towards solving the global ZIKV and DENV problem, this approach may be invaluable in investigating the human immune response in the context of natural infection and vaccination, thereby leading to the generation of ZIKV and DENV vaccines with maximal safety and efficacy.
|
10.1371/journal.pgen.1005843 | The Replisome-Coupled E3 Ubiquitin Ligase Rtt101Mms22 Counteracts Mrc1 Function to Tolerate Genotoxic Stress | Faithful DNA replication and repair requires the activity of cullin 4-based E3 ubiquitin ligases (CRL4), but the underlying mechanisms remain poorly understood. The budding yeast Cul4 homologue, Rtt101, in complex with the linker Mms1 and the putative substrate adaptor Mms22 promotes progression of replication forks through damaged DNA. Here we characterized the interactome of Mms22 and found that the Rtt101Mms22 ligase associates with the replisome progression complex during S-phase via the amino-terminal WD40 domain of Ctf4. Moreover, genetic screening for suppressors of the genotoxic sensitivity of rtt101Δ cells identified a cluster of replication proteins, among them a component of the fork protection complex, Mrc1. In contrast to rtt101Δ and mms22Δ cells, mrc1Δ rtt101Δ and mrc1Δ mms22Δ double mutants complete DNA replication upon replication stress by facilitating the repair/restart of stalled replication forks using a Rad52-dependent mechanism. Our results suggest that the Rtt101Mms22 E3 ligase does not induce Mrc1 degradation, but specifically counteracts Mrc1’s replicative function, possibly by modulating its interaction with the CMG (Cdc45-MCM-GINS) complex at stalled forks.
| Post-translational protein modifications, such as ubiquitylation, are essential for cells to respond to environmental cues. In order to understand how eukaryotes cope with DNA damage, we have investigated a conserved E3 ubiquitin ligase complex required for the resistance to carcinogenic chemicals. This complex, composed of Rtt101, Mms1 and Mms22 in budding yeast, plays a critical role in regulating the fate of stalled DNA replication. Here, we found that the Rtt101Mms22 E3 ubiquitin ligase complex interacts with the replisome during S-phase, and orchestrates the repair/restart of DNA synthesis after stalling by activating a Rad52-dependent homologous recombination pathway. Our findings indicate that Rtt101Mms22 specifically counteracts the replicative activity of Mrc1, a subunit of the fork protection complex, possibly by modulating its interaction with the CMG (Cdc45-MCM-GINS) helicase complex upon fork stalling. Altogether, our study unravels a functional protein cluster that is essential to understand how eukaryotic cells cope with DNA damage during replication and, thus deepens our knowledge of the biology that underlies carcinogenesis.
| DNA replication is a process through which cells duplicate their entire genome prior to cell division. To achieve accurate replication, eukaryotes have evolved intricate surveillance systems that allow fine-tuning of the replication machinery. In order to continually provide the replicative polymerase with a single stranded DNA template, replisomes must adapt to chromatin heterogeneities such as aberrant DNA structures, condensed chromatids, transcriptional obstacles and DNA-protein barriers [1]. In Saccharomyces cerevisiae, this adaptation is regulated by proteins such as Mrc1, Tof1, Csm3 and Ctf4, which assemble around the CMG (Cdc45-MCM-GINS) DNA helicase at replication forks. These components form the ‘Replisome Progression Complex’ (RPC), a replisome sub-assembly that exists exclusively at replication forks [2]. The RPC functions in coupling DNA polymerases to the CMG helicase [3,4], and in regulating fork progression [5–9]. Moreover, these replisome components also limit mutagenic frequency and prevent unscheduled homologous recombination (HR) events at stalled forks [10–12]. Mrc1 possesses two polymerase epsilon (Pol ε) binding sites [13] as well as an Mcm6 interaction motif [14], and is required for checkpoint activation in response to replication stress [15]. Tof1 and Csm3 help to link Mrc1 to fork components [16,17] but also have distinct functions not shared with Mrc1 [9]. Conversely, the replication fork progression defect is enhanced in mrc1Δ compared to tof1Δ and csm3Δ mutants [1], implying that Mrc1 also promotes replication functions independent of Tof1 and Csm3. Ctf4, the yeast homologue of human AND1, bridges the interaction of the primase, DNA polymerase-α, to the CMG helicase [3,5,18]. Although the coupling of the CMG to leading and lagging strand DNA polymerases preserves genome integrity during unperturbed DNA replication, this mechanism is partially disrupted when forks encounter replication stress. Indeed, significant stretches of ssDNA generated by uncoupling can promote HR-mediated replication re-start via either a template switch or break-induced replication (BIR) [19].
Growing evidence implicates Cullin-RING containing E3 ligases (CRL’s) in regulating DNA replication and repair [20]. For example, Cdc53/Cul1 in a complex with the F-box adaptor protein Dia2 (SCFDia2) promotes the ubiquitylation of the Mcm7 subunit of the CMG helicase, which triggers Cdc48/p97-dependent disassembly of CMG at the end of DNA replication [21]. In addition, SCFDia2 has been reported to antagonize Mrc1 upon replication stress, possibly by inducing degradation of phosphorylated Mrc1 to recover from checkpoint arrest following repair [22–24]. Cullin4 (CRL4)-based E3 ubiquitin ligases regulate DNA replication and repair both in yeast and mammalian cells, in part by controlling histone dynamics at active replication forks [25]. Rtt101, the budding yeast analogue of human Cul4 [26], has been reported to target Spt16, a subunit of the FACT complex that reorganizes nucleosomes during DNA replication [27]. Moreover, Rtt101 promotes replication fork progression through DNA lesions and natural pause sites [28]. This function is dependent on MMS1 and MMS22, which encode a DDB1-like linker protein and a putative substrate specific adaptor, respectively [26]. However, the underlying mechanisms and function of the Rtt101-Mms1-Mms22 complex (termed Rtt101Mms22) remain largely elusive. Recent results suggest that the sensitivity of mms1Δ, mms22Δ and rtt101Δ cells to the DNA-damaging drugs CPT and MMS could be rescued by further deleting MRC1 [18,29], possibly by de-regulating late firing origins [29].
Here we show that the Rtt101Mms22 E3 ubiquitin ligase genetically and biochemically interacts with components of the replication fork. We found that the WD40 domain of Ctf4, a protein required for coupling the CMG complex to the replicative polymerases, recruits Mms22 to active forks during S-phase. Moreover, Mms22 physically associates with Mrc1 and deletion of the MRC1 gene suppresses the defects of rtt101Δ, mms1Δ and mms22Δ cells, including genotoxic sensitivity, prolonged checkpoint activation and reduced HR rates. Importantly, our results suggest that de-repression of late replication origins is not sufficient to bypass the need of Rtt101Mms22 E3-ligase activity. Instead, MRC1 deletion promotes HR-mediated repair of replication forks that have paused in response to replication stress. Based on genetic and biochemical data we propose that the Rtt101Mms22 complex specifically counteracts the replicative, and not the checkpoint, function of Mrc1, possibly by modulating its interaction with the CMG helicase complex upon fork stalling.
To elucidate the role of the Rtt101Mms22 E3 ubiquitin ligase in DNA replication, we employed an automated SGA approach [30] to screen for genes that, when deleted, would suppress the growth defects of rtt101Δ cells exposed to either the alkylating agent methyl methanesulfonate (MMS) or the topoisomerase 1 (Top1) poison camptothecin (CPT) [31] (Figs 1A and S1). The SGA screen was performed in duplicate and only suppressors that appeared in both screens were considered for further analysis. The combined results for MMS and CPT conditions initially identified 63 suppressor genes that were scored by visual inspection as either strong, medium or weak (S1 Table). Only the strong and medium hits were validated by crossing the single deletion mutants to an independent rtt101Δ strain, in which double mutants were derived by manual tetrad dissection. Cells were then spotted in biological duplicate on MMS and CPT containing media as depicted in Fig 1C–1E. Using this workflow, we confirmed a list of 16 genes that when deleted improved the growth of rtt101Δ cells (Fig 1B). Among the most potent suppressors are genes directly involved in DNA replication, including MRC1, POL32, RAD27, TOP1, SIZ2 and DPB4. Additional suppressor mutations implicated in other biological processes were also confirmed (Fig 1B), but were not further characterized in this study. Some suppressors such as the lagging strand polymerase subunit Pol32 only restored growth on CPT containing media (Fig 1B–1D), indicating that the screening approach could isolate functional protein sub-clusters. The slow growth phenotype of other Rtt101-based E3 ubiquitin ligase component mutants, including those lacking the linker protein Mms1 (Fig 1D) as well as mutants deleted for the putative substrate specific adaptor Mms22 (Fig 1E), were rescued by deleting the identical replication gene cluster that suppressed the rtt101Δ cell phenotype. These data indicate that Rtt101 likely acts as a fully assembled E3 ubiquitin ligase in a process associated with replication stress. Indeed, deletion of MRC1 suppressed the growth defects of cells expressing the neddylation-deficient Rtt101-K791R mutant [32], implying that loss of MRC1 suppresses the phenotypes correlated with inactivation of Rtt101Mms22 E3 ligase activity (S2 Fig). In contrast, deletion of MRC1 did not rescue the MMS sensitivity of ubc13Δ cells defective for PCNA polyubiquitylation (S3 Fig), indicating that the genetic suppression is specific to the loss of Rtt101Mms22 function and not other ubiquitylation-defective mutants involved in lesion bypass repair [33].
MRC1 and DPB4 are both linked to the putative leading strand polymerase, Pol ε, and deletion of these genes showed suppression on both MMS and CPT containing media, albeit to varying extents. Furthermore, when the deletions of DPB4 and MRC1 were combined in the absence of either RTT101 or MMS22, we observed that the genetic rescue was not additive (Fig 1F and 1G). Unlike in rtt101Δ or mms1Δ cells, the deletion of POL32 did not rescue the CPT sensitivity of mms22Δ cells (Fig 1E), supporting the notion that Mms22 has additional functions independent of the E3 ligase complex [34]. Taken together, these data demonstrate that deletion of replication genes such as MRC1 and DPB4 can alleviate the growth defects associated with impaired Rtt101Mms22 E3 ubiquitin ligase activity in response to multiple genotoxic agents.
The above genetic data suggests that the Rtt101Mms22 complex may directly interact with the replisome. Since specificity of a CRL complex is mainly conferred by the substrate adaptor [35], we immunoprecipitated Mms22 from S-phase synchronized cells and identified associated proteins by an unbiased mass-spectrometry method referred to as shotgun LC-MS/MS (Fig 2A). As expected, Mms22 co-purified with the E3 ligase subunits Rtt101, Mms1 and Hrt1, with the core histones Htb2, Hhf1, Hta1, and Hht1, as well as with the FACT complex (Spt16, Pob3), a nucleosome re-organizer that likely facilitates the interactions between DNA replication and transcription. These findings are consistent with previously published data showing that Rtt101-Mms1 associates with histone H3 [25] and ubiquitylates the Spt16 subunit of FACT [27]. We also detected a cluster of replication factors, including components of the GINS- (go-ichi-ni-san), the fork protection- and the MCM helicase complexes (Fig 2A, see also S2 Table for a complete list of Mms22-interacting proteins). These results strengthen the genetic interaction clusters that were found to suppress the growth defects of rtt101Δ cells, and strongly suggest a function of Rtt101Mms22 at replisomes during S-phase.
To better characterize the interaction of Mms22 with replisome components, we immunoprecipitated functional epitope-tagged versions of Mms22 (PA-Mms22) (Fig 2C) and the GINS complex (TAP-Sld5) (Fig 2D) from cells synchronized in G1 and S-phase with normal or genotoxic stress (0.03% MMS) growth conditions (Fig 2B). We observed that Mms22 and Rtt101 interact with all tested components of the replisome progression complex during S-phase (Fig 2C and 2D). Remarkably, Ctf4 and the FACT complex showed affinity for Mms22 in both G1 and S-phase, although both interactions were more prominent during S-phase. The induction of DNA damage through MMS treatment did not alter the observed interactions, suggesting that the Rtt101Mms22 E3 ligase is not specifically recruited to replisomes upon genotoxic insult, but rather constitutively associates with the active replication machinery.
We next examined how the Rtt101Mms22 E3 ligase is recruited to active replisomes. Ctf4 was an intriguing candidate as it was previously found to interact with Mms22 [3,18,36]. Indeed, genetic analysis revealed that the growth defect of ctf4Δ cells on genotoxic drugs was epistatic with RTT101 and even slightly suppressed the sensitivity of mms22Δ cells (Fig 3A), consistent with previous findings [36]. To test whether Ctf4 tethers Mms22 to the replisome, we compared the presence of replisome components in Mms22 purifications prepared from wild-type and ctf4Δ cells (Fig 3B). Notably, Mms22 failed to interact with the replication factors Mcm2, Cdc45 and Csm3 during S-phase in ctf4Δ cells (Fig 3B), while binding to Spt16 and Pob3 FACT complex components was not perturbed (Fig 3C). Mms22 interacts with Ctf4 both by two-hybrid [3,36] and co-immunoprecipitation analysis (Fig 2A and 2C), and requires the amino-terminal WD40 domain of Ctf4 (Figs 3D and S4; [3]). Together, these results suggest that the Rtt101Mms22 E3 ligase is tethered to active replisomes via the Ctf4 scaffold (Fig 3E).
To determine if the alleviated growth defect observed in rtt101Δ mrc1Δ strains is phenocopied by the removal of other components of the fork-protection complex (FPC), we tested whether deletion of either TOF1, CSM3 alone or in combination would also rescue the sensitivity of rtt101Δ or mms22Δ cells to genotoxic agents. Neither the single nor double deletions of CSM3 and TOF1 were able to restore growth of rtt101Δ or mms22Δ cells on media containing MMS (Figs 4A and S5). These data suggest that Rtt101-Mms22 counteracts a function of Mrc1 at replication forks that is independent of its interactions with Tof1 and Csm3.
Since the replication and checkpoint functions of Mrc1 are genetically discernable, we set out to examine which functions of Mrc1 are causing lethality in the absence of Rtt101Mms22. We analyzed the genetic interaction between rtt101Δ or mms22Δ mutants and the well-characterized checkpoint-deficient MRC1-AQ allele, coding for a mutant protein in which all Mec1-dependent phospho-serines (SQ) are mutated to alanine (AQ) [37]. Interestingly, when the checkpoint-defective Mrc1-AQ variant is the only source of Mrc1 in cells lacking the Rt101Mms22 E3 ubiquitin ligase, the lethal phenotype is comparable to rtt101Δ and mms22Δ single mutant cells when exposed to genotoxic stress (Fig 4B), unlike the complete deletion of MRC1 (Figs 1C, 1E and S5). This result was surprising in light of recent data proposing that alleviation of the checkpoint-mediated late origin repression may rescue the sensitivity of rtt101Δ cells to MMS exposure [29]. To corroborate these results, we tested whether the requirement of Rtt101 to promote growth on MMS containing media could be bypassed by genetically de-repressing late origins in the sld3-37A dbf4-4A background [38]. In contrast to rtt101Δ mrc1Δ double mutants, rtt101Δ sld3-37A dbf4-4A cells were unable to rescue the sensitivity of rtt101Δ cells to MMS (Fig 4C), supporting the notion that Mrc1-dependent inhibition of late origin firing is not sufficient to explain the essential function of the Rtt101Mms22 complex in response to genotoxic stress. Conversely, the MMS-induced lethality of rtt101Δ or mms22Δ cells was suppressed by expressing an Mrc1 C-terminal truncation mutant (Mrc11-971) as the only copy of Mrc1 (Fig 4D). While this mutant has been reported to be checkpoint defective [24], the C-terminal domain of Mrc1 is known to directly interact with the C-terminal domain of Pol2 [13] and is important for the replication functions of Mrc1 [39]. To identify a bonafide separation-of-function Mrc1 allele, we thus constructed smaller C-terminal truncations of Mrc1. Strikingly, deletion of only the last C-terminal 18 amino acids (Mrc11-1078) was sufficient to confer a rescue of MMS sensitivity in both rtt101Δ and mms22Δ cells (Fig 4E, compare bottom two rows), while otherwise checkpoint defective cells expressing the Mrc11-1078 mutant protein were able to activate the replication checkpoint when challenged with replication stress (Fig 4F). Together, these results strongly suggest that loss of Mrc1’s checkpoint function is not responsible for the genetic suppression of rtt101Δ or mms22Δ cells and implicates the Mrc1 replicative functions in the observed cell toxicity. Interestingly, the inability of rtt101Δ or mms22Δ cells to extinguish the DNA damage checkpoint following MMS recovery [28] was alleviated in rtt101Δ mrc1Δ or mms22Δ mrc1Δ strains (Fig 4G), and as a consequence rtt101Δ mrc1Δ double mutants proceeded into the next G1 phase four hours post-recovery whereas the rtt101Δ cells remained largely arrested at the G2/M border as expected (Fig 4H). Based on these data we conclude that Rtt101Mms22 specifically counteracts a replicative function of Mrc1 at stalled replisomes, thereby promoting replication fork repair/restart, which leads to an eventual checkpoint termination.
Increasing evidence points towards a key HR function during replication fork restart at stalled replisomes [40]. Since Mrc1 is a known suppressor of HR [11], we hypothesized that its removal may promote restart of damaged replication forks in rtt101Δ and mms22Δ cells by an HR-dependent mechanism. To assess HR rates we used a previously described reporter system [41] that exploits a plasmid as a recombination substrate for both the single-strand invasion and annealing pathways, resulting in CAN1 gene deletion (S6 Fig). In agreement with previous studies, HR levels were abolished in rad52Δ and reduced in mms22Δ strains [42], whereas mrc1Δ cells exhibited increased recombination rates compared to wild-type controls ([12] and Fig 5A). Interestingly, mms22Δ mrc1Δ double mutants showed a level of HR comparable to mrc1Δ cells, suggesting that in the absence of MMS22, Mrc1 may block recombination (Fig 5A). In contrast, ctf4Δ mrc1Δ double mutants are inviable ([3,43], S7 Fig), implying that Mrc1 and Ctf4 share a Mms22-independent essential function during DNA replication. The increased HR phenotype seems to be specific to Mrc1, as another mutant, sgs1Δ, with increased HR levels did not alleviate the genotoxic sensitivity (S8 Fig) or the low HR frequency [34] observed in mms22Δ strains. Since the increased HR level correlated positively with the observed genetic suppression in mms22Δ mrc1Δ cells, we tested whether HR was required for this suppression by deleting the RAD52 gene, and thereby rendering cells HR defective. Indeed, the hyper-sensitivity of rtt101Δ rad52Δ and mms1Δ rad52Δ cells to MMS could no longer be rescued upon further deletion of MRC1 (Figs 5B, top and S9), consistent with the notion that an intact HR machinery is required for suppression. A similar effect was also observed in cells lacking the substrate adaptor Mms22 (Fig 5B, bottom), but in this case a slight mrc1Δ rescue was still observed in the rad52Δ background. These data imply that in contrast to rtt101Δ and mms1Δ, deletion of MRC1 in mms22Δ cells rescues MMS sensitivity, in part, in a RAD52-independent manner.
To corroborate these results, we released G1 synchronized cells endogenously expressing mCherry-tagged Rad52 (Rad52-mCherry) into media containing MMS and scored the ability of cells to form Rad52 foci, a proxy for active recombination [44]. As expected, the percentage of cells with Rad52 foci decreased in both rtt101Δ and mms22Δ cells, but strikingly, this defect was corrected by further deleting MRC1 (Fig 5C and 5D). Together, these data indicate that HR upregulation induced by the loss of Mrc1 function may play a key role in the restart/repair of defective replication forks in cells lacking Rtt101Mms22 E3 ligase activity.
Available evidence suggests that SCFDia2 targets phosphorylated Mrc1 for proteasomal degradation [22,24]. To test whether Mrc1 is degraded by a Rtt101Mms22-dependent mechanism, we monitored Mrc1 stability in synchronized cells with MMS–induced replication stress (Fig 6A). Promoter shut-off by glucose addition to cells expressing galactose-inducible 3HA-Mrc1 (Figs 6A–6C and S10) as well as cyclohexamide (CHX) chase experiments of endogenously tagged 3HA-Mrc1 (S10 Fig) showed that the degradation kinetics of phosphorylated and unphosphorylated Mrc1 in rtt101Δ or mms22Δ cells exposed to MMS were comparable to wild-type controls. These results were corroborated by quantitative mass spectrometry using selected reaction monitoring (SRM), which demonstrated that Mrc1 levels decreased with similar kinetics in wild-type, rtt101Δ and mms22Δ cells (S10 Fig). Surprisingly, deletion of DIA2 had only slight effects on the half-life of Mrc1, and thus the role of SCFDia2 in regulating Mrc1 stability remains to be further clarified [24,45]. Taken together, these data demonstrate that Rtt101Mms22 does not trigger degradation of the Mrc1 protein at stalled replication forks.
Tetrad analysis and plating assays revealed that the growth defects caused by loss of Rtt101 and Mms22 together with the loss of Dia2 function were additive, and double mutants were further impaired for growth in the presence and absence of MMS (Fig 6D and 6E). This indicates that the two E3 ligases may function independently and have non-overlapping roles during DNA replication. Interestingly, deletion of MRC1 restored some of the growth defects of mms22Δ dia2Δ and rtt101Δ dia2Δ double mutants as shown by tetrad analysis (Fig 6D) as well as spotting assays on MMS containing media (Fig 6E). Conversely, overexpression of Mrc1 resulted in toxicity in rtt101Δ dia2Δ and mms22Δ dia2Δ double mutants even in the absence of exogenous genotoxic stress (Fig 6F and 6G). Together these results indicate that these two CRLs may genetically interact with Mrc1 function but likely by distinct mechanisms.
In addition to binding DNA polymerase ε [13], Mrc1 also interacts with the Mcm6 subunit of the MCM2-7 helicase [14]. In order to disrupt the binding of Mrc1 to Mcm6, we crossed the mcm6-IL allele [14] into rtt101Δ, mms22Δ and mms1Δ cells deleted for MRC1, and compared growth of the resulting single, double and triple mutants treated with DNA damaging agents (Fig 7A). Importantly, we observed that similar to deleting MRC1 the presence of mcm6-IL was able to suppress the sensitivity of rtt101Δ, mms22Δ (albeit weakly) and mms1Δ cells to MMS (Fig 7B–7D), indicating that disrupting the association of Mrc1 with the CMG helicase is sufficient to restore growth of rtt101Δ, mms1Δ, and in part, mms22Δ mutants exposed to genotoxic stress. Importantly, deletion of MRC1 in rtt101Δ mcm6-IL or mms22Δ mcm6-IL cells did not further improve growth on MMS containing media (Fig 7B and 7C), suggesting that these mutations affect the same molecular process. We did not observe a difference in the amount of chromatin bound Mrc1 when comparing wild-type to rtt101Δ and mms22Δ cells in the absence and presence of MMS, rendering it unlikely that Rtt101Mms22 promotes Mrc1 eviction from chromatin at sites of replication stress (S11 Fig). These genetic and biochemical results suggest that the Rtt101Mms22 E3 ligase counteracts a function of Mrc1 that is linked to the replicative helicase at stalled replication forks, and possibly modulates the interaction of Mrc1 with the MCM helicase complex upon fork stalling (Fig 7E).
In this study, we have employed genome-wide genetic screening together with a proteomics approach to gain insight into how the Rtt101Mms22 E3 ubiquitin ligase regulates stalled replication forks [28]. We demonstrate that Rtt101Mms22 is recruited to replisomes during S-phase by interacting with the N-terminal WD40 domain of Ctf4. Moreover, genetic analysis revealed that the introduction of the mcm6-IL allele, or deletion of either MRC1 or the non-essential Pol ε subunit, DPB4, can restore viability of rtt101Δ, mms1Δ and mms22Δ cells exposed to various genotoxic stress conditions. Together, our data suggest that the Rtt101Mms22 complex acts directly at replisomes to promote repair and restart of stalled replication forks by counteracting a replicative function of Mrc1, and hence promoting HR.
The tight association of DNA synthesis with the unwinding activities of the replicative helicase at replisomes is important for faithful DNA replication. Mrc1 represents a plausible candidate to reinforce this association as it physically interacts with both Pol2 as well as the MCM2-7 helicase. Indeed, altering this interaction may lead to exposed stretches of ssDNA, which are vulnerable to nicking and chemical modifications, and if extensive enough, may unleash the replication checkpoint following the association of sufficient replication protein A (RPA) molecules. In accordance, in the presence of MMS we observed more extensive Rad53 phosphorylation in the absence of MRC1 (Fig 4G). Previous studies have shown that mrc1Δ cells fail to inhibit late firing origins [6,46], an effect that could be used to bypass the adverse effects of stalled replication forks. Therefore, deletion of MRC1 might conceivably alleviate the DNA damage sensitivity of rtt101Δ and mms22Δ cells by allowing the firing of additional origins and hence promoting the completion of replication [29]. In light of our results, however, this possibility seems unlikely given that the loss of Mrc1’s checkpoint functions fails to rescue the defects associated with rtt101Δ and mms22Δ cells. Moreover, the requirement of the Rtt101Mms22 complex in genotoxic stress conditions could not be bypassed by genetically de-repressing late origins using the sld3-37A dbf4-4A background [38]. Alternatively, our results suggest that Mrc1 deletion, and in turn replisome uncoupling, may promote HR-mediated fork restart at stalled replication forks [11]. In support of this idea, Rtt101, Mms1 and Mms22 have been demonstrated to stimulate HR, specifically upon exposure to genotoxic agents [42], and we found that deletion of MRC1 rescued the reduced recombination rates in cells deleted for MMS22. Indeed, an intact HR machinery is required for rtt101Δ mrc1Δ and to a lesser extent mms22Δ mrc1Δ double mutants to grow in genotoxic stress conditions. Moreover, recombination foci visualized microscopically by Rad52-mCherry were decreased in both rtt101Δ and mms22Δ cells exposed to genotoxic drugs, but were restored by additionally deleting MRC1. We thus propose that the Rtt101Mms22 E3 ubiquitin ligase promotes HR-mediated repair and restart by counteracting a replicative function of Mrc1 at stalled replication forks. Surprisingly, this role of Mrc1 is not shared with the other subunits of the replisome progression complex Tof1 and Csm3, although they are thought to help stabilize Mrc1 at replication forks. However, Mrc1 also interacts with replisomes by a Tof1/Csm3-independent mechanism [47], perhaps through its interactions with Pol2 and Mcm6. It seems that this Mrc1 pool is sufficient to inhibit HR and may need to be counteracted by the Rtt101Mms22 E3 ligase upon fork stalling (Fig 7E).
In contrast to the S-phase checkpoint defective Mrc1-AQ mutant (Fig 4B), expression of a C-terminal truncation mutant of Mrc1 (Mrc11-971) was able to suppress phenotypes linked to both SCFDia2 and Rtt101Mms22 E3 ligases ([24], Fig 4D). Importantly, we identified a Mrc1 separation-of-function mutant (Mrc11-1078), which is checkpoint-proficient but likely unable to perform the replicative function of Mrc1 that leads to toxicity when the Rtt101Mms22 E3 ligase is impaired. While the exact mechanism underlying this specific defect remains to be elucidated, we propose that the C-terminus of Mrc1 may directly regulate replisome function by binding to either Pol2 or Mcm6. This model might also help explain the higher rate of HR (Fig 5A), as mrc1Δ strains leave more unreplicated single stranded DNA stretches at stalled replication forks [11] that require post-replicative HR.
Several mechanisms allow cells to restore stalled replication forks, underlining the importance of this process (reviewed in [19]). Based on our genetic and proteomic analysis, we propose that Rtt101Mms22, presumably via a ubiquitylation event, counteracts a replicative function of Mrc1 in order to promote recombination at stalled DNA replication forks (Fig 7E). Since both the mcm6-IL allele as well as the deletion of MRC1 may affect polymerase and helicase activities (Fig 7A), the Rtt101Mms22 complex may somehow regulate the activity of these factors at stalled forks. Indeed, our genetic analysis revealed that the requirement of the Rtt101Mms22 complex to inhibit the replicative-function of Mrc1 upon genotoxic stress is epistatic to loss of its binding to the MCM helicase, suggesting that the Rtt101Mms22 E3 ligase may directly or indirectly modulate the interaction of Mrc1 with the MCM complex. Although we did not observe a difference in the total amount of Mrc1 associated with chromatin following exposure to MMS (S11 Fig), it remains possible that regulation specifically occurs at a small subset of stalled replication forks. Interestingly, a previous study has reported decreased association of Pol ε and Mrc1 with replication forks in mms1Δ mutants [48], which may represent a compensatory response. Thus, while we cannot rigorously exclude that Rtt101Mms22 regulates Mrc1 via replisome association, we favor a model by which Rtt101Mms22-dependent ubiquitination of Mrc1 or an unknown substrate leads to uncoupling of the MCM helicase at the stalled replicon, thereby promoting HR-dependent repair and restart of stalled replication forks (Fig 7E). However, we do not fully understand how Rtt101 or Mms22 interact with either the error prone TLS branch of bypass synthesis or with other means of replication fork restart (e.g. BIR and replication fork regression) [19]. Since Rad52 is synthetically-sick with Mms22 but not Rtt101 in unchallenged conditions (Fig 5B), it is conceivable that Mms22 may regulate either TLS or replication fork regression, as part of its Rtt101-independent functions. Future studies will be required to test these possibilities.
Plasmids and yeast strains are listed in S3 and S4 Tables, respectively. Standard methods were used for yeast strain construction and molecular biology. Yeast cells were grown in rich medium (YPD; 1% yeast extract, 2% peptone, 2% glucose) or synthetic medium (SD; 0.17% yeast nitrogen base, 0.5% ammonium sulphate, 2% glucose, amino acids as required). Homologous recombination frequencies were measured as described [41]. For spotting assays, the indicated strains were grown overnight at 30°C, and the cultures were diluted to OD600 0.5. Ten-fold serial dilutions were spotted using a pinning head (2 μl). The plates were incubated at 30°C and imaged using the ChemiDoc Touch Imaging System (Bio-Rad) after 2 and 3 days. For cell cycle synchronization, logarithmically growing cells were treated with 1:1000 α-factor solution (5 mg/ml + 0.1% BSA) at 24°C for 3 hours. G1 arrest was monitored by flow cytometry and microscopy (appearance of pear-shaped “shmoo” morphology of at least 95% of cells). Cells were then washed three times with YPD at room temperature and S-phase samples were collected 30 minutes after the release into fresh YPD medium.
Synthetic Genetic Array (SGA) methodology was used as described [30], with the following modifications: the non-essential heterozygous diploid S. cerevisiae knockout collection (kindly provided by M. Knop) was sporulated and crossed to a rtt101::NAT can1::STE2pr-SpHis5 strain (Y7092, C. Boone). Diploids were selected by repinning on YPD plates containing 100 μg/ml nourseothricin and 250 μg/ml of the kanamycin analogue G418. After sporulation, haploid double mutants were selected by repinning on MATa selection plates (SD-his/arg/lys + canavanine + thiolysine) followed by a repinning on MATa selection plates containing 100 μg/ml nourseothricin and 250 μg/ml G418. Colonies were then re-pinned onto SD complete, SD + 0.01% MMS and SD + 5 μM CPT, and repinned twice onto the same media after 24 h incubation at 30°C. Pictures of the last repinning were taken after 24 h incubation at 30°C. The occurrence of suppressors, i.e. double mutants that showed increased resistance to either MMS or CPT, were scored manually, and validated by tetrad analysis from independent starter strains followed by duplicate spotting assays onto drug containing media.
Culture volumes of exponentially growing cells corresponding to 0.68 OD units were collected by centrifugation (3000 rpm for 5 min at RT), resuspended in 1 ml cold 70% ethanol and stored at 4°C. Cells were washed once in 1 ml H2O (3000 rpm for 5 min at RT), resuspended in 0.5 ml 50 mM Tris-HCl (pH 8.0) and incubated with 10 μl RNase A (10 mg/ml) for 3 h at 37°C. After centrifugation (3000 rpm for 5 min at RT) cells were resuspended in 0.5 ml 50 mM Tris-HCl (pH 7.5) containing 1 mg/ml Proteinase K and incubated for 45 min at 50°C. Cells were spun down (3000 rpm for 5 min at RT) and resuspended in 0.5 ml 50 mM Tris-HCl (pH 7.5). 100 μl of cells were sonicated five times 15 sec at low intensity using the Bioruptor Twin XD10. 50 μl of cells were mixed with 1 ml 1 x SYTOX Green (Life Technologies) in 50 mM Tris-HCl (pH 7.5) to stain DNA. Cells were kept dark and analyzed immediately for DNA content using a BD FACSCanto II flow cytometer using the following filters and settings: FSC and SSC were detected with a 488 nm laser with detector settings of 318 V and 360 V, respectively. SYTOX Green was detected with a 502 nm longpass filter and 530/30 nm bandpass filter at 466 V. 20000 events per sample were analyzed in each run. Data collection and analysis was performed using BD FACSDiva software and FlowJo v10.0.6 (Miltenyi Biotec) software.
2 OD600 units of exponentially growing cells were pelleted at 13’000 rpm for 2 min, and if necessary stored at -20°C. Cell pellets were resuspended in 150 μl of Solution 1 (0.97 M 2-mercaptoethanol, 1.8 M NaOH) and incubated on ice for 10 min. 150 μl of Solution 2 (50% TCA) were added, cells were incubated 10 min on ice and centrifuged at 13’000 rpm for 2 min at 4°C. Pellets were resuspended in 1 ml acetone, centrifuged at 13’000 rpm for 2 min at 4°C and the pellets resuspended in 100 μl urea buffer (120 mM Tris-HCl pH 6.8, 5% glycerol, 8 M urea, 143 mM 2-mercaptoethanol, 8% SDS, bromophenol blue indicator). Protein extracts were incubated 5 min at 55°C, centrifuged at 8’000 rpm for 30 sec, separated by SDS-PAGE and transferred onto nitrocellulose. Membranes were blocked with 5% milk and 1% BSA and incubated with appropriate antibodies: Rabbit peroxidase anti-peroxidase (1:10000), mouse monoclonal antibody against c-Myc (1:3000), HA (1:3000), Mcm2 (1:2000), Pgk1 (1:200000), Rad53 (1:16, EL7.E1, gift from M. Foiani), mouse polyclonal antibody against Orc6 (1:500, gift from H. Ulrich). Replisome antibodies are from sheep polyclonal antiserum: Ctf4 (1:2000), Cdc45 (1:1000), Mcm6 (1:1000), Sld5 (1:1000), Psf2 (1:250), Psf3 (1:3000), Csm3 (1:1000), Spt16 (1:3000), Pob3 (1:3000).
Cell harvesting was performed at 3000 rpm (Multifuge 3 5-R) for 3 min at RT. Samples were first washed with 20 mM Tris-acetate pH 9.0, then with lysis buffer (75 mM (or 100 mM in Fig 2C and 2D) Tris-acetate pH 9.0, 50 mM KOAc, 10 mM MgOAc, 2 mM EDTA, 2 mM NaF, 2 mM β-glycerophosphate, 1× Roche protease inhibitor cocktail, 1× sigma inhibitors). The cell pellets’ mass was weighted and re-suspended in 3 volumes of lysis buffer. The cell suspension was shock-frozen in liquid nitrogen as “droplets” and stored at -80°C. All cell manipulations and collection were performed at 4°C, if not specified otherwise. Equal weight of “droplets” was grinded with a cryogenic impact mill (Freezer-mill 6870 Large SamplePrep), using 5 min pre-cool followed by 6 cycles of 2 min milling at 12 CP and 2 min cooling down. Cells were thawed for 5 min at RT and 0.25 volume of glycerol mix (75 mM (or 100 mM in Fig 2C and 2D) Tris-acetate pH 9.0, 300 mM KOAc, 50 mM MgOAc, 2 mM EDTA, 0.5% NP40, 1 mM DTT, 2 mM NaF, 2 mM β-glycerophosphate, 1× Roche protease inhibitors, 1× yeast inhibitors) was added to the lysate. DNA was digested by 800 Units/ml DNA nuclease (Benzonase Novagen) at 4°C for 30 min, followed by 30 min centrifugation at 15’000 rpm (25’000g) at 4°C (Sorvall RC26 Plus, SS-34 rotor) and 60 min ultracentrifugation at 25’000 rpm (100’000g) at 4°C (Beckman Coulter Optima LE80K, SW-41 rotor) to remove insoluble material. From the resulting extract, 50 μl was used as whole cell extract (WCE) and the remaining was used for affinity-precipitation. 50 μl of WCE was dissolved in 100 μl 1.5×SDS-loading buffer (1× buffer: 50 mM Tris-Cl pH 6.8, 100 mM DTT, 2% SDS, 0.1% bromophenol blue, 10% glycerol) and boiled at 95°C for 5 min. 4 μl of sample was used to load on a Bis-Tris acrylamide gel. Washed IgG coupled dynabeads (M-270 Epoxy; 14302D, Life Technologies) were added to extracts for immuno-precipitation. Samples were incubated for 2 hours on a rotating platform at 4°C. Beads were washed 4 times at RT with 1 ml wash buffer (100 mM Tris-acetate pH 9.0, 100 mM potassium acetate, 10 mM magnesium acetate, 2 mM EDTA) and protein was eluted with 50 μl of 1× SDS-loading buffer.
At the indicated time points, cells expressing Rad52-mCherry (0.08 OD600 units) were pelleted at 3’000 rpm for 3 min, resuspended in 300 μl SD-Trp containing 0.03% MMS, and then transferred into one chamber of a Nunc Lab-Tek coverglass (Thermo Fisher Scientific) coated with 2 mg/ml Concanavalin A (Sigma Aldrich). Images were obtained on a Leica AF7000 widefield microscope using a 63x/1.4 oil objective. Brightfield and fluorescent images were taken along the z-axis, and Rad52-mCherry foci counted in all focal planes for at least 400–600 cells per strain (n = 2 biological replicates).
G1-arrested cell cultures were split and released into YPD media containing DMSO solvent or 0.03% MMS. S-phase cells were collected after 30 min (untreated) or 1 hour (MMS-treated) at 30°C, stopped with 0.1% sodium azide and harvested at RT by centrifugation for 5 min at 4000 rpm. The pellet was resuspended in 1.5 ml pre-spheroplasting buffer (100 mM PIPES, pH 9.4, 10 mM DTT, 0.1% sodium azide), pelleted again after 10 min at room temperature, and resuspended in 1 ml spheroplasting buffer (50 mM potassium phosphate buffer, pH 7.5, 0.6 M sorbitol, 10 mM DTT, 0.2 mg/ml zymolyase (>200 units/mg)). After 1 h incubation at 30°C, spheroplasts were spun down at 4°C for 1 min at 2500 rpm, washed with 1 ml wash buffer (100 mM KCl, 50 mM HEPES-KOH, pH 7.5, 2.5 mM MgCl2, 0.4 M sorbitol) and resuspended in 100 μl extraction buffer (100 mM KCl, 50 mM HEPES-KOH, pH 7.5, 2.5 mM MgCl2, 1× Roche protease inhibitor cocktail). The suspension was split into three aliquots of 50 μl each (whole cell extract, soluble fraction, chromatin bound fraction), cells lysed by adding 0.25% Triton X-100 and 5 min incubation on ice, and the cell extract treated with 1 μl of a 1:50 dilution of benzonase (NEB). After 15 min incubation on ice, NuPAGE LDS sample buffer was added, the soluble fraction centrifuged at 4°C for 10 min at 14000 rpm, and the supernatant transferred to a new reaction tube. The chromatin bound fraction was underlayed with 30% sucrose solution and centrifuged at 4°C for 10 min at 14000 rpm. The supernatant was discarded and the pellet resuspended in 50 μl extraction buffer with 0.25% Triton X-100. This was repeated once, the resuspended final pellet treated at 4°C with 1 μl of a 1:50 dilution of benzonase, and the reaction stopped after 15 min by the addition NuPAGE LDS sample buffer. All samples were incubated for 10 min at 70°C, cleared by centrifugation for 10 min at 13000 rpm, and the samples analyzed by immunoblotting using 4–15% pre-cast polyacrylamide gels (Bio-Rad).
|
10.1371/journal.pgen.1004044 | A Nonsense Mutation in TMEM95 Encoding a Nondescript Transmembrane Protein Causes Idiopathic Male Subfertility in Cattle | Genetic variants underlying reduced male reproductive performance have been identified in humans and model organisms, most of them compromising semen quality. Occasionally, male fertility is severely compromised although semen analysis remains without any apparent pathological findings (i.e., idiopathic subfertility). Artificial insemination (AI) in most cattle populations requires close examination of all ejaculates before insemination. Although anomalous ejaculates are rejected, insemination success varies considerably among AI bulls. In an attempt to identify genetic causes of such variation, we undertook a genome-wide association study (GWAS). Imputed genotypes of 652,856 SNPs were available for 7962 AI bulls of the Fleckvieh (FV) population. Male reproductive ability (MRA) was assessed based on 15.3 million artificial inseminations. The GWAS uncovered a strong association signal on bovine chromosome 19 (P = 4.08×10−59). Subsequent autozygosity mapping revealed a common 1386 kb segment of extended homozygosity in 40 bulls with exceptionally poor reproductive performance. Only 1.7% of 35,671 inseminations with semen samples of those bulls were successful. None of the bulls with normal reproductive performance was homozygous, indicating recessive inheritance. Exploiting whole-genome re-sequencing data of 43 animals revealed a candidate causal nonsense mutation (rs378652941, c.483C>A, p.Cys161X) in the transmembrane protein 95 encoding gene TMEM95 which was subsequently validated in 1990 AI bulls. Immunohistochemical investigations evidenced that TMEM95 is located at the surface of spermatozoa of fertile animals whereas it is absent in spermatozoa of subfertile animals. These findings imply that integrity of TMEM95 is required for an undisturbed fertilisation. Our results demonstrate that deficiency of TMEM95 severely compromises male reproductive performance in cattle and reveal for the first time a phenotypic effect associated with genomic variation in TMEM95.
| Impaired male fertility is a prevalent condition in many species and is often explained by aberrant semen quality. In some cases, male fertility is severely compromised although semen quality is without any apparent pathological findings (i.e., idiopathic male subfertility). The genetic mechanisms underlying idiopathic male subfertility often remain unexplained. In the present paper, we report a recessively inherited variant of idiopathic male subfertility in a cattle population. We use 650,000 genome-wide SNP markers genotyped in >7900 artificial insemination bulls to pinpoint the underlying genomic region. We take advantage of whole-genome re-sequencing data of 43 animals to identify a causal loss-of-function mutation in TMEM95 encoding a nondescript transmembrane protein. We demonstrate that transmembrane protein 95 is located at the plasma membrane of spermatozoa of fertile animals whereas it is absent in spermatozoa of subfertile animals. Our results indicate that integrity of transmembrane protein 95 is required for an undisturbed fertilisation. This is the first report to reveal a phenotypic effect associated with genomic variation in TMEM95 in any organism.
| Impaired reproductive performance is a prevalent condition in both sexes of many species and up to 15% of couples are affected in humans [1], [2]. The disability to reproduce is defined as infertility (i.e., sterility), whereas subfertility refers to any form of reduced fertility [3].
Low sperm concentration (i.e., oligospermia) and the absence of spermatozoa (i.e., azoospermia), respectively, are frequently diagnosed in males with impaired fertility [4]. Further aberrant semen quality traits (e.g., abnormal sperm morphology [5], reduced motility [6], [7]) account for another substantial fraction of reduced male fertility. However, semen analysis of a considerable number of males with impaired reproductive performance remains without any apparent pathological findings (i.e., unexplained/idiopathic infertility) [8], [9].
Semen quality traits have low to medium heritability in cattle populations [10]. Numerous genetic variants underlying routinely assessed semen quality traits have been identified so far in humans [11], [12], model species [13] and livestock populations [14]. However, the number of known genetic mechanisms causing idiopathic male subfertility is very small [15], [16] and identified polymorphisms explain only a small fraction of its genetic variation [17].
Artificial insemination (AI) is predominant over natural service in most cattle populations and all ejaculates are closely examined immediately after semen collection. Only semen samples without any apparent abnormalities, such as low sperm count, reduced progressive motility, low viability, abnormal morphology of spermatozoa, are used for insemination. However, the reproductive performance indicated by the proportion of successful inseminations varies considerably among AI sires [18], [19]. So far, genome-wide association studies (GWAS) for male reproductive traits were of limited success in cattle populations [20], [21] and only one putatively causative mutation has been identified [22].
Here we report a new recessively inherited variant of idiopathic male subfertility in the Fleckvieh (FV) cattle population. The mapping of the underlying genomic region was facilitated by using high-density genotypes in a large sample of artificial insemination bulls with phenotypes for reproductive performance assessed based on 15 million artificial inseminations. Exploiting whole-genome re-sequencing data revealed a causative loss-of-function mutation in the transmembrane protein 95 encoding gene TMEM95.
Phenotypes for male reproductive ability (MRA) were obtained for 7962 bulls of the FV population based on 15.3 Mio artificial inseminations (AI). The values for MRA range from −40 to +13 and reflect the bulls' reproductive performance as percentage deviation from the population mean. Male reproductive ability is highly correlated (r = 0.59) with the 56-day non-return rate (NRR56) in cows. The NRR56 is the proportion of cows that are not re-inseminated within a 56-day interval after the first insemination. After visual inspection of the distribution of MRA, forty-nine bulls with exceptionally poor reproductive performance (MRA<−20) were considered as subfertile (Figure S1 and Table 1). Animals with values for MRA below −20 ( = five standard deviations below the population mean) were used as case group in a case-control design.
Using MRA as quantitative trait in a genome-wide association study (GWAS) yielded a strong association signal on bovine chromosome (BTA) 19 (P = 4.38×10−20, Figure S2). However, the association signal was more pronounced using 49 subfertile animals (MRA<−20) as case group and the remaining 7913 animals as controls (Table 1 and Figure 1A). The most significantly associated SNP is located at 30,220,186 bp (ARS-BFGL-NGS-11488; P = 4.08×10−59).
Autozygosity mapping revealed a common 1386 kb segment (26,580,096 bp–27,956,634 bp) of extended homozygosity in 40 subfertile bulls containing 80 genes (Figure 1B and Table S1). None of 7913 bulls with normal reproductive performance was homozygous for the 1386 kb segment, indicating recessive inheritance. Semen samples of 40 homozygous bulls had been used for 35,671 artificial inseminations with an average of 892 inseminations per bull. This is a typical number for test inseminations performed with semen samples of young bulls in progeny testing based breeding programmes. However, only 619 (1.74%) of those inseminations were successful (Table S2).
There was no evidence for the presence of large structural variants (i.e., copy number variations) within the segment of extended homozygosity (Figure S3). The proportion of missing genotypes did not significantly differ between cases and controls (P>0.09) for all SNPs located within the associated region.
The frequency of the subfertility-associated haplotype amounts to 7.2%. Of 7962 genotyped bulls with phenotypes for MRA, 1068 (13.41%) carry the deleterious haplotype in heterozygous state. The carrier frequency increased considerably within the last years (P = 0.0002, Figure S4). The reproductive performance of heterozygous bulls is normal, indicating recessive inheritance (Figure 2). Of 1952 primiparous cows, 291 are heterozygous and 17 are homozygous for the subfertility-associated haplotype. The haplotype neither affects reproductive performance nor milk production traits in females (Table S3). The haplotype distribution does not deviate from the Hardy-Weinberg equilibrium, neither in females (P = 0.303) nor in males (P = 0.817).
Both, haplotype and pedigree analysis allowed to trace the mutation back to the bull HAXL (*1966) (Figure S5). HAXL appears in the pedigrees of 7779 out of 7962 bulls (97.70%) and can be considered as the most important ancestor of the current FV population [23].
Whole genome re-sequencing of 43 animals and subsequent multi-sample variant calling yielded genotypes at 17.17 million sites [23]. Among them, 5965 (5287 SNPs and 678 INDELs) are located within the subfertility-associated region on BTA19 (26,580,096 bp to 27,956,634 bp). Six of the 43 re-sequenced animals were identified as carriers of the associated haplotype via high-density genotypes. The sequence data were filtered for variants compatible with the supposed recessive inheritance, i.e., heterozygous in carriers and homozygous for the reference allele in non-carriers (see Material & Methods, Figure S6). After filtering, 26 SNPs and six INDELs were retained as candidate causal mutations (Table S4 and S5). The functional effects of those variants were predicted based on the gene annotation of the UMD3.1 assembly of the bovine genome [24]. Four of the 32 compatible variants were located in coding regions (Table 2). Among them, we considered a nonsense mutation in TMEM95 (rs378652941, c.483C>A, p.Cys161X, Chr19: 27,689,622 bp) as the prime candidate causal mutation (Figure 3A and 3B). The nonsense mutation was subsequently confirmed in the re-sequenced animals by classical Sanger sequencing (Figure S7 and S8).
Genotypes for two non-synonymous substitutions in ACDVL and KIF1C and for the nonsense mutation in TMEM95 were obtained for cases and controls using TaqMan genotyping assays (Table 3). Only c.483C>A, introducing the premature stop-codon in TMEM95 (p.Cys161X), was perfectly associated. All animals, which are homozygous for the subfertility-associated haplotype, are homozygous for the non-reference allele, whereas none of 1396 FV bulls with normal reproductive performance are homozygous. The polymorphism is present in the FV breed only; 277 Holstein-Friesian and 278 Braunvieh animals are homozygous for the reference allele. The c.483C>A-mutation is not segregating among 15 Jersey, 47 Angus and 129 Holstein-Friesian animals which were sequenced in the context of the 1000 bull genomes project [25].
TMEM95 encodes a highly conserved single-pass type I transmembrane protein consisting of 183 amino acids with a predicted extracellular N-terminal signal peptide, a 23-amino acid transmembrane domain (amino acid position 153 to 175) and a 8-amino acid intracellular C-terminal domain (Figure 3C and Figure S9 and S10). The premature stop codon introduced by the c.483C>A-mutation is located within the predicted transmembrane domain and truncates the protein by 22 amino acids.
Semen quality (morphology, vitality, total motility) was analysed using cryopreserved semen samples of 30 bulls (10 wt/wt, 10 wt/mt, 10 mt/mt). In all ejaculates, spermatozoa showed less than 20% morphological alterations and less than 5% morphological alterations of the head. Total motility after thawing ranged from 50 to 65%. Statistical analysis showed no significant differences in the proportion of motile spermatozoa from wt/wt, wt/mt and mt/mt bulls (Table 4). As shown by eosin staining, 40 to 70% of the spermatozoa were viable after thawing. There were no significant differences in the percentages of viable spermatozoa between wt/wt, wt/mt and mt/mt bulls. Additionally, ejaculate volume, sperm concentration and progressive motility were assessed in fresh semen samples of 203 AI bulls (177 wt/wt, 21wt/mt, 5 mt/mt). Ejaculate volume was above 5 ml, sperm count was above 1.42 Mio/µl and the proportion of spermatozoa with progressive motility was above 70% for all animals (Table 5).
A mouse-derived polyclonal antibody generated against human transmembrane protein 95 was used to locate its position in spermatozoa of 33 bulls (10 wt/wt, 10 wt/mt and 13 mt/mt). In spermatozoa of wt/wt bulls, TMEM95 was distinctly located on the plasma membrane of the acrosome (Figure 4A). Staining was also visible on the equatorial segment of the head. The sperm neck was regularly labelled. Spermatozoa of wt/mt and wt/wt bulls showed an identical staining pattern (Figure 4B), whereas spermatozoa of mt/mt bulls did not show any staining on the head (Figure 4C). Weak fluorescence was detected in the midpiece of the tail in spermatozoa of all animals due to the autofluorescence of the mitochondria. In the negative controls, there was no signal detectable on the sperm head whereas the midpiece of the tail showed weak autofluorescence (Figure S11).
The genome-wide association study (GWAS) with imputed genotypes for 7962 artificial insemination bulls identified a genomic region on BTA19 for male reproductive ability (MRA) in the FV population. Autozygosity mapping revealed a common 1386 kb segment of extended homozygosity in 40 bulls with unexplained exceptionally poor reproductive performance. None of the bulls with normal reproductive performance was homozygous indicating recessive inheritance. Only 1.74% of inseminations performed with semen samples of affected bulls were successful, although semen quality parameters were within a normal range, reflecting idiopathic subfertility [26]. The newly identified congenital defect is denominated as “Bovine Male Subfertility” and accounts for 82% of FV bulls with exceptionally poor reproductive performance. However, we cannot exclude the possibility that homozygous males are infertile and that the very low proportion of successful inseminations reflects errors in parentage recording which might be as high as 10% in dairy cattle breeding programmes [27]. In progeny testing based breeding programmes, semen doses of young bulls are used for approximately 1000 test inseminations [28]. These artificial inseminations are performed within very short time, precluding the early identification of subfertile/infertile bulls. Identifying and culling bulls with poor fertility prognosis (i.e., homozygous bulls) before they are used for artificial insemination is now possible. There was no evidence for any additional genomic region underlying idiopathic male subfertility in the FV population, although the reproductive ability of nine bulls which are not homozygous for the c.483C>A-mutation, is very low. However, the number (n = 9) of subfertile bulls not attributable to the BTA19 locus might not be sufficient for detecting additional loci (Figure S12 and Table S6).
The potential of targeted or whole genome re-sequencing for the identification of causal trait variants has been demonstrated in several species (e.g., [29]–[31]) including cattle [32]–[34]. Causal trait variants for monogenic disorders are traditionally identified by sequencing case/control-panels and by subsequently comparing allele counts in affected and unaffected individuals. However, the concept of the present study is different: the identification of the underlying mutation was based on whole genome re-sequencing data of 43 unaffected FV animals explaining a vast majority of the population's genomic variation [23]. As the frequency of the mutation was reasonably high (7.2%), the affected haplotype was present in heterozygous state in six of the re-sequenced animals. Filtering the re-sequencing data for variants compatible with the supposed recessive inheritance pattern revealed a plausible candidate causative loss-of-function mutation in TMEM95 encoding the transmembrane protein 95.
The nonsense mutation was perfectly associated in 1990 animals representing three different breeds. To our knowledge, this is the first report of a phenotypic effect associated with variation in TMEM95 in any organism. So far, there are no clues about the precise function of TMEM95. However, it seems likely that TMEM95 is involved in sperm-egg interactions, which has been shown to be the main function of sperm-specific transmembrane proteins (e.g., [35], [36]). The phenotype in the present study resembles phenotypic patterns of Caenorhabditis elegans resulting from an impaired function of sperm-specific transmembrane proteins [37], [38]. Taken together, our findings evidence genomic variation within TMEM95 to severely compromise the reproductive performance in cattle.
The causative polymorphism (c.483C>A, rs378652941) introduces a premature stop codon in TMEM95 (p.Cys161X). The affected codon resides within the predicted transmembrane domain of TMEM95 most likely resulting in a disturbed anchorage of the truncated protein in the sperm plasma membrane bilayer. It is also likely that the resulting truncated protein is absent due to nonsense-mediated mRNA decay [39]. Our data show no evidence that the mutation affects any of the routinely assessed semen quality parameters in vitro. However, we cannot exclude the possibility that the mutation affects semen quality parameters, e.g., vitality and motility, in vivo [40], [41].
Transmembrane protein 95 is primarily located on the acrosomal membrane of the sperm head indicating that it may be involved in the acrosome reaction. Spermatozoa of mt/mt animals showed no fluorescence at the acrosomal membrane implying deficiency of TMEM95. Thus, successful fertilization by spermatozoa of mt/mt animals might be compromised. This is supported by the fact that the equatorial segment of the acrosome, which provides the first contact of the spermatozoon with the cell membrane of the oocyte [42], is also labelled in spermatozoa of fertile animals. The sperm neck contains the centriole and is essential for cell division and development of the early embryo [43]. Labelling of the neck indicates an additional potential role of TMEM95 after fertilization during the first cell divisions of the early embryo.
Although spermatozoa of mt/mt animals showed fluorescence, neither on the acrosomal membrane and the equatorial segment nor at the centriole, weak unspecific fluorescence was observed at the midpiece of the tail. This fluorescence pattern is also present in spermatozoa of wt/wt and wt/mt animals. The midpiece is the only region of spermatozoa that contains mitochondria [44]. The weak fluorescence of the midpiece is due to unspecific autofluorescence of the mitochondria, which has been described in several organs and species [45]–[47].
Male subfertility is also present in other species besides cattle [9], [48]–[50], and compromised sperm surface proteins account for a substantial number of males with distinctly reduced reproductive ability in humans [15], [40]. Our results demonstrate that TMEM95 is another sperm surface protein, which is likely to be involved in sperm-egg plasma membrane interactions. Its protein sequence is highly conserved among species (Figure S10) and genetic variants disrupting TMEM95 are likely to induce male subfertility also in other species than cattle. Numerous polymorphic sites have been identified in human TMEM95, among them several potential loss-of-function variants (Figure S13). Based on our findings it is highly recommended to systematically survey variants in TMEM95 as potentially causal for idiopathic male in- or subfertility in any species.
Frequencies of variants that disrupt protein-coding genes are usually low in human populations [51], [52]. However, in livestock populations, the frequency of deleterious alleles might increase rapidly because individual sires can generate tens of thousands of progeny by artificial insemination (e.g., [53], [54], [32]). The loss-of-function mutation in TMEM95 can be traced back to HAXL (*1966), the most important ancestor of the current FV population. Within eight generations, the frequency of the deleterious allele increased to 8.9% and 1443 (13.92%) animals of the present study are heterozygous. This increase of the allele frequency had been possible because phenotypic effects become apparent in homozygous males only. There are no phenotypic effects detectable neither in heterozygous nor in homozygous females (Table S3).
In agreement with previous findings in livestock [55] and humans [15], our results evidence that standard assessment of spermatozoa (i.e., morphology, motility and vitality) is not sufficient to reliably anticipate male reproductive performance. All routinely assessed semen parameters of bulls homozygous for the nonsense mutation in TMEM95 comply with current requirements for artificial insemination in cattle [56]. It might be advisable to develop functional assays, e.g., for the integrity of sperm surface proteins, for an efficient prospective monitoring of male fertility.
Semen samples were collected by approved commercial artificial insemination stations as part of their regular breeding and reproduction measures in cattle industry. No ethical approval was required for this study.
Male reproductive ability (MRA) was evaluated in 7962 AI bulls of the German FV population. Semen samples of those bulls were used for 15,321,171 artificial inseminations with an average of 1924 artificial inseminations per bull. The phenotypes for MRA were obtained from routine breeding value estimation for reproductive traits, which are jointly estimated for males and females [57]. The resulting phenotypes for MRA represent the bulls' reproductive performance adjusted for environmental and genetic effects (i.e., year, season, flock, female mating partner). The lower the value for MRA, the worse is the bull's reproductive performance (i.e., the smaller the proportion of successful inseminations).
A total of 3545 animals (1475 AI bulls, 2070 primiparous cows) of the FV population were genotyped with the Illumina BovineHD Bead chip comprising 777,962 SNPs. Another 7073 AI bulls were genotyped with the BovineSNP50 Bead chip comprising ∼54,000 SNPs. The chromosomal position of the SNPs was determined according to the UMD3.1 assembly of the bovine genome [58]. Mitochondrial, Y-chromosomal and those SNPs with unknown chromosomal position were not considered for further analyses. Stringent quality control was carried out for each dataset separately using PLINK v1.07 [59]. Animals and SNPs with call-rate <0.95 were excluded, as well as SNPs with minor allele frequency <0.5% and those SNPs deviating significantly from the Hardy-Weinberg equilibrium (P<10−6). The pairwise genomic relationship [60] was compared with the pedigree relationship tracing pedigree records back to 1920 [61]. Animals showing major discrepancies of the pedigree and the genomic relationship were removed from the dataset, as such patterns indicate sample swaps. After quality control, the high-density dataset contained 3332 animals and 652,856 SNPs with an average per-individual call-rate of 99.17%. The medium-density dataset contained 7031 animals and 42,907 SNPs with an average per-individual call-rate of 99.75%. Genotype imputation was performed to extrapolate medium-density genotypes to higher density using a pre-phasing based approach. Haplotypes were inferred for the two datasets separately using Beagle [62] and subsequent haplotype-based imputation was performed with Minimac [63]. This approach yields high imputation accuracy in cattle [64]. The imputed dataset comprised 10,363 animals (8411 AI bulls/1952 primiparous cows) and genotypes for 652,856 SNPs. Phenotypic records for MRA were available for 7962 bulls.
Genome-wide association studies were performed applying a variance component based approach to account for population stratification. We used EMMAX [65] to fit the mixed model , where Y is a vector of phenotypes, b is the SNP effect, X is a design matrix of imputed SNP genotypes, u is a vector of additive genetic effects assumed to be normally distributed with mean 0 and (co)variance , with being the additive genetic variance and G being the realized genomic relationship matrix (GRM) of the 7962 bulls with phenotype information built based on 635,224 autosomal SNPs (see above) and where e is a vector of random normal deviates .
The genomes of 42 key and contemporary animals of the FV population were sequenced at low- to medium coverage (ø 7.4-fold) and one animal was sequenced at high coverage (25-fold) using Illumina GA IIx and HiSeq 2000 instruments [66], [23]. Paired-end reads were obtained and mapped to the bovine reference sequence (UMD3.1 [58]) using the Burrows-Wheeler Aligner (BWA) [67]. PICARD (http://picard.sourceforge.net) was used to mark PCR-duplicates. Subsequent multi-sample variant calling with mpileup [68] yielded genotypes at 17.17 million sites. The re-sequencing data were contributed to the 1000 bull genomes project [25] and all variants were submitted to dbSNP [23]. Beagle phasing and imputation within the 43 sequenced animals improved the primary genotype calls (a detailed overview of the entire variant calling pipeline and all obtained variants is presented in Jansen et al. [23]). Of 17.17 million sites, 5287 SNPs and 678 INDELs were located within the 1386 kb segment (26,580,096 bp to 27,956,634 bp) of extended homozygosity on BTA19. Of the 43 sequenced animals, six were identified as carriers of the subfertility-associated haplotype via high-density genotypes, among them the animal sequenced at high coverage. Assuming perfect correlation between the subfertility-associated haplotype and the causal mutation, the allele frequency of the causal mutation should be 7% (6 of 86 affected alleles) in the sequence-derived genotypes. To account for inaccurately genotyped variants due to the low-coverage sequence data (e.g., mis-calling of heterozygous genotypes for rare variants [69]) [70] and for possible phasing errors, a conservative mutation scan was performed to identify variants compatible with recessive inheritance (Figure S6). The 5965 polymorphic sites were filtered for variants that met three conditions: (i) the frequency of the non-reference allele is below 10%, (ii) the variant is heterozygous in the animal sequenced at high coverage and (iii) the variant is present in heterozygous state in at least three of five carrier sequenced at low coverage.
PCR primers (5′-CACCCTGCCTTGTCTTTCAT-3′ and 5′-AGGCTCTGTCCTCGTTTTCA-3′) were designed for exon 6 of TMEM95 to scrutinize the rs378652941-polymorphism by classical Sanger sequencing in the re-sequenced animals as recommended by Jansen et al. [23]. Genomic PCR products were sequenced using the BigDye Terminator v1.1 Cycle Sequencing Kit (Life Technologies) on the ABI 3130x1 Genetic Analyzer (Life Technologies). Genotypes for rs378652941 (TMEM95:c.483C>A, p.Cys161X, Chr19:27689622), rs381722524 (KIF1C:c.197A>G, p.Gln66Arg, Chr19:27042848) and rs385135118 (ACADVL:c.706C>A, p.Pro236Thr, Chr19:27570146) were obtained by TaqMan genotyping assays (Life Technologies) in 1990, 1222 and 1206 animals, respectively, representing three different breeds (BV, FV, HF). The primer and probe sequences are listed in Table S7.
The topology of bovine transmembrane protein 95 (NCBI reference sequence: XP_002695846.1) was predicted with SPOCTOPUS [71] and PHOBIUS [72]. Both methods predict the transmembrane protein topology while accounting for the existence of a N-terminal signal peptide sequence. The protein topology was visualized with SOSUI [73]. ClustalW [74] was used for multi-species alignment of the protein sequence and for the prediction of conserved regions.
Cryopreserved (−196°C) sperm specimens of 30 bulls (10 wt/wt, 10 wt/mt, 10 mt/mt) were obtained from artificial insemination companies. Two different ejaculates of each bull were evaluated with two straws pooled per ejaculate. After thawing (37°C, 30 s), sperm morphology was assessed by staining with Diff-Quik (Siemens Healthcare, Germany). Sperm total motility was investigated immediately after thawing by phase-contrast microscopy using the Leica DM 1000 microscope (Leica, Germany). Viability of spermatozoa was investigated after thawing by mixing 5 µl of the thawed ejaculate with aqueous Eosin Y solution (Sigma Aldrich, Germany) in a volume ratio of 1∶1. Intact and viable spermatozoa stay colourless whereas spermatozoa with disturbed membrane integrity stain red. Counting of viable sperm was done within 10 seconds after mixing. Two hundred spermatozoa in at least two different fields of view were investigated at a magnification of 400× to analyse morphology, viability and motility.
Fresh semen traits (ejaculate volume, sperm count and progressive motility) of 203 AI bulls (177 wt/wt, 21wt/mt, 5 mt/mt) were provided by an artificial insemination company. Semen quality was analysed based on 10,682 ejaculates with an average of 52.6 ejaculates per bull. At semen collection, the age of the bulls ranged from 1.1 to 2.5 years.
Immunohistochemistry on cryopreserved sperm specimens of 33 bulls (10 wt/wt, 10 wt/mt,13 mt/mt) was repeated for 3 times. After thawing (37°C, 30 s), spermatozoa were washed in phosphate buffered saline (PBS) twice and were diluted in PBS to a concentration of 500,000 spermatozoa/ml. Thereafter, drops of 7 µl were placed on 3-aminopropyl-ethoxysilane-coated slides and dried on a heating-plate at a temperature of 38°C. Subsequently, the slides were fixed in Bouin's solution for 7 min and washed in PBS twice. Non-specific binding was blocked by incubation in blocking buffer (0.1% bovine serum Albumin in PBS, Sigma-Aldrich, Germany) for 5 minutes and in normal goat serum (dilution 1∶5 in PBS, Invitrogen, Germany). Next, the spermatozoa were incubated with the first antibody Yomics Ab989 (mouse-derived against human TMEM95, Primm, USA) in a dilution of 1∶200 in blocking buffer at 4°C overnight. The secondary antibody was the Fluorescein (FITC)-conjugated AffiniPure Goat anti Mouse IgG (H+L) (Dianova, Germany, dilution 1∶200). Negative controls were done by a) replacing the first antibody with PBS and b) using a non-relevant anti-mouse antibody directed against Villin (1∶75, Beckman Coulter). Specimens were evaluated by using a confocal laser scanning microscope (Leica DM IRBE) in magnifications from 400 to 800.
|
10.1371/journal.pgen.1004085 | Metabolic QTL Analysis Links Chloroquine Resistance in Plasmodium falciparum to Impaired Hemoglobin Catabolism | Drug resistant strains of the malaria parasite, Plasmodium falciparum, have rendered chloroquine ineffective throughout much of the world. In parts of Africa and Asia, the coordinated shift from chloroquine to other drugs has resulted in the near disappearance of chloroquine-resistant (CQR) parasites from the population. Currently, there is no molecular explanation for this phenomenon. Herein, we employ metabolic quantitative trait locus mapping (mQTL) to analyze progeny from a genetic cross between chloroquine-susceptible (CQS) and CQR parasites. We identify a family of hemoglobin-derived peptides that are elevated in CQR parasites and show that peptide accumulation, drug resistance, and reduced parasite fitness are all linked in vitro to CQR alleles of the P. falciparum chloroquine resistance transporter (pfcrt). These findings suggest that CQR parasites are less fit because mutations in pfcrt interfere with hemoglobin digestion by the parasite. Moreover, our findings may provide a molecular explanation for the reemergence of CQS parasites in wild populations.
| Chloroquine was formerly a front line drug in the treatment of malaria. However, drug resistant strains of the malaria parasite have made this drug ineffective in many malaria endemic regions. Surprisingly, the discontinuation of chloroquine therapy has led to the reappearance of drug-sensitive parasites. In this study, we use metabolite quantitative trait locus analysis, parasite genetics, and peptidomics to demonstrate that chloroquine resistance is inherently linked to a defect in the parasite's ability to digest hemoglobin, which is an essential metabolic activity for malaria parasites. This metabolic impairment makes it harder for the drug-resistant parasites to reproduce than genetically-equivalent drug-sensitive parasites, and thus favors selection for drug-sensitive lines when parasites are in direct competition. Given these results, we attribute the re-emergence of chloroquine sensitive parasites in the wild to more efficient hemoglobin digestion.
| Drug resistance is a critical issue facing worldwide malaria control. The spread and persistence of chloroquine-resistant (CQR) Plasmodium falciparum has rendered chloroquine, an inexpensive and potent drug, ineffective throughout most of the world [1]. In sub-Saharan Africa [2] and the island of Hainan (China) [3], where CQR parasites formerly accounted for 85–98% of the population, the coordinated cessation of chloroquine treatment resulted in a dramatic reduction (40–100%) in the prevalence of CQR parasites over 10 years. Although the reemergence of chloroquine sensitive (CQS) parasites is a major development with regard to human health, the underlying molecular mechanisms behind this phenomenon are unknown.
The importance of drug resistance to world health has prompted a half century of intensive research into the parasite's mechanisms of resistance [4]. These efforts identified the predominant gene responsible for chloroquine resistance, the P. falciparum chloroquine resistance transporter (pfcrt, pf3D7_0709000) [5]. PfCRT is a multiple pass membrane protein that is localized to the digestive vacuole [5], [6]. Mutations associated with chloroquine resistance have been mapped [7]–[10] and a single polymorphism in the first transmembrane domain (K76T) has been shown to be essential for drug resistance [11]. Recently, the resistant form of PfCRT was found to transport chloroquine under physiologically relevant conditions. Wildtype PfCRT is also assumed to function as a transporter [12], but its native substrates are unclear and the impact of CQR alleles on PfCRT's normal function remains a mystery.
Although mutations in pfcrt are necessary and sufficient to confer chloroquine resistance, several other genes have also been implicated in drug tolerance. The interactions between pfcrt and these other loci, including the P. falciparum multiple drug resistance gene (pfmdr1) [9], [13]–[15] and P. falciparum multiple resistance protein (pfmrp1) [16], are not clearly understood. One possibility is that mutations at secondary loci interact with pfcrt to modulate drug resistance. Alternatively, mutations at other loci may compensate for loss of function associated with CQR forms of PfCRT. Understanding how pfcrt mutations affect parasite physiology is an essential step towards unraveling these polygenic contributions to chloroquine resistance.
Given PfCRT's transmembrane structure, its localization to the digestive vacuole, and its ability to transport chloroquine in vitro, we hypothesized that wildtype PfCRT functions as a transporter. Furthermore, we predicted that mutations in pfcrt, and other CQR-associated genes, alter steady-state metabolite levels in PfCRT-associated pathways. Identifying these phenotypes and linking them to specific genes is difficult because 1) metabolites are often derived from multiple pathways, 2) steady-state levels of compounds can be affected by small perturbations far upstream or downstream of a particular compound and 3) metabolic regulation often involves complex interactions between nonlinear factors, such as covalent modification of enzymes, feedback inhibition, and allosteric regulation.
Quantitative trait locus (QTL) mapping is a powerful tool for unraveling complex metabolic networks and tracing metabolic regulation to specific genes [17], [18]. QTL mapping uses the segregation of alleles across a phenotypically and genetically diverse population to measure the genome-wide contribution of individual alleles to a phenotype (e.g. a metabolite concentration) [19]–[24]. Recently, this approach has been integrated with untargeted metabolomics to study metabolic regulation on a genome-wide scale [17]. This emerging metabolic QTL (mQTL) strategy is of obvious applicability to malariology. However, only three genetic crosses of P. falciparum have ever been completed because of serious logistical challenges [25]. One of these efforts crossed the CQS HB3 parasite clone with the CQR Dd2 clone. The haploid progeny from this cross provide a unique opportunity to investigate the metabolic consequences of drug resistance and the role of compensatory mutations in maintaining metabolic homeostasis.
In this study, we use high resolution mass spectrometry to measure the global metabolic profiles of progeny from the HB3×Dd2 genetic cross. Using mQTL mapping, we identify a family of hemoglobin-derived peptides that accumulate in parasites carrying CQR pfcrt alleles. We show that this phenotype can be recapitulated in transgenic parasite lines in which the native pfcrt gene has been replaced with a recombinant CQR or CQS pfcrt allele. In addition, we show that two independently evolved CQR alleles of pfcrt confer a fitness cost. From these data, we propose that CQR imparts a fitness cost on parasites by disrupting hemoglobin catabolism.
We combined genome-wide QTL mapping with mass spectrometry (MS)-based metabolomics to identify genetic loci in P. falciparum that have a significant influence over steady-state metabolite levels. To achieve this, synchronous trophozoite-stage cultures (24 hour post invasion) of the 34 haploid progeny and two parental lines from the HB3×Dd2 genetic were grown using established in vitro methods [26]. Metabolites were harvested from each of the cultures and high-resolution mass data were collected on an LTQ-Orbitrap. For maximum sensitivity, mass data were peak-picked near the noise threshold (minimum signal/noise = 3) and biologically relevant data were identified using a two-stage assignment routine. In the first stage, promising signals in the untargeted mass list (N = 124,020) were identified by QTL analysis and genetic linkages with LOD scores greater than 3 (N = 1,707) were manually curated to remove artifacts and correct for errors in the automated MS data peak picking algorithm. Untargeted mass data are highly redundant because electrospray ionization generates numerous adducts and in-source fragments for each input metabolite. Consequently, we employed a second assignment stage to condense redundant data into a single representative parent mass for each compound. Curated signals (N = 279) were clustered by coelution, signal covariance, and mass difference relative to common adducts/isotopomers/fragments. The most intense signal from each group was designated as the parent mass. A total of 15 signals passed this two-stage filtering routine. Each of these signals had LOD scores above their permutation-established thresholds for genome-wide significance (α = .05), and all but 3 of these signals were significant after Bonferroni correction (α = .05/24) for multiple hypothesis testing (Table 1). Notably, all but one of these 15 signals have LOD scores above the 5% false discovery threshold (LOD = 5.5) established by Q-value analysis for the original unfiltered mass list. Surprisingly, all Bonferroni-compliant signals were linked to a single 22 cM genomic region on chromosome 7 (Figure 1) containing pfcrt, the gene responsible for conferring chloroquine resistance [5].
Tentative metabolite assignments were generated for each of the PfCRT-linked masses by submitting observed signals to the Madison Metabolomics Consortium Database [27] and the Human Metabolome Database [28]. The resulting list of putative IDs was evaluated by analyzing ms/ms fragmentation spectra from enriched parasite extracts (Figure 2). Eleven of the 15 significant compounds (α = .05) had exact masses and fragmentation spectra consistent with small peptides. Peptide assignments were empirically validated by co-elution of the parasite-derived signal with synthetic peptide standards. These experiments were conducted using single reaction monitoring (SRM) mass spectrometry, a robust analytical method for confirming specific metabolite assignments in complex mixtures [29]. Each of the PfCRT-linked peptides co-eluted with their respective synthetic standards. Moreover, the intensity of the parasite-derived signal changed in a concentration dependent manner when standards were added to parasite extracts (Figure 2).
QTL analysis demonstrated that the peptide accumulation phenotype observed in CQR parasites maps to a 36 kb locus on chromosome 7 containing pfcrt and eight other genes (Figure 3). To determine if the pfcrt gene is responsible for the peptide accumulation phenotype, we analyzed transgenic parasites in which the native pfcrt gene has been replaced with either a CQS (C2; HB3 allele) or CQR (C4, Dd2 allele; C6, 7G8 allele) variant of pfcrt [11]. The two CQR alleles we tested have distinct evolutionary origins but all of the transgenic parasites share the CQS GC03 background (a progeny clone from the HB3×Dd2 cross). MS analysis demonstrated that parasites carrying either of the CQR pfcrt alleles accumulate peptides over the 48-hour intraerythrocytic life cycle to much higher levels (>32-fold) than parasites carrying the CQS allele (Figure S1). Furthermore, a survey of diverse parasite genotypes showed that all of the parasites that accumulate peptides carry the critical PfCRT-K76T polymorphism that is required for chloroquine resistance [30] (Figure 4).
Hemoglobin catabolism is an essential activity that provides amino acids and the physical space parasites need to grow [31]. All of the PfCRT-linked peptides identified by QTL analysis are found in, but not unique to, hemoglobin. Because PfCRT is located on the digestive vacuole membrane, which is the organelle where hemoglobin metabolism occurs, we hypothesized that PfCRT polymorphisms directly affect hemoglobin catabolism. To test this hypothesis, we conducted a comprehensive peptidomics analysis of parasites to monitor PfCRT-related effects on the hemoglobin digestion pathway. Erythrocytes infected with either CQS (C2) or CQR (C4, C6) parasites were purified by Percoll density gradient and endogenous peptides present in parasite extracts were analyzed by high-resolution nanospray LC-MS/MS. This analysis identified 362 endogenous peptides ranging from trimers to 32-mers that exactly correspond to sequences found in either the α or β chain of hemoglobin (Figure 5, Figure S3, and Figure S4). The majority of these peptides (e.g. VHLTPEE) have sequences that are unique to hemoglobin (i.e. have no other exact matches in either the P. falciparum or human genomes) and exist in overlapping clusters of structurally related peptides. The peptide clustering we observed is consistent with the parasite's semi-ordered hemoglobin digestion cascade, which involves protein degradation via a series of proteases (plasmepsins, falcipains, and falcilysin) and aminopeptidases [32]. The boundaries of most of the peptide clusters we observed coincide with established proteolytic cleavage sites [33]. In addition, we observed several sequence breaks that may suggest previously unmapped cut sites (Figure 5). Quantitative analysis identified 87 peptides that show evidence for differential accumulation between CQS and CQR lines (|z-score|>4; Figure 5). These peptides include the mQTL-linked peptides (e.g. PEE and DLS), other structurally related peptides (e.g. VHLTPEEK and HFDLS), and novel classes of peptides that fell below the analytical limit of detection in the original mQTL analysis (e.g. DPENFR in β-hemoglobin).
We interpret peptide accumulation in CQR parasites as evidence for impaired hemoglobin catabolism. Because hemoglobin catabolism is essential to P. falciparum [31], we anticipated that CQR parasites may have alterations in other metabolic pathways. To determine the degree to which CRQ pfcrt alleles affect metabolic homeostasis, we quantified metabolites present in extracts from density-purified samples of RBCs infected with transgenic CQS (C2) and CQR (C4, C6) allele exchange parasites. Using high resolution HPLC-MS, we quantified 80 metabolites, including representative central carbon metabolites, nucleotides, cofactors, amino acids, and peptides. Surprisingly, the only compounds showing consistent steady-state metabolic differences between the CQS and CQR lines were hemoglobin-derived peptides (Figure S5). These data suggest that altered hemoglobin catabolism is the most significant metabolic consequence of CQR mutations.
Given the significance of hemoglobin catabolism in the parasite's blood stage development, we hypothesized that CQR-induced peptide-accumulation would be associated with a fitness cost. To test this hypothesis, we conducted long-term competition experiments between CQS and CQR transgenic parasites grown for 70 days in mixed cultures containing either two (C2, C4; C2, C6) or three (C2, C4, C6) allele-exchange parasite lines in the same flask. This experiment differs from previous in vitro studies in our use of the transgenic pfcrt allele exchange parasites, which controls for the polygenic contributions from genetic background, and the long timeframe over which cultures were allowed to compete (∼35 generations). DNA was harvested every 48 hours and the abundance of each pfcrt allele was quantified by Sanger sequencing (Figure S6 and Figure S7). Quantitative DNA sequencing showed that mixed populations of CQS/CQR parasites converted to nearly pure populations of CQS parasites after 70 days (Figure 6). This result was consistent across both Asian (Dd2, C4) and South American (7G8, C6) CQR pfcrt alleles and the outcome was not influenced by the starting ratio of the mixed populations (Figure 6 and Figure S8).
Our competition assay showed a transient increase in CQR allele abundance that peaked at ∼20 days. This transient peak is attributable to a difference in cell cycle length between CQR and CQS parasites. Differences in cycle length accumulate across generations and thus progressively offset cycle stages of competing populations. Since parasites amplify their DNA midway through their 48 hour cell cycle, and daughter progeny do not all successfully re-invade, the population with the longer generation-to-generation replication time will have more DNA (but fewer infected cells) when populations are offset across generations (i.e. the leading population is at the ring stage whereas the trailing population is at the schizont stage). To quantify this phenomenon, we constructed a computer model of in vitro P. falciparum competition. Our model showed that a 2 hour difference in cell cycle length and a 13% overall fitness cost per generation can explain both the transient increase in CQR DNA and the subsequent disappearance of CQR alleles from mixed cultures (R2 = .98, Figure 6). This computational assessment is empirically supported by competition experiments involving asynchronous populations of parasites, which showed the same long-term population dynamics but lack the transient increase in CQR allele abundance (Figure S8).
Since CQR parasites are less fit (Figure 6 and Figure S8) and have altered hemoglobin metabolism when compared to CQS parasites (Figure 6), we predicted that CQR parasites would have more difficulty in using hemoglobin-derived amino acids for biomass production. To test this, growth assays were conducted in a modified RPMI medium lacking all amino acids except isoleucine (which is not present in hemoglobin), which forces parasites to use hemoglobin to supply its amino acid needs. In agreement with the literature [31], [34] parasites incubated in isoleucine-only medium grew more slowly than those incubated in rich medium. However, CQR parasites were significantly more impaired than CQS parasites (p = 0.0075, Figure S9), suggesting that CQR-induced fitness changes are linked to hemoglobin catabolism.
This study provides four independent lines of evidence linking chloroquine resistance to hemoglobin catabolism: I) mQTL analysis demonstrated that elevated hemoglobin-derived peptides co-segregate with the CQR-encoding pfcrt locus (Figure 1 and Figure 3), II) levels of hemoglobin-derived peptides observed in parasite extracts predict chloroquine susceptibility in genetically diverse parasite strains from around the world (Figure 4), III) genetically identical parasite lines that differ only by CQR versus CQS isoforms of PfCRT recapitulate the peptide phenotypes observed in wild isolates (Figure S1 and Figure S2), and IV) forcing parasites to rely on hemoglobin as an amino acid source affects CQR parasites more severely than CQS parasites. In addition, we show that chloroquine resistant alleles affect the levels of 87 of the 362 observable peptides in the parasite's hemoglobin catabolism pathway (Figure 5). Finally, we demonstrate that the peptide accumulation phenotype is associated with a 2 hour increase in cell cycle duration and a 13% overall fitness cost in transgenic parasites that only differ by their pfcrt allele. Together, these data argue that the significant fitness disadvantage observed in CQR parasites is attributable to impaired hemoglobin metabolism.
These results provide the first molecular explanation for the reemergence of CQS parasites in wild populations following the cessation of chloroquine treatment. Parasites degrade ∼75% of the host cell hemoglobin over the course of their 48 hour intraerythrocytic developmental cycle [35], [36]. Any impairment in the hemoglobin digestion pathway directly affects I) the amino acid pool available for new protein synthesis [32], II) the osmotic stability of infected cells [37], and III) may reduce the number of developing merozoites that can fit within the physical confines of the infected erythrocyte. Any of these mechanisms could account for the increased cycle length and lower generation-to-generation fecundity we observed [37]. Although this is the first report of a metabolic perturbation inherent to CQR alleles of pfcrt, similar fitness-linked phenomena are associated with other drug resistance genes [13], [38]. Epidemiological analysis of the parasite population changes in Malawi and Hainan estimated the fitness cost of CQR to be ∼5% [39], [40]. Our in vitro competition studies support this conclusion and show that the fitness cost may actually be much higher in the absence of compensatory mutations (Figure 6).
While PfCRT isoforms are clearly the main contributor to the peptide-accumulation phenotype, our data also show that genetic background modulates PfCRT's effects on hemoglobin metabolism. The wide distribution of peptide levels observed across the HB3×Dd2 progeny (Figure 3), and the consistent secondary peaks observed in the mQTL analysis (Figure 1), suggest that loci other than PfCRT are contributing to peptide accumulation. Similarly, the CQR allele-exchange parasite lines (C4 and C6) both showed more extreme phenotypes than their respective parental lines (Dd2 and 7G8; Figure 4) despite having identical pfcrt sequences. The PfCRT/hemoglobin catabolism link we describe here, along with our peptidomics approach, provide a framework for investigating compensatory mutations elsewhere in the genome.
Identifying the native function of PfCRT is a subject of considerable interest to the parasitology community. One possible interpretation of the peptide accumulation phenotype is that wildtype PfCRT functions as a peptide transporter and that CQR mutations interfere with this activity [41]. This interpretation is supported by a recent report of glutathione transport in Xenopus oocytes expressing PfCRT [42], and by work in Arabidopsis, which showed that a plant PfCRT homolog mediates glutathione transport [43]. However, the broad diversity of sizes (2–32mers) and physical properties of peptides accumulated by CQR parasites are inconsistent with the relatively narrow range of substrates carried by most peptide transporters [44]. An alternative interpretation of the peptide phenotype is that CQR-associated mutations affect hemoglobin catabolism indirectly by altering the permeability of the digestive vacuole membrane. Resistance mutations in PfCRT, and perhaps other membrane proteins, may cause the digestive vacuole to leak protons [45], glutathione [42], heme, or other osmolytes, which thereby alter the solution conditions of the vacuolar compartment. Given that protease activity can be very sensitive to solution conditions [46], even modest changes in vacuolar conditions could interfere with hemoglobin catabolism by the parasite. Similarly, perturbations in solution conditions may affect protein-protein interaction and thereby disrupt the recently described hemoglobin degradation complex [47].
In conclusion, this study demonstrates that chloroquine resistance, impaired hemoglobin catabolism, and reduced parasite fitness are linked to polymorphisms in PfCRT. This surprising linkage provides a molecular explanation for the reemergence of CQS parasites in Africa and Asia. Our results suggest that co-formulating chloroquine with a P. falciparum protease inhibitor [48] may be an effective strategy for slowing the emergence of resistant parasites.
Synchronous parasites were grown using established methods [26] in RPMI 1640 supplemented with 25 mM HEPES, 100 µM hypoxanthine (all from Sigma), 10 µg/mL gentamycin (Gibco) and 2.5 g/L Albumax II (Gibco). A total of 47 parasite strains were analyzed in this study; these strains include 36 lines from the HB3×Dd2 cross (34 progeny and 2 parental), three pfcrt allele swap lines (C2, C4, and C6) prepared in an isogenic GC03 background [11], and 8 out-group lines (V1/S, PAD, 7G8, GB4, 3D7, D10, SL/D6, and 106/1) used to measure metabolic phenotypes in diverse genetic backgrounds. All cultures were maintained at 5% hematocrit in a 37°C incubator with an atmosphere of 5% CO2, 6% O2, and 89% N2. Cultures were triple synchronized using consecutive treatments with 5% sorbitol at 0, 48, and 56 h of culture. Invasion time was determined by preparing blood smears every two hours starting 34 h after the last sorbitol treatment. The zero time point was designated when cultures reached 95% rings. Samples were harvested at 24 h post invasion for the HB3×Dd2 cross study, 38 h for the parasite enrichment studies, and at several time points throughout the cyclic 48-hour asexual blood stage (12, 24, and 36 h) for the out-group analysis.
To confirm our PfCRT-related phenotypes and improve our analytical sensitivity, erythrocytes infected with late trophozoite-stage parasites (38 h post invasion) were separated from uninfected erythrocytes using an established density gradient method [49]. Briefly, bulk cultures were suspended at 30% hematocrit in RPMI. Cultures were layered over dual Percoll layers (70% lower, 30% upper) diluted in 1× RPMI (final concentration). Samples were centrifuged (2,000× g, 15 min) and the infected erythrocyte layer was collected from the 30%/70% interface. Infected cells were washed with 50 volumes of RPMI, then suspended at 0.4% hematocrit in RPMI. Parasitemias of the purified samples were checked by blood smear and the purified samples were allowed to recover for 4 h in a 37°C incubator prior to metabolite extraction (Text S1 provides a step-by-step protocol).
Our metabolite extraction protocol is adapted from a previously established method [50]. Metabolites were extracted by suspending 50 µL packed cells in 1 mL 4°C 90% methanol. Samples were vortexed and briefly sonicated, if necessary, to disrupt the cell pellet and generate a uniform homogenate. Homogenates were centrifuged (13,000× g, 10 min) and the supernatants were harvested. Samples were stored at −80°C as 90% methanolic extracts until metabolite analysis. Just prior to analysis, extracts were dried under a stream of N2 gas and resuspended in 200 µL H2O (Text S1 provides a step-by-step protocol).
Metabolite extracts were analyzed by high performance liquid chromatography (HPLC) mass spectrometry (MS). The chromatographic conditions used in this study have been described in detail elsewhere [50]. Briefly, metabolites were separated by reverse phase C18 chromatography run over a 50 minute (HB3×Dd2 cross study and coelution assays) or 25 minute gradient (all other studies) using tributylamine as an ion pairing agent. General metabolite analyses were conducted using negative-mode electrospray ionization on a Thermo Scientific LTQ-Orbitrap (HB3×Dd2 cross) or Thermo Exactive (all other studies). Metabolite assignments were validated by single reaction monitoring (SRM) on a Thermo TSQ Quantum Discovery Max triple quad. For peptidomics analysis, aliquots of each metabolite extract were harvested, diluted 1∶4 in a 3% acetonitrile and 0.1% formic acid solution (final concentrations), and analyzed in positive mode on an LTQ-Orbitrap using nanospray from a 120 minute hydrophilic interaction liquid chromatography (HILIC) gradient. Scans were conducted at both low (150–500 m/z) and high (450–1800 m/z) mass ranges to accommodate multiple charge states and MS2 scans were automatically conducted on fragments from each of the top seven signals observed at any given time. Both the original cross dataset and the peptidomic analyses were collected at the Princeton mass spectrometry facility; all other data were collected in house.
Genome-wide scans were performed using pseudomarker [51] to detect QTLs associated with metabolite levels in the HB3×Dd2 genetic cross. Intensities of mass signals were log-transformed and the median signal for each mass across replicates was used as a phenotype. Batch number was included as an independent covariate [52] to correct for run-to-run changes in MS instrument sensitivity. Genome-wide significance thresholds were determined by permutation testing (n = 1000 permutations) [53] and the strength of each linkage was expressed as a LOD score. We accounted for multiple hypothesis testing using established methods [21]. Briefly, false discovery rates were calculated from p-values using the q-value approach [54]. QTL-based significance scores were used to filter the large untargeted mass list to a manageable subset of putative signals. Final LOD scores and significance thresholds were computed using R/qtl [55] (interval mapping parameters step = 1, n.draws = 64; QTL mapping method = hk). The custom R code (Text S2 and Text S3) and data tables used for this analysis (Table S2 and Table S3) are included in the supplemental materials.
Mass data were visualized and analyzed using MAVEN, a freely available software package for MS-based metabolomics [56]. Data were peak picked using a permissive threshold (S/N = 3) and raw LOD scores generated by QTL mapping were then used to identify the most promising subset of signals. The extracted ion chromatogram of each signal with a LOD score greater than 3 was visually inspected and data originating from peak picking errors, thermal noise, elution artifacts, or associated with the void and wash volumes were excluded. Coeluting adducts, fragments, and isotopomers were condensed into their respective parent masses and the intensities for each of the final parent masses were hand-verified to correct for peak picking errors.
A list of potential compound identities was generated by searching the Madison Metabolomics Consortium Database [27] and the Human Metabolome Database [28] for metabolites matching the QTL-identified masses. Putative IDs were evaluated by ms/ms fragmentation analysis. Final compound identities were confirmed by coelution of the parasite-derived compounds with standards purchased from Sigma and the Proteomics Resource Center at Rockefeller University. The final assignment and quantification steps were conducted by single reaction monitoring (SRM) on a triple quadrupole mass spectrometer.
Peptide assignments were conducted using a comprehensive hemoglobin digestion library loaded into Mascot proteomics software (Matrix Science). Only assignments with mass defects of less than 10 ppm, matching scores greater than nine, and observable peptide signals in all nine of the extracts (N = 3 per genotype) were included. All assignments based on parent masses that mapped to adducts or fragments of hemoglobin peptides were excluded. A custom MAVEN-format standards library was generated using the Mascot results and the extracted ion chromatogram of each assignment was visually inspected and quantified in MAVEN. Peak intensities for each peptide were compiled and aligned to both the hemoglobin α and β primary sequences using custom software written in R.
Synchronous cultures of isogenic pfcrt allele exchange parasites (C2, C4, C6) were grown to the late trophozoite phase and magnetically purified from uninfected cells using a MACS column. The parasitemia of each enriched sample was determined by microscopy and cell counts were determined by hemocytometer. Mixed culture flasks containing either two (C2, C4; C2, C6) or three (C2, C4, C6) genotypes were constructed at mixing ratios of 50∶50 (C2/C4, C2/C6), 25∶75 (C2/C6), or 50∶15∶35 (C2, C4, C6). Each two-way competition experiment was run as a single biological replicate whereas the three-way competition was replicated in three independent flasks (established from a single seed culture) run in parallel. The entire experimental procedure was repeated a second time using asynchronous populations of parasites. Flasks were maintained continuously for 70 days under standard culturing conditions. Culture flasks were cut back 1∶10 every 48 h (to ∼0.5% parasitemia) and DNA was harvested from the excess parasites via Saponin lysis (0.1%) followed by genomic DNA isolation (DNeasy kit, Qiagen). The pfcrt allele present in each sample was PCR amplified (primers: CGAGCGTTATAGAGAATTAG, ACAACATCACCGGCTAAGAA). Products were then Sanger sequenced using independent diagnostic primers (GGCTCACGTTTAGGTGGAGG, ACAACATCACCGGCTAAGAA). Sequencing results were analyzed using online tools from Genewiz and allelic abundances in each flask were quantified using diagnostic single nucleotide polymorphisms in PfCRT amino acid positions 74–76, and 98 (C2: ATG AAT AAA AAC, C4: ATT GAA ACA AGC, C6: ATG AAT ACA GAC, Figure S6).
Allele frequencies observed in long-term competition experiments were fit using a custom model of P. falciparum in vitro growth. All modeling and regression analyses were conducted using custom software written for the R statistical software environment. Our model makes the following assumptions: 1) long-term changes in allele frequencies follow exponential kinetics, 2) parasite clones can differ with respect to life cycle length, 3) DNA abundance follows a sigmoidal accumulation over the life cycle with peak DNA synthesis occurring mid lifecycle, 4) most of the DNA synthesized in one generation is not amplified in the following generation because not all merozoites successfully reinvade, 5) parasite life cycle synchronicity follows a Gaussian distribution that becomes progressively broader with each generation. Using these assumptions, the relative DNA content expected in mixed culture flasks was modeled for each point in the 70 day competition experiment. Initial differences in allele frequencies were set according to the empirical mixing ratio, then life cycle lengths and exponential growth rates were sampled by grid search. A best-fit multiple regression model was identified by iterative grid searches with progressively finer increments of cycle lengths and growth rates. The custom R code (Text S2 and Text S3) and data table (Table S4) used for this analysis is provided in the supplemental materials.
|
10.1371/journal.pntd.0002982 | Risk Factors for Adverse Prognosis and Death in American Visceral Leishmaniasis: A Meta-analysis | In the current context of high fatality rates associated with American visceral leishmaniasis (VL), the appropriate use of prognostic factors to identify patients at higher risk of unfavorable outcomes represents a potential tool for clinical practice. This systematic review brings together information reported in studies conducted in Latin America, on the potential predictors of adverse prognosis (continued evolution of the initial clinical conditions of the patient despite the implementation of treatment, independent of the occurrence of death) and death from VL. The limitations of the existing knowledge, the advances achieved and the approaches to be used in future research are presented.
The full texts of 14 studies conforming to the inclusion criteria were analyzed and their methodological quality examined by means of a tool developed in the light of current research tools. Information regarding prognostic variables was synthesized using meta-analysis. Variables were grouped according to the strength of evidence considering summary measures, patterns and heterogeneity of effect-sizes, and the results of multivariate analyses. The strongest predictors identified in this review were jaundice, thrombocytopenia, hemorrhage, HIV coinfection, diarrhea, age <5 and age >40–50 years, severe neutropenia, dyspnoea and bacterial infections. Edema and low hemoglobin concentration were also associated with unfavorable outcomes. The main limitation identified was the absence of validation procedures for the few prognostic models developed so far.
Integration of the results from different investigations conducted over the last 10 years enabled the identification of consistent prognostic variables that could be useful in recognizing and handling VL patients at higher risk of unfavorable outcomes. The development of externally validated prognostic models must be prioritized in future investigations.
| In contrast to other clinical presentations of leishmaniasis in Latin America, American visceral leishmaniasis (VL) can lead to death in 5-10% of patients under treatment. The fatality rates associated with this disease have remained stable at a high level over the years in Brazil and are neither recorded in under-treatment patients from endemic countries of the Old World nor from non-endemic countries where such cases are imported. Since VL-induced lethality can occur even after the implementation of recommended therapy, the understanding of individual, clinical and laboratory factors that predispose to an unfavorable outcome might represent an important feature for informing better practice in the clinical management of cases. The present systematic review with meta-analysis brings together information on various prognostic variables associated with the severity of VL. Potential predictors identified in the studies surveyed were grouped according to the strength of evidence available, and 13 were considered to be of significant relevance. The gaps in the existing knowledge and the need for the development of externally validated prognostic models were also discussed. The results presented herein could be useful in identifying patients at higher risk of unfavorable evolution or death from VL, and might provide an aid in decision-making regarding the clinical management of VL cases.
| Visceral leishmaniasis (VL) constitutes a serious public health problem in endemic regions, especially in the Indian sub-continent, in North and East Africa, and in South America. However, VL is one of the most neglected diseases in the world [1], closely associated with poverty, for which effective and affordable chemotherapies remain scarce [2], [3]. In Brazil, American VL was originally confined almost entirely to rural areas in the northeast of the country, but since the 1980s the disease has spread to large cities in the northeast, southeast and center-west regions of the country [4]. During the first decade of the 21st century, some 40,000 cases of VL and 2,500 VL-related deaths were reported in the country with no signs of a significant reduction in the fatality rates [5], [6].
In the Americas, the transmission of VL to humans occurs through the bite of female phlebotomine sandflies of the genus Lutzomyia, which hosts the promastigote form of Leishmania infantum [7]. After a relatively long incubation period of 3 to 8 months, the disease manifests itself through signs and symptoms that include irregular or remittent fever, cough, tiredness, weakness, loss of appetite and weight, together with those caused by invasion of the parasite into the phagocytic system such as enlargement of lymph nodes, liver and spleen [8]. The evolution of VL varies from case to case, and some infected individuals may never exhibit any signs of the disease [9], [10]. In cases of VL-related mortality, the outcome results predominantly from hemorrhage or co-infection [11].
Treatment options for VL in Brazil are pentavalent antimonial compounds and formulations of amphotericin B [12]. Although amphotericin B exhibits stronger antileishmanial activity than pentavalent antimonials, the treatment practice employed in Latin America is based on weak scientific evidence [4] and may induce parasite resistance [13] or be subject to host-related limitations associated with unresponsiveness, drug toxicity or prolonged parenteral administration [14].
The lack of reduction in the fatality rates of VL in Brazil can be explained not only by the limitations in therapy applied and the delay in diagnosis [12], but also by the lack of adequate management provided to individuals at higher risk of an unfavorable evolution of the disease. In this context, the identification of prognostic factors associated with VL might be a valuable tool for clinical practice. Prognostic factors are defined as variables that predict the course of a disease, possible outcomes and the frequency with which they can be expected to occur. Knowledge about such factors is essential in medicine, prompting the selection of the most appropriate diagnostic tests and treatments to be applied, assisting in the development of new medical interventions, aiding in the monitoring of disease progression, and facilitating the counseling of patients regarding their future health condition [15]–[17]. In the case of VL, prognostic indicators of disease severity could also be used to establish if treatment should be carried out in primary health care units or in specialized care centers, and would be of considerable value in prescribing specific interventions for patients at most risk of a lethal outcome [12], [18].
Generally, prognostic factors have received less research attention than etiological factors and therapeutics [15], [19], although in some medical fields, particularly those related to oncology, several prognostic models have been published [20], [21]. In Brazil, a number of studies have been performed with the purpose of identifying individual, clinical and laboratory factors associated with poor evolution of VL and/or lethal outcome [12]. However, to the best of our knowledge, no systematic review articles have been published summarizing the current state of understanding of VL prognostic factors and indicating the most consistent predictors.
Considering the relevance of predictors of clinical evolution in reducing the number of VL-induced deaths, and the need for reliable prognostic models (developed and validated according to appropriate methodologies), the present systematic review with meta-analysis seeks to bring together information reported in studies of the potential predictors of death and other adverse outcomes of American VL. In addition, based on the analysis of the limitations of the published studies and of existing knowledge we propose possible improvements that might be incorporated into future research.
Independent literature searches were conducted between March and September 2011 by two of the authors (VSB and DSB) using the databanks and keywords listed in Fig. 1. Additional studies were identified by contacting experts in the field and by searching reference lists within selected publications. The titles and abstracts of all articles identified in the searches were subjected to an initial evaluation, and the full texts of those considered potentially relevant by at least one of the authors were analyzed.
The systematic review encompassed epidemiological studies containing data that allowed us to estimate measures of association relating to predictors of death or of adverse prognosis independent of the occurrence of death (sets of signs and/or symptoms characterizing the continued evolution of the initial clinical conditions of the patient following the implementation of treatment) in individuals diagnosed with VL. No restrictions were made regarding the age or gender of the patients or of the language of the publication. The exclusion criteria proposed were (i) studies performed outside Latin America; (ii) reports published as proceedings of symposiums or conferences; (iii) studies restricted to the description of signs and symptoms observed in VL-infected individuals without comparisons regarding the evolution of the disease; (iv) studies that simply described the existence of statistically significant (or not) associations without reporting at least the calculated P values or crude data that made possible the calculation of effect sizes (provided such information had not been obtained directly from the authors); (v) studies containing confusing text or incomprehensible analyses; (vi) studies exhibiting bias or inconsistencies that invalidated the results; and (vii) studies of prognostic factors related to genetic features or to quantification of cytokines.
The extraction of data from the publications was performed by one of the authors (VSB) and verified by the co-authors. Attempts were made to contact the authors of original reports when further information was required in order to calculate measures of association for possible inclusion in the meta-analysis. Data pertaining to individual patients were not requested.
The selected studies were separated into two main groups according to the outcomes, namely: (i) adverse evolution of the disease independent of death (as defined in the last section), (ii) evolution of the disease resulting in death. The first group of studies encompassed various possible outcomes and the information concerning each of the clinical or laboratory predictors identified was, if considered plausible (i.e. if the issue examined, the cut-off points and methods of analysis were not divergent), combined through meta-analysis of one sized P-values using the Stouffer method, weighted proportionally to the inverse of the study squared standard error [22]. In the second group, meta-analysis of measures of association were performed when cut-off points for defining variable categories employed in primary studies had close values. In this case, the effect-sizes adjusted by the greatest number of variables in each study were pooled regarding the odds ratio (OR). However, when there was divergence regarding the cut-off points, or when the predictors were defined differently but were relatively similar, meta-analysis of P-values was carried out as for the first group. For both groups of studies, we conducted theoretical discussions about variables that could not be submitted to meta-analysis, either because of the small number of studies involved or because of the non-uniform manner in which the data were presented or analyzed among the primary studies.
Measures of association were combined using the random effects model, except when the number of studies was less than four in which case the fixed effects model was employed [3]. The occurrence of heterogeneity in measures of effect between studies was analyzed using the I2 test, which describes the percentage of total variation across studies associated with real dispersion in effect-sizes (inter-study variation) rather than random error (intra-study variation). For each prognostic factor, the studies were separated according to the ages of the participants (adults and children) and evaluations were performed separately for each group. When the measures of association were similar in the two groups the data were combined, otherwise the combination of data was performed only within the specific group.
Meta-P software was employed for the meta-analysis of P-values, while CMA software version 2.0.057 was used for all other meta-analyses.
The relative strength of each of the clinical and laboratory variables as a predictor of the severity of VL was evaluated according to defined criteria which were, in decreasing order of weight: (i) force of summary measure obtained through meta-analysis; (ii) pattern of data (direction of association and heterogeneity in studies where the outcome was death); (iii) number of statistically significant studies in which the control for confounding variables had been performed; and (iv) pattern of associations in studies where the outcome was unfavorable clinical evolution independent of death.
There is no universally accepted or standardized tool for the identification of limitations or potential risks of bias in the analysis and/or presentation of data in studies relating to prognostic factors. Thus, in order to analyze the quality of studies reviewed we opted to use five publications [15], [16], [23]–[25] describing principles and methods for the development of prognostic models. Additionally, the STROBE statement, the aim of which is to strengthen the reporting of observational studies in epidemiology, was used to complement these resources [26]. Based on these publications, a set of 17 conditions was established in order to evaluate the adequacy of the methodology employed and the clarity of presentation of the results described in the included studies (Fig. 2).
Of the 2945 studies identified and screened as part of the comprehensive survey, only 14 prognostic studies [11], [18], [27]–[38] complied fully with the inclusion criteria (Fig. 3). Although the survey covered studies conducted in all Latin America, the 14 selected publications originated from Brazil. Ten publications described death as the outcome of interest, while three referred to the clinical evolution of the patients independent of death, and one targeted both outcomes. The sources of information used in these studies were medical records (11/14), direct interviews with the patients during hospitalization (2/14) and the Brazilian Information System on Disease Notification (Sistema de Informação de Agravos de Notificação; SINAN; 1/14) as shown in Table S1.
Each of the 14 studies reviewed employed appropriate criteria for selecting the study populations and defining the cases, and all except one [36] observed fully the premises for the statistical analysis of the data. Only two studies [27], [29] failed to employ any control for confounding factors and to describe the treatment adopted (which was always based on the recommendations of Brazilian Ministry of Health [12]), although a number of studies presented limitations regarding the definitions of variables [27], [31], [33], [36], [37], extraction of data from medical records [11], [27], [33], [34], [36], [37], selection of variables for the regression models [27], [29], [32], [36], [37], and description of the results [11], [27], [33], [36], [37]. Eight articles failed to provide information regarding missing data in the medical records/SINAN or sample losses [11], [27], [29], [30], [34], [36]–[38] and three [18], [32], [35] of the six studies that described these aspects did not treat the matter in the correct manner. Only one study [32] employed adequate criteria for the stratification of continuous variables. The statistical power was generally low and the treatment of data and the description of the methods employed for the construction of models were poorly described in most articles. For example, testing of interaction effects was described in only one study [18], while multicollinearity testing was fully described in just two studies [28], [38]. Additionally, more than half of the studies (9/14) ignored completely calibration and discrimination procedures [27], [29], [31], [33]–[38]. None of the studies addressed the issue of validation of the predictive regression models in populations other than that of the original study (Fig. 4).
All predictors of adverse evolution of VL and/or related mortality for which it was possible to perform meta-analysis (Text S1) are presented and classified according to strength in Table S2. Nine potentially strong predictors (Groups I and II in Table S2) were identified, namely, jaundice, thrombocytopenia, hemorrhage, HIV coinfection, diarrhea, severe neutropenia, age <5 years, age >40–50 years, dyspnoea and bacterial infection. All but the last three factors mentioned above presented summary measures significantly associated with mortality, consistency of effects in the direction of adverse evolution and mortality, and statistical significance in the majority of the multivariate analyses. While age >40–50 years, dyspnoea and bacterial infection were also strong predictors of death, their strength with respect to adverse evolution could not be assessed owing to the lack of studies exploring this outcome independent of death.
Apart from hepatomegaly, splenomegaly and weight loss, which were considered weak predictors, there was a predominance of statistically significant summary measures that showed, however, no significance in the majority of multivariate analyses (Groups III–V in Table S2).
Additionally, separate analyses of the variables in adults and children showed that there were no differences between the two groups except for the gender of participants and the interval between onset of symptoms and hospital admission, indicating that, in general, the predictors pointed in the same manner and direction independent of age group.
Prognostic factors that could not be submitted to meta-analysis did not form part of the classification of evidence adopted in this review. For example, Costa [32] reported that inappetence, kidney failure and high levels of alkaline phosphatase were highly associated with the risk of death, while other studies [29], [31] showed that VL-infected individuals with proteinuria had increased risk of unfavorable evolution and death. Moreover, Madalosso [33] demonstrated an association between mortality and positive myelogram, tuberculosis, dehydration, cardiovascular anomalies, asthenia, diabetes, splenectomy, myocardiopathy and abdominal pain. In addition, Costa et al. [31] demonstrated that VL-infected individuals with creatinine levels above 1.2 mg/100 mL exhibited high mortality risk, while Alvarenga et al. [27] showed that VL-infected individuals with comorbidities (HIV infection, liver and kidney diseases, cardiopathy, and other non-defined problems) had less chance of survival, similar to the findings of Araujo [28] for patients with other comorbidities (weakness and tuberculosis). Finally, Cavalcante [30] reported that individuals who recovered from VL within 20 days of treatment presented a higher mean eosinophil count as compared with those that did not recover, while individuals whose outcome was death exhibited higher mean values of prothrombin time and erythrocyte sedimentation rate compared with those that recovered. On the other hand, Braga [29] and Souza [38] showed that there was no significant difference between individuals that recovered and those that did not recover within 20 days of treatment regarding the mean lymphocyte count as well as when some cut-off point was used for this parameter.
The present systematic review identified, combined and analyzed information reported in studies addressing the factors associated with adverse prognosis of American VL and associated mortality. It was possible to identify a set of variables that ought to be considered in the clinical practice in order to improve disease management of patients and deserve further evaluation in future etiological and interventional studies in order to increase the empirical evidence on which to base their causal role.
The occurrence of jaundice was the strongest risk factor for severity of VL, demonstrating the relevance of hepatic impairment in disease progression. The association between jaundice and blood clotting disorders suggests the existence of a common hepatic mechanism [32], while liver dysfunctions in association with thrombocytopenia may lead to severe hemorrhage that could be responsible for the increased risk of death [39]. Considering that pentavalent antimonials, which represent the first-line of treatment of VL, are known to cause hepatotoxic side effects [40], VL-diagnosed individuals with jaundice or altered liver disease markers should be treated with amphotericin B-based pharmaceuticals rather than with antimonials. Inexplicably, this approach is not always followed, as exemplified by the patients investigated by Alvarenga et al. [27].
Hemorrhage was also a strong prognostic factor for adverse evolution of VL, and complications arising from this condition were major causes of death. Thus, the detection of bleeding at the first diagnosis or during the course of treatment is crucial in the identification of severity. According to Costa [32], hemorrhage is a consequence of the VL-induced inflammatory process, since pathogenesis of the disease is based on a cascade of events comprising activation of the inflammatory response, development of systemic endothelial lesions, activation of intravascular clotting, hypoperfusion, hypoxemia and, ultimately, cell death. Although the present review did not take into account the number of bleeding sites, it has been demonstrated that the greater the number of hemorrhagic points the higher is the risk of death [11], [32], suggesting that such relationship must be further investigated.
Thrombocytopenia was the second most important predictor of VL-induced death, although it is not possible to state with certainty if this variable is a cause or a consequence of hemorrhage. Splenic sequestration of platelets is possibly the main cause of a low platelet count [41], but this hypothesis only partially explains the disruption of homeostasis [32], and it has been suggested that thrombocytopenia is directly associated with the systemic inflammation induced by disseminated intravascular clotting [42]. From the reviewed data, it would appear that counts lower than 100,000 platelets/mm3 are indicative of high risk of adverse evolution, although a cut-off point of 50,000 platelets/mm3 is associated with an even higher risk. Thus, rather than attempting to define a standard limit of thrombocytopenia, it is of greater importance to assess each case separately in order to decide which is the most appropriate hemotherapeutic approach. In this context, the manual issued by the Ministry of Health of Brazil [12] recommends platelet transfusion only for VL patients presenting counts lower than 10,000 platelets/mm3.
Leishmania-HIV coinfection was a relevant prognostic factor for the adverse evolution of VL, since all studies analyzed and all multivariate analyses performed showed that coinfected individuals had a higher risk of poor prognosis. Considering that HIV induces the replication of Leishmania, that Th1-type immune response changes into Th2-type in both VL and HIV infection, and that HIV as well as Leishmania infect and multiply within cells of myeloid or lymphoid origin, the damaging effects of HIV and VL on the cellular immune system are not only synergistic but also reciprocally modulate pathogenesis [43]–[46]. According to Jarvis and Lockwood [47], the use of pentavalent antimonials is no longer recommended for HIV/VL-coinfected individuals owing to the high rates of failure and the level of toxicity associated with the treatment. These researchers emphasized the need for clinical tests to accelerate the development of more effective combined therapies and the planning of secondary prophylactic strategies. The Ministry of Health of Brazil [12] recommends HIV testing and the treatment with liposomal amphotericin B for all VL patients.
Together with hemorrhagic complications, the presence of bacterial infections is known to be an important cause of death among VL-infected individuals [11]. Even though this review included only studies that analyzed the occurrence of infections at the time of admission, the presence of coinfections represented a strong predictor of adverse evolution. This finding indicates the importance of preventing general infections and of treating VL patients isolated from individuals with bacterial infections, furthermore it calls attention to the damaging impact of late diagnosis on increased lethality of VL. Unfortunately, a large proportion of patients seeking medical assistance at hospitals or health units already presented opportunist infections and, possibly, VL at an advanced stage.
Severe neutropenia, characterized by the cut-off point of 500 neutrophils/mm3 [48], also constituted a strong predictor of VL severity. Patients with this condition had a higher risk of VL-related death, possibly because they were more susceptible to bacterial infections. In such cases, the use of antibiotics and the constant monitoring of this parameter are mandatory throughout the course of treatment.
Interestingly, diarrhea was a strong predictor of mortality. However, according to Werneck et al. [11], the occurrence of melena may be incorrectly interpreted as diarrhea or enteric microorganisms may be responsible for the sepsis associated with clotting abnormalities. Dyspnoea was also a good indicator of increased risk of unfavorable evolution of VL, and assessment of this condition, together with that of diarrhea, should be a routine priority in clinical practice since evaluation of these two parameters is rapid and straightforward, and their presence is possibly the result of more severe complications [35].
Although age of the subject was a strong indicator of poor clinical course of VL, most studies included in the review used dissimilar cut-off points, and few analyzed age as a continuous variable. Nevertheless, the data revealed that children of less than five years (especially those less than one year) and adults above 40 years (especially those older than 50 years) are more likely to have an adverse evolution. The distribution of lethality with peaks among children and older adults suggests that different factors may be involved in the acquisition of infections and complications at different ages [32]. In particular, the elderly are more frequently affected by comorbidities, such as cardiovascular diseases and weaker immunological resistance, which may increase the risk of death [49], [50]. On the other hand, children exhibit increased interleukin-10 levels and L-arginine secretion, which are factors associated with parasite persistence and greater VL severity. These parameters, coupled with the immaturity of the immune system, could explain the poor prognosis for this age group [42], [51]–[54].
Together with the strong prognostic factors of groups I and II (Table S2), it is worth considering the importance of the other groups of variables in the clinical evaluation of patients and for the purposes of improved disease management. For example, group III variables (Table S2) were statistically significant according to meta-analysis and some (but not the majority) of the multivariate analysis. In particular, the presence of edema emerged as a relevant indicator of VL severity since, although not significant in half of the multivariate analyses, it was strongly associated with death, similarly to the presence of vomiting.
The reduced strength of some relationships may be attributed to the specific therapeutic measures employed in some cases. For instance, individuals presenting hemoglobin levels below 7 g/dL would have received transfusions of packed red cells, as recommended by the Ministry of Health of Brazil [12], and this strategy may have diminished not only the degree of anemia but also the strength of the association between hemoglobin and VL severity. Nevertheless, low hemoglobin concentration was strongly associated with death and, therefore, it represented a relevant prognosis factor. Regarding undernutrition, there is evidence suggesting that this condition is more a consequence of the wasting syndrome in VL rather than a risk factor for severity. Furthermore, the control of Leishmania replication by the adaptive immunosystem, particularly by Th1 cells, of undernourished patients could explain the lack of association between undernutrition and mortality risk [32]. Some other laboratory variables, such as leukocyte count and levels of albumin, alanine transaminase (ALT) and aspartate transaminase (AST), constituted prognosis factors of intermediary evidence in the prediction of poor prognosis.
Several potentially relevant variables could not be included in the categories of evidence proposed herein because of the scarcity of studies. Among these are factors that can be readily assessed in clinical practice with minimal cost and must be better evaluated in future research, for example, mean cell volume, eosinophil count, serum creatinine, inappetence, weakness or asthenia, dehydration, lymphadenopathy and occurrence of comorbidities such as diabetes, tuberculosis, heart or renal diseases and dengue fever. In this context, it is noteworthy that the influence of helminthiasis on the clinical evolution of VL was not investigated in any of the reviewed studies even though infection by intestinal parasitic worms is highly prevalent in urban and rural areas of Brazil [55], [56]. It is well known that helminths can modulate and even suppress the immune response and, consequently, modify the clinical manifestations of diseases associated with the immune system [57], [58], hence this topic also should be included in future research. Other variables that require more specific and expensive tests, including myelogram, cardiovascular abnormalities, bilirubin levels, prothrombin time and partial thromboplastin time, have also received little research attention.
The present review provides a reliable source of information for the identification of risk factors of adverse prognosis and mortality in VL and should be used as an aid in decision-making in clinical practice. It is important to emphasize, however, that the results presented herein do not directly allow the creation and validation of prognostic scores based on the signs and symptoms presented by patients. Thus, studies should be carried out with the specific purpose of developing such scores and performing external validation of prognostic models already proposed, along with the incorporation of prognostic factors or additional biomarkers as recommended by Pencina et al. [59]. Assessment of the quality of the studies reviewed herein revealed that only five developed scores based on data obtained from the study populations, and no external validation of any kind was performed in these investigations. Prognostic models may present poor reproducibility and predictive performance when applied to other populations owing to the possibility of overfitting, the exclusion of some significant predictor, or differences between the characteristics of patients, health services or diagnostic methods [23]. According to Steyerberg et al. [60], a prognostic model is only useful if it is able to predict with accuracy the outcome of a patient who was not a member of the source population, i.e. the cohort employed in the development of the model, and studies that do not include at least some form of internal validation procedure (such as cross validation or bootstrapping) are rarely acceptable for publication. The manual issued by the Ministry of Health of Brazil [12] describes the implementation of a validation of the prognostic model developed by Costa [32], but does not include details regarding the procedures employed. For this reason, it is not possible to evaluate the score structure proposed in the model or to evaluate its potential applicability. However, this constitutes the first step in the formulation of a consistent prognostic model, with an impact that could be properly assessed, for application in different contexts in Brazil.
Concerning other limitations in the analyzed studies, the procedures adopted to deal with the problem of missing information from medical records were generally unclear. According to Little and Rubin [61], restricting an analysis to participants presenting complete records not only reduces the statistical power of the study but may also introduce bias. The pitfalls caused by missing data can be circumvented by the use of sophisticated statistical approaches especially designed for the imputation of missing information [62]–[65]. Such procedures should be employed as an alternative in all future studies whenever a set of values of variables are absent from a cohort.
The majority of studies considered in the present review failed to define the criteria adopted for the stratification of continuous variables. The quality of studies could be improved by adoption of credible and unequivocal clinical and analytical stratification criteria [66], or by analyzing continuous variables according to their original scale [67], [68] and by the implementation of appropriate procedures for the analysis of the functional form of their relationship with the outcome [69], [70].
Although the majority of the reviewed studies can be considered acceptable with respect to the adequacy of case definitions, statistical methods and multivariate analyses adopted, there were limitations in the models in cases where no interaction or multi-colinearity tests between the predictor variables were performed. In most of the studies, various prognostic factors were analyzed and many of them could be correlated, thereby producing the same explanation of variability in outcome [71]. In such cases, it is not correct to maintain all of the correlated variables in modeling procedures and, in view of the low statistical power of these studies, the exclusion of redundant explanatory variables would be helpful in increasing the accuracy of the multivariate model.
Considering the limitations of the present review, none of the studies conducted in other parts of the world were analyzed since those studies would reflect specific clinical, social and epidemiological characteristics distinct from those of VL in the Americas. Other relevant issues included the problem of combining data acquired from distinct populations (in terms of areas and characteristics) as well as the inability to explore the causes of heterogeneity of effect sizes between studies, and the impracticality of determining the existence of publication bias. Most studies described the results for all of the variables analyzed, but four articles did not provide data regarding some associations, particularly for non-significant variables, and this may have modified the true effect of some of the calculated summary measures. The force of these measures may also have been overestimated because of the use of odds ratio as a proxy for the relative risk [72], [73]. Additionally, there is the limitation regarding the sources of information, since most of the primary studies used retrospective data collected from medical records. The consistency and accuracy of such data is often a topic of discussion among researchers [74] because of the differences that exist in standards and in methods of registering data from one hospital to another. This does not mean that the use of medical records for research purposes should be abandoned, but that information derived from them should be examined with caution, and that those responsible for managing and for completing the records should be encouraged to improve the quality of information provided.
This is the first systematic review with meta-analysis on the prognosis factors relating to VL severity. The integration of information from different investigations conducted in Brazil in the last 10 years led to the identification of consistent predictor variables that might be useful in clinical practice for designing distinct therapies for patients at risk of an unfavorable outcome of the disease. The analysis of the quality of the published studies may be of assistance in future research, since positive features have been highlighted while logical criticism of the flaws, mainly relating to the external validation of multivariate prognostic models, has been offered. Similar assessments in different regions of the globe would be highly relevant since lethality of VL and the impact of this disease on our society can only be diminished by using consistent evidence-based medical approaches.
|
10.1371/journal.pbio.2003981 | Pseudomonas aeruginosa exoproducts determine antibiotic efficacy against Staphylococcus aureus | Chronic coinfections of Staphylococcus aureus and Pseudomonas aeruginosa frequently fail to respond to antibiotic treatment, leading to significant patient morbidity and mortality. Currently, the impact of interspecies interaction on S. aureus antibiotic susceptibility remains poorly understood. In this study, we utilize a panel of P. aeruginosa burn wound and cystic fibrosis (CF) lung isolates to demonstrate that P. aeruginosa alters S. aureus susceptibility to bactericidal antibiotics in a variable, strain-dependent manner and further identify 3 independent interactions responsible for antagonizing or potentiating antibiotic activity against S. aureus. We find that P. aeruginosa LasA endopeptidase potentiates lysis of S. aureus by vancomycin, rhamnolipids facilitate proton-motive force-independent tobramycin uptake, and 2-heptyl-4-hydroxyquinoline N-oxide (HQNO) induces multidrug tolerance in S. aureus through respiratory inhibition and reduction of cellular ATP. We find that the production of each of these factors varies between clinical isolates and corresponds to the capacity of each isolate to alter S. aureus antibiotic susceptibility. Furthermore, we demonstrate that vancomycin treatment of a S. aureus mouse burn infection is potentiated by the presence of a LasA-producing P. aeruginosa population. These findings demonstrate that antibiotic susceptibility is complex and dependent not only upon the genotype of the pathogen being targeted, but also on interactions with other microorganisms in the infection environment. Consideration of these interactions will improve the treatment of polymicrobial infections.
| Accurate prediction of antimicrobial efficacy is essential for successful treatment of a bacterial infection. While many studies have considered the impacts of genetically encoded mechanisms of resistance, nongenetic determinants of antibiotic susceptibility during infection remain poorly understood. Here we show that a single interspecies interaction between 2 human pathogens, S. aureus and P. aeruginosa, can completely transform the antibiotic susceptibility profile of S. aureus. Through multiple distinct mechanisms, P. aeruginosa can antagonize or potentiate the efficacy of multiple classes of antibiotics against S. aureus. We identify the exoproducts responsible for altering S. aureus susceptibility to antibiotic killing, and furthermore demonstrate that these compounds are produced at varying levels in P. aeruginosa clinical isolates, with dramatic repercussions for S. aureus antibiotic susceptibility. Finally, we use a mouse model of P. aeruginosa–S. aureus coinfection to demonstrate that the presence of P. aeruginosa significantly alters the outcome of S. aureus antibiotic therapy in a host. These findings indicate that the efficacy of antibiotic treatment in polymicrobial infection is determined at the community level, with interspecies interaction playing an important and previously unappreciated role.
| S. aureus is responsible for numerous chronic and relapsing infections such as osteomyelitis, endocarditis, and infections of the cystic fibrosis (CF) lung, as well as many penetrating trauma and burn infections, venous leg ulcers, pressure ulcers, and diabetic foot ulcers. These infections are notoriously difficult to treat, despite isolates frequently exhibiting full sensitivity to administered antibiotics, as measured in vitro using a Minimum Inhibitory Concentration (MIC) assay. This suggests that environmental factors present in vivo may influence the pathogen’s susceptibility to antibiotic killing. While these factors can include physical barriers to antibiotic activity, such as tissue necrosis and low vascularization at a site of infection, or bacterial replication within host phagocytes, treatment failure cannot be fully explained by poor drug penetration [1]. Instead, environmental determinants, such as interactions with the host, can induce phenotypic responses or genetic adaptations in bacteria that reduce antibiotic sensitivity [2,3].
Similarly, within complex polymicrobial communities such as those encountered in chronic skin infections, burn wound infections, and chronic colonization of the CF lung, inter- and intraspecies interactions can influence the pathogenicity and antibiotic susceptibility of individual organisms [4–6]. The presence of the fungal pathogen Candida albicans, for instance, can induce S. aureus biofilm formation and thus decrease the bacterium’s susceptibility to antibiotic killing [4]. Furthermore, antibiotic deactivation by resistant organisms within a population can lead to de facto resistance of all members of the community [7–10].
In such polymicrobial infections, S. aureus is commonly co-isolated with the opportunistic pathogen P. aeruginosa [11]. These co-infections are generally more virulent and/or more difficult to treat than infections caused by either pathogen alone [12–14]. The interaction between these 2 organisms is complex, with P. aeruginosa producing a number of molecules that interfere with S. aureus growth, metabolism, and cellular homeostasis. These molecules include the secondary metabolites 4-hydroxy-2-heptylquinoline-N-oxide (HQNO), pyocyanin, and hydrogen cyanide (HCN), all of which inhibit S. aureus respiration [15–17]. Additionally, P. aeruginosa produces rhamnolipids, biosurfactants that interfere with the S. aureus cell membrane, and an endopeptidase, LasA, that cleaves pentaglycine bridges in S. aureus peptidoglycan [18–20].
These anti-staphylococcal compounds allow P. aeruginosa to quickly eliminate S. aureus during in vitro coculture but do not prevent co-colonization in vivo. Recent findings suggest that within the CF lung, P. aeruginosa strains evolve to be less competitive with S. aureus, resulting in more stable coinfection of the same spatial niche [21]. Additionally, work by Wakeman et al. has shown that the presence of the abundant innate immune protein, calprotectin, induces a phenotypic switch in P. aeruginosa that promotes stable P. aeruginosa and S. aureus interaction through the chelation of zinc and manganese ions at the site of infection. This in turn represses P. aeruginosa metabolic toxin production, resulting in significantly less HQNO and pyocyanin [22]. Similarly, Smith et al. recently demonstrated that S. aureus can tolerate in vitro coculture with P. aeruginosa in the presence of serum albumin through the inhibition of P. aeruginosa lasR quorum sensing and thus LasA expression [23]. Despite these findings, P. aeruginosa LasA, rhamnolipids, HQNO, and pyocyanin are routinely detected at significant concentrations in burn wounds and in CF sputum samples and thus likely influence S. aureus physiology [24–28].
We hypothesized that interaction with P. aeruginosa may antagonize or potentiate S. aureus antibiotic susceptibility and could explain the frequent occurrence of treatment failure in infections involving otherwise drug-susceptible strains. Furthermore, we hypothesized that such interactions could be exploited to improve antibiotic treatment outcome. Here we demonstrate that secreted P. aeruginosa factors dramatically alter S. aureus susceptibility to killing by multiple antibiotic classes, and identify several mediators of S. aureus antibiotic antagonism or potentiation. Importantly, the production of these molecules is highly strain dependent, thus implicating the genotype of coinfecting P. aeruginosa strains as critical determinants of antibiotic treatment outcomes for S. aureus infections. Ultimately, we demonstrate in a mouse model of S. aureus, P. aeruginosa coinfection that the presence of P. aeruginosa can significantly alter the outcome of S. aureus antibiotic treatment. Overall, this work highlights the importance of considering the microbial context of the infection environment during the treatment of polymicrobial infection.
To investigate the impact of P. aeruginosa on S. aureus antibiotic susceptibility, we measured the bactericidal activity of 3 antibiotics against S. aureus in the presence of supernatants from 12 P. aeruginosa clinical isolates; 7 from the lungs of CF patients and 5 from burn wounds, as well as 2 laboratory strains; PAO1 and PA14. We were interested in examining how P. aeruginosa-secreted exoproducts can impact the susceptibility of S. aureus to vancomycin, tobramycin, and ciprofloxacin. Vancomycin is the frontline antibiotic for the treatment of methicillin-resistant S. aureus (MRSA). Ciprofloxacin and tobramycin are commonly used to treat P. aeruginosa during coinfection.
S. aureus strain HG003 was grown to exponential phase and treated with 500 μL of sterile supernatant from overnight (18 h) cultures of HG003 (control) or one of the 14 P. aeruginosa strains prior to antibiotic challenge. After 24 h, cells were washed and plated to enumerate survivors. We found that the individual bactericidal activities of all 3 antibiotics against S. aureus were affected by P. aeruginosa supernatants. More specifically, we observed 3 P. aeruginosa isolates that significantly protected S. aureus from killing by tobramycin (BC239, BC312, and BC252) and one P. aeruginosa isolate (BC310) induced over a 10-fold increase in tobramycin killing of S. aureus (Fig 1A). We also observed that the majority of P. aeruginosa supernatants were antagonistic towards ciprofloxacin killing (Fig 1B). Furthermore, supernatants from 8 P. aeruginosa strains (PAO1, PA14, BC238, BC310, BC249, BC250, BC251, and BC252) dramatically potentiated vancomycin killing of S. aureus, resulting in 100–1,000 times more killing than the control culture (Fig 1C). These data highlight the variable and strain-dependent influence of P. aeruginosa on the susceptibility of S. aureus to different antibiotics; however, the mechanism(s) by which P. aeruginosa alters S. aureus antibiotic susceptibility remained unclear.
Previous studies have shown that during coculture the presence of P. aeruginosa results in increased S. aureus resistance to tobramycin through the activity of HQNO [5]. In agreement with this, we observed that supernatants from BC239 and BC312 and BC252 protected S. aureus from tobramycin killing (Fig 1A). Paradoxically, however, we observed that the majority of our clinical isolates had no significant impact on tobramycin bactericidal activity. Even more striking, isolate BC310 appeared to potentiate tobramycin bactericidal activity against S. aureus (Fig 1A). We hypothesized that the impact of P. aeruginosa on S. aureus tobramycin susceptibility was multifactorial, with an unidentified factor increasing tobramycin bactericidal activity.
Tobramycin uptake is dependent on proton-motive force (PMF) [29]. P. aeruginosa HQNO collapses S. aureus PMF by inhibiting electron transport, thus abolishing tobramycin uptake into the cell [5]. To explore the possibility that an additional factor within P. aeruginosa supernatant may influence the bactericidal activity of tobramycin against S. aureus, we examined S. aureus susceptibility to tobramycin in the presence of supernatant from a PA14 ΔpqsLphzShcnC strain. This strain cannot produce the respiratory toxins HQNO, pyocyanin, or HCN, all of which inhibit S. aureus respiration and deplete PMF. Strikingly, we found that PA14 ΔpqsLphzShcnC mutant supernatant led to the rapid eradication of a S. aureus population following tobramycin treatment (Fig 2A) (S1A Fig). Heat-inactivation of P. aeruginosa PA14 supernatant had no impact on its ability to alter tobramycin activity, ruling out heat-labile proteins as potentiators of tobramycin killing (S1B Fig).
P. aeruginosa produces surfactant molecules called rhamnolipids that inhibit growth of competing gram-positive bacteria. These amphiphilic molecules increase cell permeability by interacting with the plasma membrane [30]. We hypothesized that rhamnolipid interaction with the membrane may facilitate tobramycin entry into otherwise tolerant, PMF-depleted persister subpopulations. To investigate this possibility, we deleted the rhlA gene in PA14, which is essential for rhamnolipid biosynthesis. Supernatant from a PA14 ΔrhlA mutant conferred full protection to S. aureus against tobramycin killing (Fig 2A) (S1A Fig). Furthermore, during tobramycin treatment, the exogenous addition of a 50/50 mix of purified P. aeruginosa mono- and di-rhamnolipids at 30 μg/ml facilitated the rapid eradication of the S. aureus population and decreased the MIC of tobramycin for S. aureus 8-fold (Fig 2A) (S1 Table). This concentration is within the physiological range of rhamnolipids likely encountered by S. aureus during coinfection with P. aeruginosa, as previous work by Bjarnsholt et al. found that clinical isolates produce a range of 2.4 μg/ml to 72.8 μg/ml rhamnolipids when grown in vitro, and Read et al. reported rhamnolipid concentrations as high as 64 μg/ml in a CF lung explant [24,31]. At the concentrations used in this study, rhamnolipids did not display antibacterial activity in the absence of antibiotic (S1C Fig). Further, incubation with a similar concentration of L-rhamnose, the glycosyl head constituent of rhamnolipids, had no effect on tobramycin killing, ruling out metabolite-stimulated PMF generation as the mechanism of tobramycin potentiation (S1D Fig). Finally, we found that 30 μg/ml purified P. aeruginosa rhamnolipids led to increased uptake of Texas Red-conjugated tobramycin as determined by flow cytometry (Fig 2B).
We next measured the relative amount of HQNO and rhamnolipids produced by each P. aeruginosa isolate using mass spectrometry [32] and a drop-collapse assay, respectively [33]. We observed a large variance in the production of both HQNO (Table 1) and rhamnolipids between isolates (Fig 2C). Importantly, the potentiator of tobramycin activity, BC310, was the only strain shown to be a high rhamnolipid producer without detectable HQNO production. In contrast, strain BC239, the strongest tobramycin antagonist, was among the highest HQNO producers, and did not produce rhamnolipids. Together, these data show that P. aeruginosa has the capacity to both positively and negatively influence S. aureus tobramycin uptake and bactericidal activity through the action of rhamnolipids and respiratory toxins, respectively. The presence of these 2 opposing factors may be responsible for the apparent disconnect between the P. aeruginosa-mediated increase in tobramycin resistance reported previously [5], and lack of protection from tobramycin killing following treatment with supernatant from the majority of P. aeruginosa strains observed in this study. Indeed, deletion of either P. aeruginosa respiratory toxins or rhamnolipids in a P. aeruginosa laboratory strain resulted in supernatants that facilitate complete sterilization or protection of S. aureus cultures, respectively (Fig 2A). Furthermore, similar trends were observed when S. aureus MRSA strain JE-2 was challenged with tobramycin following treatment with P. aeruginosa supernatant, supporting the relevance of this phenomenon in the clinical treatment of S. aureus infection (S6A Fig).
In addition to the ability of HQNO to inhibit uptake of aminoglycosides, we made an interesting and somewhat unexpected observation during our investigation. HQNO production in P. aeruginosa isolates correlated perfectly with protection against ciprofloxacin killing (Fig 1B) (Table 1). As ciprofloxacin uptake is PMF-independent, we wondered if P. aeruginosa HQNO was conferring ciprofloxacin tolerance in S. aureus via an alternate mechanism.
Antibiotic tolerance generally refers to a population-wide decrease in antibiotic susceptibility, often following exposure to external mediators of bacterial metabolism or physiology. In contrast, persister cells are generally described as antibiotic-tolerant subpopulations that form stochastically in an otherwise susceptible population. We recently demonstrated that both phenomena are specifically associated with cells entering a low ATP state [34,35]. Subpopulations of low energy cells give rise to persisters, while changes in the environment can lead to a low energy antibiotic-tolerant state in the entire population. HQNO inhibits respiration, the most efficient mechanism for ATP generation in S. aureus. We hypothesized that P. aeruginosa inhibition of S. aureus respiration induces a low ATP, multidrug-tolerant state of the entire population. In support of this, no protection from antibiotic killing was observed following pre-treatment with PA14 supernatant during anoxic growth (S2A Fig). We then cloned the fermentation-specific promoter for pyruvate acetyltransferase (pflB) from S. aureus upstream of gfp in a low-copy plasmid. Expression of pflB only occurs under anaerobic conditions or when respiration is inhibited [36]. We found that transcription of the pflB promoter was activated in response to supernatant from all of the P. aeruginosa strains with the exception of PA14 ΔpqsLphzShcnC (negative control) and 4 of the clinical isolates, BC236, BC308, BC310, and BC251. Importantly, these were the only clinical isolates that did not induce significant protection from ciprofloxacin killing (Fig 1B). Activation of pflB during aerobic growth demonstrates that respiration is inhibited in these conditions (Fig 3A). Direct intracellular ATP quantification of cultures treated with P. aeruginosa or S. aureus supernatant revealed that P. aeruginosa supernatant induces significant depletion of S. aureus intracellular ATP (Fig 3B).
We found that mutation of pqsL (HQNO negative) drastically reduced the capacity of PA14 supernatant to protect S. aureus from ciprofloxacin killing, suggesting tolerance to ciprofloxacin killing is mediated by HQNO (Fig 3C). Individually, mutations to the biosynthetic pathways for pyocyanin (phzS) and hydrogen cyanide (hcnC) had no influence on P. aeruginosa-conferred protection from ciprofloxacin killing. However, supernatants from a ΔpqsLphzS and a respiratory toxin-null mutant (ΔpqsLphzShcnC) were further reduced in their capacity to protect S. aureus from ciprofloxacin killing (S2B Fig) (Fig 3C). Together, these data demonstrate that P. aeruginosa confers protection from ciprofloxacin killing to S. aureus through respiration inhibition and depletion of ATP. Further, treatment with HQNO, pyocyanin, and HCN at concentrations detected within the sputum of CF patients with active P. aeruginosa infection [26,27,37] induced tolerance of S. aureus to ciprofloxacin (Fig 3D). Surprisingly, similar levels of tolerance were observed for other classes of antibiotics including tobramycin and vancomycin, with HQNO inducing the most robust tolerance to antibiotic killing (S2C and S2D Fig).
The presence of purified HQNO protects S. aureus from vancomycin killing (S2D Fig). However, P. aeruginosa supernatant from the majority of isolates tested significantly potentiated vancomycin killing of S. aureus (Fig 1C). We hypothesized that, similar to what was observed with S. aureus susceptibility to tobramycin, an additional factor present in P. aeruginosa supernatant is capable of overcoming the protective effects of HQNO to potentiate vancomycin killing of S. aureus. Heat denaturation of PAO1 supernatant completely abrogated the potentiating effect, suggesting the involvement of heat-labile extracellular protein(s) in the phenotype (Fig 4A). Bacteriolytic assays revealed that the PAO1 supernatant combined with vancomycin-induced dramatic lysis of the population that was absent in the presence of either factor alone (Fig 4B). This led us to examine the potential role of the P. aeruginosa extracellular lytic enzyme, LasA, in mediating vancomycin killing. LasA cleaves pentaglycine cross bridges in S. aureus peptidoglycan and has been shown to attack the cell wall of S. aureus during in vivo competition [18]. We examined the capacity of supernatant from a PAO1 lasA mutant to potentiate vancomycin killing. The lasA mutant supernatant did not potentiate killing by vancomycin compared to a 3-log reduction in S. aureus cfu in the presence of the PAO1 wild-type supernatant (Fig 4A) (S4A Fig). Similar trends were observed in a S. aureus MRSA strain JE-2 (S6B Fig). As it was previously shown that S. aureus can degrade HQNO [38], the absence of which could result in a more dramatic LasA-dependent potentiation effect in our supernatant experiments, we examined vancomycin killing in a coculture model where P. aeruginosa is present to continually produce HQNO. Again, we found that the presence of wild-type PAO1 resulted in a 3-Log reduction in cfu following vancomycin challenge, and that this potentiation was not observed in the presence of a PAO1 lasA mutant, where we observed 100-fold more survivors at 24 h (S7 Fig).
Next, we measured the levels of LasA in the supernatants of each clinical isolate via western blot and an additional assay developed previously to quantify LasA activity [39] (Fig 4C). Seven of the clinical isolates and both laboratory strains were positive for LasA. Of these, only BC253, the lowest LasA producer, and BC312, a high HQNO producer, did not induce at least a 10-fold increase in killing by vancomycin (Fig 1C). Of the 5 LasA negative strains, only one, BC251, significantly potentiated vancomycin killing, although no lysis of the culture was observed (S4B Fig). These data suggest that P. aeruginosa potentiates the vancomycin killing of S. aureus via at least 2 distinct mechanisms, only one of which is LasA-dependent.
Our observation that purified HQNO induces multidrug tolerance in S. aureus agrees with recent findings that P. aeruginosa protects S. aureus biofilm from vancomycin killing [40]. However, we have demonstrated that under planktonic growth conditions the protective effects of P. aeruginosa HQNO on S. aureus vancomycin susceptibility (S2D Fig) can be overcome by the lytic activity of LasA to potentiate vancomycin killing. In order to determine whether the protective effects of HQNO or the potentiating effects of LasA predominated in vivo, we adapted a previously described murine model of burn injury for S. aureus and P. aeruginosa coinfection [41]. Briefly, groups of mice were inflicted with a 20% total body surface area burn, then after 24 h were infected subcutaneously at the wound site with approximately 105 CFU S. aureus, HG003 alone, or in combination with 103 PAO1 or 103 PAO1 lasA∷tet. Mice were then treated daily with vancomycin and harvested 72h post infection.
LasA has been shown to mediate P. aeruginosa epithelial cell invasion and has been shown to be essential for corneal infections [42,43]. Interestingly, in our burn model, it appeared that the presence of PAO1 resulted in a higher burden of S. aureus, which is also dependent on lasA. However, for this study, we were interested solely on the impact of PAO1 presence on vancomycin sensitivity of S. aureus. While we observed no significant vancomycin efficacy in S. aureus mono-infected mice, relative to an untreated control group, we observed a 2-Log reduction in S. aureus burden following vancomycin treatment in mice coinfected with P. aeruginosa PAO1 (Fig 5A and 5B). Furthermore, no potentiation of vancomycin killing was observed in mice coinfected with the PAO1 lasA transposon mutant (Fig 5A and 5B). Importantly, P. aeruginosa appeared to be unaffected by vancomycin treatment, and burden was similar for both wild type and PAO1 lasA∷tet-infected mice (S8 Fig). Finally, we observed that PAO1 transcription of lasA is strongly up-regulated (approximately 200-fold) during in vivo coinfection (Fig 5C). Up-regulation of lasA transcription was also observed during P. aeruginosa monoinfection, suggesting that lasA expression is independent of the presence or absence of S. aureus during burn wound infection.
Together, these data demonstrate that the presence of P. aeruginosa can potentiate vancomycin killing of S. aureus during infection through the production of LasA. To our knowledge, these data represent the first evidence of P. aeruginosa altering S. aureus antibiotic susceptibility in vivo and underlines the importance of deciphering interspecies interactions to improve the antibiotic treatment of polymicrobial infections.
Polymicrobial infections are associated with exacerbated morbidity, accelerated disease progression and poor treatment outcome [44–47]. Antibiotic therapies are often selected to specifically target individual pathogens within a polymicrobial community without consideration of how interspecies interactions may alter a target organism’s antibiotic susceptibility. S. aureus and P. aeruginosa are 2 major human pathogens that frequently coexist within chronically colonized patients, and these infections are often impossible to resolve through conventional antibiotic therapy. We find that P. aeruginosa dramatically alters the susceptibility of S. aureus to the killing activities of commonly used and clinically relevant antibiotics through 3 distinct pathways governed by rhamnolipids, HQNO, and LasA, and that these molecules are produced at different levels by P. aeruginosa clinical isolates resulting in vastly different impacts on antibiotic efficacy against S. aureus (Fig 6; summarized in S2 Table). We found that P. aeruginosa staphylolytic activity correlates with vancomycin potentiation, and that P. aeruginosa HQNO production correlates with ciprofloxacin antagonism (S9A and S9B Fig). Correlation analysis with rhamnolipid production is not appropriate as the measurement of biosurfactant activity is qualitative. Overall, our results imply that antibiotic efficacy is strongly influenced by interactions between bacterial species, which may have major implications for future susceptibility determination and antibiotic treatment of polymicrobial infection.
Aminoglycosides are used routinely for the treatment of P. aeruginosa infection. Though aminoglycosides are effective against susceptible populations of S. aureus, bactericidal activity is limited against anaerobic, small colony variant (SCV), biofilm-associated, or persister subpopulations due in part to decreased respiration and thus PMF-dependent drug uptake [48,49]. Stimulating tobramycin uptake has been proposed as a way to eradicate these recalcitrant populations [49]. We have observed that P. aeruginosa-produced rhamnolipids sensitize the entire S. aureus population to tobramycin killing, leading to total eradication of otherwise tolerant populations. P. aeruginosa rhamnolipids may represent a promising new avenue for potentiating aminoglycoside killing of recalcitrant S. aureus and possibly other bacterial populations. Interestingly, a recent study has revealed that S. aureus increases tobramycin resistance in P. aeruginosa in an in vitro biofilm model, further emphasizing the importance of interaction between these organisms in dictating aminoglycoside susceptibility [50].
Vancomycin is a frontline antibiotic in the treatment of MRSA. To exert bactericidal activity against S. aureus, vancomycin must specifically bind the D-Ala-D-Ala residues of lipid II during cell wall biosynthesis [51]. However, vancomycin will also bind D-Ala-D-Ala residues of mature peptidoglycan. Thus, vancomycin exhibits limited bactericidal activity against dense populations of S. aureus cells due to the increased number of “decoy” targets available in late exponential or stationary phase populations of cells. Our data demonstrate that through LasA, P. aeruginosa can restore vancomycin efficacy against otherwise tolerant S. aureus populations. In support of this, we found that strains capable of increasing vancomycin lysis of S. aureus were LasA producers while the inert strains, generally, were not. This variance in LasA production may be due to mutations in lasR, an activator of lasA expression, which acquires mutations at high frequency during chronic P. aeruginosa infection [52]. We hypothesize that the combined action of cell wall degradation by LasA and inhibition of de novo peptidoglycan biosynthesis by vancomycin leads to cell wall destruction and a potent bactericidal effect.
P. aeruginosa HQNO induces tolerance of S. aureus to multiple antibiotic classes through respiration inhibition and depletion of intracellular ATP. Recent work by Orazi et al. found that in a bronchial epithelial tissue culture system, P. aeruginosa inhibited the killing activity of vancomycin through HQNO [40]. In agreement with these findings, we observed that the addition of exogenous HQNO protects S. aureus from vancomycin killing (S2D Fig) However, during planktonic growth the protective effect of HQNO was overshadowed by LasA-mediated potentiation of vancomycin killing. We were interested in examining whether P. aeruginosa antagonized or potentiated vancomycin in vivo, and found that P. aeruginosa expresses LasA at high levels during infection, and significantly potentiates the activity of vancomycin against S. aureus. This effect was abrogated in a P. aeruginosa lasA mutant suggesting that LasA plays a role in potentiating vancomycin killing of S. aureus during polymicrobial infection.
Similarly, we observed in vitro that P. aeruginosa-produced rhamnolipids can negate the protective effect of HQNO to restore or even increase S. aureus susceptibility to tobramycin killing. The opposing influences of HQNO and rhamnolipids on aminoglycoside activity against S. aureus may have major implications for aminoglycoside treatment of S. aureus during coinfection where production of one factor may dominate, leading to inhibition or potentiation of tobramycin activity against S. aureus. Indeed, we observed that clinical isolates produce a range of HQNO, LasA, and rhamnolipids, and production of each factor determines an isolate’s ability to potentiate or antagonize antibiotic killing.
P. aeruginosa strain variation and the impact of this variation on S. aureus antibiotic susceptibility suggests that a personalized approach to antibiotic therapy may be necessary to identify the ideal therapy to eradicate infection in an individual patient based on the genotype of S. aureus and the genotype of the bacteria it’s interacting with. Recent work has demonstrated that within the infectious environment, the production of HQNO, LasA, and rhamnolipids is highly variable. P. aeruginosa isolates from chronic CF infections frequently harbor mutations associated with decreased quorum-sensing activities and increased alginate production [52]. These mutations are attributed to the conversion to a mucoidal phenotype of P. aeruginosa that is significantly less competitive with S. aureus [53]. P. aeruginosa mucoidy is rarely associated with acute infection, thus the impact of P. aeruginosa on antibiotic susceptibility of S. aureus may differ during acute versus chronic coinfection. Future studies are necessary to identify genetic hallmarks of P. aeruginosa strains that potentiate or antagonize the activities of different antibiotic classes against S. aureus towards the goal of improving antibiotic efficacy against currently unresolvable coinfections.
It has long been observed that S. aureus is the dominant pathogen in the early life of CF patients with P. aeruginosa eventually dominating later in life [54]. It is interesting to consider a possible role of altered antibiotic susceptibility of S. aureus contributing to these dynamics, where vancomycin or tobramycin treatment in the presence of LasA- or rhamnolipid-producing P. aeruginosa strains may be particularly efficacious, resulting in a decrease in relative S. aureus abundance.
In summary, we characterized 3 distinct P. aeruginosa-mediated pathways altering S. aureus antibiotic susceptibility. HQNO induces a low-energy multidrug-tolerant state while LasA and rhamnolipids overcome this tolerance in cooperation with vancomycin and tobramycin, respectively. Exploitation of these newly discovered pathways may lead to better prediction of antibiotic efficacy in vivo and improved treatments for chronic S. aureus infection.
P. aeruginosa CF isolates were provided by an IRB-approved biospecimen bank (IRB#02–0948). P. aeruginosa burn wound isolates were from a previous study and use was deemed exempt by IRB study number #17–0836. All mice used in the study were maintained under specific pathogen-free conditions in the Animal Association of Laboratory Animal Care-accredited University of North Carolina Department of Laboratory Animal Medicine Facilities. All protocols were approved by the Institutional Animal Care and Use Committee at the University of North Carolina, protocol number 17–141, and all experiments were performed in accordance with the National Institutes of Health. Animals were anesthetized by inhalation of vaporized isoflurane. A subcutaneous injection of morphine was given prior to burn injury for pain control, and an intraperitoneal injection of lactated Ringer’s solution was given immediately after burn injury for fluid resuscitation. Animals were provided morphinated water ab libitum and monitored twice a day.
S. aureus strains HG003 and JE-2 were cultured aerobically in Mueller-Hinton broth (MHB) at 37°C with shaking at 225 rpm. HG003 is a well-characterized model strain of S. aureus, while JE-2 is a well-characterized USA300 S. aureus associated with community-acquired MRSA infection. For anaerobic growth, overnight cultures were washed twice with PBS and diluted into 5 ml of pre-warmed (37°C) TSB + 100 mM MOPS (pH7) to an OD600 of 0.05. Cultures were prepared in triplicate in 16 x 150 mm glass tubes containing 1 mm stir bars. Following dilution, cultures were immediately transferred into a Coy anaerobic chamber and grown at 37°C with stirring. P. aeruginosa strains were grown aerobically in MHB at 37°C with shaking at 225 rpm. Burn wound isolates represent the first positive Pseudomonal wound cultures obtained from 5 unique patients admitted to the NC Jaycee Burn Center between Nov 2015 and April 2016 with a total body surface area burn ≥ 20% and/or inhalational injury after obtaining informed consent. CF isolates were collected from 5 patients at the UNC medical center. Isolates were cultured from sputum or bronchoalveolar lavage (BAL) from patients with CF after obtaining informed consent. All burn and CF isolates were grown on Pseudomonas isolation agar (BD Difco) and verified with 16S ribosomal sequencing using the primer pair 5’-AGTATTGAACTGAAGAGTTTGATCATGG-3’ and 5’-CTGAGATCTTCGATTAAGGAGGTGA-3’ for PCR amplification.
The PAO1 lasA mutant (PW4282 lasA-H03∷ISlacZ/hah) was from the PAO1 knockout library [55]. PA14 deletion mutants were constructed by singly or sequentially deleting the coding sequences of rhlA, pqsL, phzS, and hcnC. Briefly, flanking primers were designed to anneal 800–1,200 bp upstream and downstream of the coding region. The resulting PCR product was inserted into plasmid pEX18Gm in accordance with the NEBuilder HiFi DNA Assembly protocol (New England Biolabs). Mutant alleles were integrated onto the chromosome of PA14 as described previously [56]. Briefly, pEX18Gm containing the in-frame deletion and gene-specific flanking regions was mated into P. aeruginosa via Escherichia coli S17-λpir. Primary integrants were selected for with gentamycin and irgasan, then grown for 4 h in LB without selection to allow for recombination. Dilutions of P. aeruginosa were plated on LB containing 8% sucrose for counterselection (loss of plasmid). Deletion strains were confirmed through PCR and sequencing (Genewiz). Plasmid PpflB∷gfp was constructed as follows, 298 bp upstream of the pflB coding region was amplified from HG003 genomic DNA using primers flanked with EcoRI and XbaI sites and cloned upstream of gfpuvr in plasmid pALC1434 [57].
To prepare sterile supernatants, S. aureus and P. aeruginosa strains were grown in MHB at 37°C with shaking at 225 rpm for approximately 20 h. The cultures were pelleted and supernatants were passed through a 0.2 μm filter. HG003 or JE-2 was grown to approximately 5 x 107 (for cell wall acting antibiotics) or approximately 2 x 108 cfu/ml (for all other antibiotics) in 3 ml MHB under aerobic or in 5 ml TSB + 100 mM MOPS under anaerobic conditions. Cells were pre-treated with 0.5 ml sterile supernatant (or 0.83 ml for anaerobic cultures) and returned to the incubator for a further 30 min. A 30-min pre-exposure was routinely used as we attempted to emulate the situation in vivo, where a population that has encountered and reacted to the relevant metabolites is subsequently exposed to antibiotic treatment. Where appropriate, cells were treated with 0.5 ml of P. aeruginosa culture taken directly from stationary phase (18 h) cultures in place of sterile supernatant. Coculture experiments were performed in the presence of 5% bovine serum albumin (BSA) to facilitate S. aureus/P. aeruginosa coexistence. An aliquot was plated to enumerate cfu before the addition of antibiotics. Antibiotics were added at concentrations similar to the Cmax in humans at recommended dosing; ciprofloxacin 2.34 μg/ml [58], tobramycin 58 μg/ml [59], vancomycin 50 μg/ml [60]. The Cmax of vancomycin is physiologically relevant for bacteremia and infections with good blood supply. The Cmax of vancomycin in serum is likely higher than that reached in the lung during IV infusion, however, it is certainly within the range experienced during inhaled therapy where clinical trials observed a Cmax of 270 μg/ml in sputum of CF patients. The Cmax of tobramycin is 58 μg/ml. Regarding lung concentrations, work by Ruddy et al. has found that inhaled tobramycin therapy results in sputum concentrations of between 17.2 and 327.3 μg/ml. The concentration we use is well within this range. Also, this therapy fails to eradicate P. aeruginosa in vivo and thus, we believe, is physiologically relevant for this study [61]. The ciprofloxacin blood Cmax used in this study is 2.34 μg/ml. This is also well within the physiologically relevant concentration for lung infection where ciprofloxacin concentration is actually higher than corresponding blood serum levels [62].
Ciprofloxacin concentration was increased to 4.68 μg/ml when cells were grown in TSB + 100 mM MOPS to account for any decrease in pH where ciprofloxacin killing activity is reduced. At indicated times, an aliquot was removed and washed with 1% NaCl. Cells were serially diluted and plated to enumerate survivors. We routinely used 2 time points to enumerate survivors, 16 h and 24 h after antibiotic challenge as we previously found that, in S. aureus, susceptible cells are killed and a stable subpopulation of survivors emerges between 16 and 24 h of exposure to antibiotics [35,63]. Where indicated sterile supernatant was heat-inactivated at 95°C for 10 min before addition to culture. Where indicated, pyocyanin 100 μM, HQNO 11.5 μM, sodium cyanide 150 μM, or rhamnolipids 10–50 μg/ml (50/50 mix of mono- and di-rhamnolipids, Sigma) or L-rhamnose 10–50 μg/ml were added in place of supernatant. Concentrations of respiratory toxins represent levels detected in the sputum of CF patients [26,27].
S. aureus strain HG003 harboring gfp promoter plasmid PpflB∷gfp was grown to approximately 2 x 108 cfu/ml in 3 ml MHB containing chloramphenicol 10 μg/ml. Cultures were treated with 0.5 ml supernatant from HG003, PAO1, PA14, or P. aeruginosa clinical isolates as indicated. 200 μl culture was added to the wells of a clear bottom, black side 96-well plate. The plate was placed in a Biotek Synergy H1 microplate reader at 37°C with shaking. Absorbance (OD600) and GFP fluorescence (emission 528 nm and excitation 485 nm) were measured every 1 h for 16 h. GFP values were divided by OD600.
HG003 was grown to approximately 2 x 108 cfu/ml in 3 ml MHB and pre-treated with 0.5 ml sterile supernatant from S. aureus HG003 or P. aeruginosa PAO1 or PA14. ATP levels of the cultures were measured after 1.5 h, as described previously using a Promega BacTiter Glo kit according to the manufacturer’s instructions [35]. P-values are indicated.
HG003 was grown to approximately 2 x 108 cfu/ml in 3 ml MHB and pre-treated with 0.5 ml sterile supernatant from S. aureus HG003, P. aeruginosa PAO1, PA14, or P. aeruginosa clinical isolates as indicated. Cells were incubated for a further 30 min before addition of vancomycin 50 μg/ml. 200 μl aliquots were added to the wells of a clear 96-well plate and placed in a Biotek Synergy H1 microplate reader. Absorbance (OD600) was measured every 1 h for 16 h.
P. aeruginosa strains were grown in MHB media for ~ 20 h. Cultures were normalized to OD600 2.0 and pelleted, and supernatants were passed through a 0.2 μm filter. Supernatants were boiled in SDS-sample buffer and run on a 4%–12% bis-tris acrylamide gel (Invitrogen). Protein was transferred onto a PVDF membrane, and LasA was detected using rabbit polyclonal anti-LasA antibodies (LifeSpan BioSciences, Inc.).
Staphylolytic assay was modified from Grande et al. [39]. Briefly, stationary phase S. aureus strain HG003 was heat killed at 95°C for 20 min. Cells were pelleted and resuspended in 20 mM Tris-HCl (pH 8.0) at an OD595 0.8–1. P. aeruginosa strains were cultured in MHB media for approximately 20 h. Cultures were normalized to OD600 2.0, pelleted and supernatants were passed through a 0.2 μm filter. Seventeen microliters of sterile supernatant were added to a 100 μl heat-killed cells. OD595 was measured at time 0 and after 2 h, and percent cell lysis was determined. The values shown represent the average of biological triplicates.
Tobramycin-Texas Red was made as described previously [64,65]. S. aureus strain HG003 was grown to mid-exponential phase and then incubated with or without 30 μg/ml rhamnolipids for 30 min. Cells were plated to enumerate cfu prior to addition of Texas-Red tobramycin at a final concentration of 58 μg/ml. After 1 h, an aliquot of cells was removed, washed twice in 1% NaCl, and plated to enumerate survivors. The remaining aliquot was analyzed for Texas Red uptake on a BD Fortessa flow cytometer. Thirty thousand events were recorded. Figures were generated using FSC Express 6 Flow.
P. aeruginosa rhamnolipid production was quantified utilizing a drop collapse assay, as previously described [33]. Briefly, clarified supernatants from overnight cultures of P. aeruginosa strains were serially diluted (1:1) with deionized water plus 0.005% crystal violet for visualization. Twenty-five microliters of aliquots of each dilution were spotted on to the underside of a petri dish plate and tilted to a 90° angle. Surfactant scores represent the reciprocal of the highest dilution at which a collapsed drop migrated down the surface of the plate.
Five hundred microliters of aliquots of P. aeruginosa supernatant were extracted 3 times with 1 ml of ethyl acetate containing 0.01% acetic acid. For each extraction, samples were vortexed for 30 s then centrifuged at 15,000 xg for 2 min. The organic phases were removed and combined in a separate tube and evaporated to dryness in a TurboVap under a gentle stream of nitrogen at 50°C. Dried samples were reconstituted in 250 μl acetonitrile and a portion diluted by a factor of 100 prior to analysis by liquid chromatography tandem mass spectrometry (LC-MS/MS). Quantitative analyses were performed on a Quantum Ultra triple quadrupole mass spectrometer (Thermo Scientific, Waltham, MA) equipped with an Acquity ultra-performance liquid chromatography (UPLC) system (Waters Corp., Milford, MA). A sample injection volume of 10 μl was separated on a 2.1 x 100 mm, 1.7 μm, CSH Fluoro-Phenyl UPLC column (Waters Corp., Milford, MA) at a flow rate of 250 μl per minute and a column temperature of 40°C. Mobile phase solvents consisted of 0.1% acetic acid in deionized water (A) and methanol (B). Separation was achieved with a linear gradient from 30% to 95% B over 5 min with a total run time of 10 min. Column effluent was diverted to waste from 1–3 and 7–10 min, and HQNO eluted at a retention time of 5.8 min. The mass spectrometer was operated in positive ion electrospray mode (3,000 V; 250°C), and data were acquired by selected reaction monitoring (SRM) in centroid mode using a mass transition of 260.1 to 159.3 m/z and a collision energy of 26 eV.
Animals were purchased from Taconic Farms and housed in specific pathogen-free conditions in the Animal Association of Laboratory Animal Care at the University of North Carolina’s Department of Laboratory Animal Medicine Facilities. All protocols were approved by the Institutional Animal Care and Use Committee at the University of North Carolina, and all experiments were performed in accordance with the National Institutes of Health. Wild type C57BL/6 mice were burned and infected as previously described [41]. Briefly, a 65 g copper rod was heated to 100°C and used to create a full-contact burn of approximately 20% of total body surface area through 4 applications of the rod to the anesthetized animal’s dorsal region. In preparing the inoculum, overnight cultures of bacterial strains were subcultured into fresh MHB and grown for 2.5 h to mid-exponential phase. An aliquot of each culture was centrifuged and washed with 1 ml of PBS + 1% protease peptone. Approximate bacterial density was calculated by absorbance (OD600), and cultures were diluted to obtain the desired concentration. Inoculum was verified through CFU enumeration. Mice were infected subcutaneously in the mid-dorsal region in unburned skin surrounded by the wound 2 4h after burn. Mice were administered vancomycin intraperitoneally at 110 mg/kg once daily for 2 d, then sacrificed 48 h post infection. At the time of sacrifice, 5 mm tissue biopsy of the bacterial infection site were aseptically removed, homogenized with 3.2 mm stainless steel beads and a Bullet Blender (Next Advance; Averill Park, NY), then bacterial burden was enumerated by plating serial dilutions of the homogenates on selective media. Mice that mostly cleared P. aeruginosa (<103 CFU/g tissue) were discounted from analysis.
Two hundred and fifty microliters of homogenized tissue were suspended in 1 ml of Trizol (ThermoFisher) and frozen at −80°C until extraction. RNA was extracted following manufacturer’s protocol. Extracted RNA was DNase treated with 10 units RQ-1 RNase free DNase (Promega) following manufacturer’s protocol. RNA was purified after DNase treatment using RNA Clean and Concentrator-25 (Zymo) and eluted in 30 μl of H2O, and quantified using a NanoDrop spectrophotometer. Five hundred nanograms of total RNA were used to generate cDNA using SuperScript II Reverse Transcriptase (ThermoFisher) and Random Primer 9 (NEB) following manufacturer’s protocol. Copy number of lasA and gyrA were quantified using iTaq Universal Sybr Green master mix (Bio-Rad) in 20 ul reaction volumes on a Roche LightCycler 96, using the following primer pairs: lasA_RT_5 5’-CCTGTTCCTCTACGGTCGCG-3’, lasA_RT_3 5’-GGTTGATGCTGTAGTAGCCG-3’, gyrA_5_RT 5’-GAAGCTGCTCTCCGAATACC-3’, gyrA_3_RT 5’-CAGTTCCTCACGGATCACCT-3’.
MICs were determined using the microdilution method. Briefly, approximately 5 x 105 cfu were incubated with varying concentrations of ciprofloxacin, tobramycin, or vancomycin in a total volume of 200 μl MHB in a 96-well plate. Where indicated, 34 μl MHB was replaced with sterile P. aeruginosa or S. aureus supernatant or purified HQNO or rhamnolipids at final concentrations of 11.5 μM and 30 μg/ml, respectively. MICs were determined following incubation at 37°C for 24 h.
Statistical data analysis was performed using Prism GraphPad software (San Diego, CA) version 5.0 b. Differences in S. aureus intracellular ATP concentration or surviving S. aureus CFU following P. aeruginosa supernatant treatment and antibiotic challenge were compared using a one-way ANOVA with Tukey’s multiple comparisons post test or the Student t test where appropriate. Differences in tissue burden following S. aureus and P. aeruginosa coinfection were compared with a Mann-Whitney test. Finally, the statistical significance of each correlation analysis was determined with a two-tailed Pearson’s chi-squared test. Differences with a p-value ≤ 0.05 were considered significant.
|
10.1371/journal.pgen.0030217 | Localization of Candidate Regions Maintaining a Common Polymorphic Inversion (2La) in Anopheles gambiae | Chromosomal inversion polymorphisms are thought to play a role in adaptive divergence, but the genes conferring adaptive benefits remain elusive. Here we study 2La, a common polymorphic inversion in the African malaria vector Anopheles gambiae. The frequency of 2La varies clinally and seasonally in a pattern suggesting response to selection for aridity tolerance. By hybridizing genomic DNA from individual mosquitoes to oligonucleotide microarrays, we obtained a complete map of differentiation across the A. gambiae genome. Comparing mosquitoes homozygous for the 2La gene arrangement or its alternative (2L+a), divergence was highest at loci within the rearranged region. In the 22 Mb included within alternative arrangements, two ∼1.5 Mb regions near but not adjacent to the breakpoints were identified as being significantly diverged, a conclusion validated by targeted sequencing. The persistent association of both regions with the 2La arrangement is highly unlikely given known recombination rates across the inversion in 2La heterozygotes, thus implicating selection on genes underlying these regions as factors responsible for the maintenance of 2La. Polymorphism and divergence data are consistent with a model in which the inversion is maintained by migration-selection balance between multiple alleles inside these regions, but further experiments will be needed to fully distinguish between the epistasis (coadaptation) and local adaptation models for the maintenance of 2La.
| A chromosomal inversion occurs when part of the chromosome breaks, rotates 180 degrees, and rejoins the broken chromosome. The result is a chromosome carrying a segment whose gene order is reversed. Whereas the physical rearrangement itself may have no direct consequences on gene function, recombination between alleles in the rearranged and wild type segments is suppressed. If multiple alleles inside the inverted or original orientations are well adapted to contrasting environmental conditions, suppressed recombination provides a mechanism to keep beneficial allelic combinations from being shuffled between different genetic backgrounds. Working with wild populations of flies, Dobzhansky provided the first evidence that selection was key to maintaining inversion polymorphism. Subsequently, examples of inversion polymorphisms under selection in other organisms have been found, notably in the mosquito that transmits most cases of human malaria, Anopheles gambiae. However, the genes or gene regions conferring fitness advantages have yet to be discovered. In this study, the authors used modern genomics tools to map such regions in an inversion at an unprecedented level of detail, and show that these regions are likely to be responsible for the maintenance of the inversion polymorphism in natural populations. This study lays the groundwork for future efforts to identify the genes themselves and their role in adaptation.
| Dobzhansky's studies of chromosomal inversion polymorphisms in natural populations of Drosophila provided the first evidence that selection played an indispensable role in their maintenance, helping to spark the neo-Darwinian synthesis [1,2]. More recent studies implicate selection in maintaining inversion polymorphisms in a diversity of eukaryotes, including humans [3–6]. A mechanism thought to facilitate their maintenance is reduced recombination. In inversion heterozygotes (heterokaryotypes), recombination between alternate arrangements may be inhibited both by asynapsis and because single crossovers within an inversion loop result in aneuploid meiotic products [7]. Such reduced recombination binds together favorably interacting genes (coadapted gene complexes) and/or multiple genes that are individually adapted to local conditions, and stabilizes them against gene exchange with migrants from other genetic backgrounds [1,8]. Stabilization of these allelic combinations allows the inversion to establish and spread, and consequently organisms can become adapted to highly divergent environmental conditions. Although selection has been invoked repeatedly to explain the maintenance of chromosomal inversions—and in some cases associated phenotypic traits have been identified [4]—the genes or regions involved remain elusive.
If a small subset of genes within an inversion were under selection and there was no gene flux at all between arrangements (sensu [9]), it would be impossible to identify specific genes or even regions affected by selection. Fortunately, inhibition of gene exchange between alternative gene arrangements is not absolute. Except near inversion breakpoints, gene conversion is unaffected and double crossovers can result in balanced recombinant gametes [10,11]. Working together, both recombinational processes (gene flux, [10]) gradually break up linkage disequilibrium within arrangements and homogenize sequence variation between them, unless countered by selection. The interaction of gene flux and selection is expected to produce a mosaic of more- and less-differentiated regions inside the inversion and away from breakpoints, exactly the pattern observed from some molecular studies of inversion polymorphisms in natural populations (e.g., [12,13]). These observations suggest that regions affected by selection can be identified, if not the precise genes and mutations involved. It is expected that such regions will be in significant linkage disequilibrium with the inversion and with each other even when they are not adjacent. However, a neutral explanation for patterns of linkage disequilibrium also exists: regions significantly associated with the inversion polymorphism may simply represent historical remnants of the genetic background upon which mutations arose. These alternative hypotheses can be tested using estimates of the rate of genetic exchange in heterokaryotypes and the age of inversion polymorphism.
Mapping of divergent regions between chromosomal arrangements is a prerequisite to identifying candidate genes under selection and ultimately elucidating the molecular basis of adaptations conferred by inversions. The reduced level of recombination in heterokaryotypes renders a traditional QTL (quantitative trait locus) mapping approach impractical. Instead, genomic scans of nucleotide divergence in natural populations take advantage of recombination over many generations. Previous scans for divergence and linkage disequilibrium in inversions have been hindered by the low resolution afforded by limited numbers of genetic markers. Application of modern genomics tools to the classical study of inversions has the potential to both accelerate and refine the mapping of diverged regions. Recent studies have demonstrated the utility of Affymetrix GeneChip arrays as high density genetic markers [14–17]. Emitted fluorescence from individual probes on the array directly correlates with the sequence similarity between hybridized DNA and the probe. In this manner, divergence between two genetic classes (e.g., alternative chromosomal orientations) can be examined at high resolution. Using this technique, we examined genic differentiation between individual Anopheles gambiae mosquitoes bearing alternate chromosomal arrangements—2La and 2L+a—on the left arm of the second chromosome, as a first step in identifying candidate regions maintaining inversion 2La in natural populations.
The 2La inversion system in A. gambiae is of interest not only as a model for understanding the adaptive role of inversions, but also for its epidemiological importance. A. gambiae is the most proficient vector of human malaria in the world, causing more than one million malaria-related deaths in sub-Saharan Africa each year [18]. Abundant inversion polymorphisms on chromosome 2 appear to play a key role in the ecological success of this species, as different inversion combinations are nonrandomly associated with both natural and anthropogenic environmental heterogeneities [3,19,20]. Inversion 2La is of particular interest and significance. First, it is the only inversion polymorphic on 2L, simplifying its analysis. Second, this inversion was acquired from an arid-adapted sibling species, A. arabiensis, by introgressive hybridization [3,21,22]. The nonrandom association of 2La with degree of aridity points to the adaptive value of this polymorphism in A. gambiae. In many different locations across Africa, 2La frequency exhibits strong and stable geographic clines from near fixation in arid zones to complete absence in humid rainforests [23–26]. Similarly, its frequency varies seasonally and microspatially according to patterns of rainfall and microclimate. Thus mosquito carriers of 2La are more likely than carriers of 2L+a to rest inside houses at night where a saturation deficit exists, affecting the probability of vector–human contact at peak blood feeding times (reviewed in [27]). From the standpoint of epidemiology and human health, the 2La polymorphism has increased malaria transmission by A. gambiae across diverse ecoclimatic zones and it could mitigate the efficacy of control measures that assume uniform indoor resting and biting behavior, such as bednets and indoor insecticide (or fungicide) application. Its study is facilitated by several recent developments, including a completely sequenced reference genome [28] and a resulting Affymetrix GeneChip array already proven to be an effective population genomics tool [16]. Furthermore, the breakpoints of the 2La inversion have been characterized molecularly [29] and the rate of genetic exchange on 2L between 2La/+a heterozygotes has been estimated in laboratory crosses [30].
In the following we address the selective maintenance of the 2La inversion polymorphism by multiple experiments. (1) We applied the Affymetrix GeneChip to map, at unprecedented resolution, highly diverged regions between alternate arrangements of a chromosomal inversion (2La). (2) We validated the principal microarray findings by targeted DNA sequencing, and used the resulting nucleotide polymorphism data to ask whether the introgressed chromosomal arrangement rose to its current frequency via adaptive natural selection. (3) We used known recombination rates in inversion heterokaryotypes to assess the likelihood that linkage disequilibrium between diverged regions and the inversion is maintained by selection.
The Anopheles probes on the Affymetrix GeneChip Anopheles/Plasmodium Array were designed mainly from the reference A. gambiae genome assembled from the chromosomally standard (uninverted) PEST strain [28]. The 25 bp probes (11 per transcript) interrogate ∼14,900 putative transcripts predicted from an early gene build (GeneBuild 2, 2003). From this core set of probes, those with more than one perfect match or with single nucleotide mismatches elsewhere in the genome were excluded from the analysis. The remaining 151,213 unique probes were distributed across the genome roughly in proportion to chromosome arm length. The 49 Mb chromosome 2L was represented by 33,892 probes, of which 13,984 mapped within the 22 Mb 2La inversion.
To minimize the contribution of ecological or geographic diversity to genetic variation, samples of A. gambiae homozygous for alternative 2L arrangements (2La/a and 2L+a/+a) were collected simultaneously from one village in central Cameroon where 2La is highly polymorphic and inversion heterozygotes are common (the 2La frequency was 46% in our 2005 sample of 70 mosquitoes, and in samples collected from the same village in 2002–2003 its frequency was 39%; F. Simard, unpublished data). The specimens used in this study were all identified as the S molecular form, one of two assortatively mating incipient species of A. gambiae [31]. Three of five 2La/a specimens were polymorphic for inversions on 2R; all other specimens carried the 2R standard arrangements. Labeled genomic DNA from each of five 2La/a and 2L+a/+a mosquitoes was hybridized to individual arrays (ten in total) to map nucleotide divergence, measured in terms of single feature polymorphisms (SFPs). SFPs were defined as probes whose hybridization intensities were significantly different between the five carriers of each 2L arrangement, as determined by two-tailed t-tests with a threshold of p < 0.01 [16]. A significant difference in hybridization intensity between samples reflects underlying differences in the target nucleotide sequences interrogated by the probes on the array.
Genome-wide, 1,352 probes (0.89%) were SFPs between 2La- and 2L+a-carriers. Of these, 444 were found on 2L, 283 of which were found in the rearranged region. The proportion of SFPs was notably higher on 2L (1.29%) than across the other four chromosome arms (0.79%; p < 1×10−21), which show consistent levels of differentiation (Figure 1). Along 2L, SFPs were distributed disproportionately within as compared to outside the rearranged region (1.98% versus 0.80%; p < 1×10−20; Figure 2). Indeed, the level of differentiation on 2L outside the rearranged region was indistinguishable from that of the other four chromosome arms (p = 0.75).
The distribution of putative SFPs on 2L was also explored by implementing a hidden Markov model (HMM) to identify differentiated and homogenized regions independent of a priori information about 2La breakpoint locations (cf. [16]). The HMM identified a 22 Mb diverged region corresponding almost precisely to the rearranged region, beginning at the first probe inside the proximal breakpoint and ending only ∼386 kb (243 probes) beyond the distal breakpoint.
To test for regional clustering of SFPs between the breakpoints of the rearrangement, a sliding window analysis was performed that revealed two regions in which the observed number of SFPs was greater than expected by chance (Figure 3). The first (proximal) cluster (p = 0.001) extends ∼1.0 Mb, approximately from 2L coordinates 21.1–22.1 Mb and as measured from its midpoint, ∼1.1 Mb from the proximal breakpoint (labeled “1” in Figure 3). Though the boundaries of the clusters are necessarily imprecise, the first cluster spans roughly 32 genes and contains 17 SFPs. The second (distal) cluster (p < 1×10−5) extends between coordinates ∼38.8–40.5 Mb, ∼2.5 Mb from the distal breakpoint (labeled “2” in Figure 3). The 63 SFPs in this ∼1.7 Mb cluster span approximately 178 genes. Both clusters remained significant (p < 0.05) after correction for the number of windows tested. The genes predicted in each cluster according to Genebuild AgamP3.4 (http://agambiae.vectorbase.org/index.php) are listed in Supplementary Tables S2 and S3.
The microarray analysis suggested two patterns. Most striking was heightened divergence between the rearranged region and little elsewhere on 2L. In addition, two significant clusters of SFPs inside the rearrangement were found near, but not directly adjacent to, its proximal and distal ends. Because of the limited sample sizes in the microarray analysis, we sought to confirm and extend these results by targeted sequencing of an additional 24–34 chromosomes carrying each gene arrangement, sampled from the same Cameroon population of A. gambiae. Eleven genes were chosen based on their location within or outside of the rearranged region (Figure 3; Table 1). Two were located ∼1 Mb outside of the proximal and distal breakpoints; inside, one was located centrally, one within the proximal cluster, three within the distal cluster, and two just outside of and flanking each cluster. Wherever possible, the corresponding genes were also sequenced from 2–6 chromosomes of two sibling species: a sympatric population of A. arabiensis (fixed for 2La) and an allopatric population of A. quadriannulatus (fixed for 2L+a). We used three approaches to explore whether DNA sequences supported the microarray results: comparing numbers of shared versus fixed differences between chromosomal arrangements, summary statistics of nucleotide differentiation, and gene tree reconstruction.
The numbers of polymorphisms shared between alternative arrangements in A. gambiae was high at the two genes outside the rearranged region (24 and 31), while the corresponding values for genes inside were much lower (ranging from 1–9) (Table 2). A small proportion of all shared polymorphisms may be due to recurrent mutation (Table 2; [32]), but most are shared because of gene flux (see below). As expected, the number of fixed differences followed a trend opposite to that of shared polymorphisms. No fixed differences occurred outside the rearranged region; inside, five of nine genes had fixed differences. The three genes nearest the proximal breakpoint all show relatively high numbers of fixed differences, even a gene distal to the proximal cluster (asph). These data therefore indicate that the boundaries of the proximal cluster were likely underestimated, though sequences at the breakpoint are expected to show fixed differences because of low levels of gene flux (see below). At the other end of the rearranged region, two of the three genes within the distal cluster show fixed differences, while no fixed differences were observed in either of the flanking genes, not even the one nearest the distal breakpoint. The ratio of fixed differences to shared polymorphisms for three genes within the distal cluster (cpr34, cpr63, srpn2; 9:14) was significantly higher than that for three genes outside and flanking the region (hdac, nwk, depcd5; 0:21; Fisher's exact test, p = 0.001). Because this test was conducted post hoc the results should be interpreted with some caution.
Differentiation between arrangements was also gauged by estimating FST values and net divergence along 2L. Pairwise FST values for all loci on 2L were significantly different from zero, but values within the rearranged region were roughly an order of magnitude larger than those in collinear regions (Table 2). Net divergence (Da) followed the same general pattern, with far greater levels of divergence observed inside the rearranged region. The average number of net nucleotide substitutions per site was 1.7% between arrangements and only 0.1% outside them. Moreover, those genes showing the highest levels of divergence were located in and around the proximal cluster and—with the exception of cpr63—inside the distal cluster. The lowest level of divergence was measured at depcd5 nearest the distal breakpoint.
Gene trees reconstructed from sequences at each locus yielded three basic patterns that were consistent with those that emerged from measures of divergence and fixed/shared variation (Figure 4). Owing to recombination, these branching patterns do not represent the exact evolutionary history of the genes sampled, but they do portray contrasting pictures of the extent of genetic exchange between arrangement classes. The first pattern, exemplified by lys-c in Figure 4A, shows the complete intertwining of sequences from inverted and standard arrangements, as expected if gene exchange has been frequent. This pattern was shared by both genes located outside of the inversion, as well as two inside at the distal end: cpr63 (in the distal cluster) and depcd5. These latter two genes showed the lowest level of net nucleotide divergence of any genes inside the inversion and correspondingly reduced FST values. The second pattern, common to five genes within the inversion (asph, endp, srpn2, unk and cpr34), shows reciprocally monophyletic 2La and 2L+a sequences, as expected if they are largely isolated. Three of four genes sampled from both clusters showed this pattern, including the srpn2 gene illustrated in Figure 4B. Two remaining genes inside the inversion (hdac and nwk) gave a third pattern indicative of limited gene exchange, such that a single sequence of one arrangement clustered together with sequences of the opposite arrangement as illustrated for nwk in Figure 4C).
A notable feature of all gene trees where outgroup sequences were available was the embedding of A. arabiensis (2La) and A. quadriannulatus (2L+a) sequences inside of A. gambiae 2La and 2L+a clades, respectively. A. gambiae is considered to have evolved from an A. quadriannulatus-like ancestor in a recent human-influenced speciation event in the central African rain forest [33]. If so, it would have carried only 2L+a, as A. quadriannulatus does. Although the 2La arrangement is ancestral in the A. gambiae sibling species complex [29,33], 2La likely passed into A. gambiae subsequent to the emergence of this species, following contact and genetic introgression with A. arabiensis [3,21,22]. Further evidence of the close genetic relationship between the same arrangement of 2L in different species can be seen from the contrasts presented in Table 2. In the rearranged region, greater differentiation exists between alternative arrangements within A. gambiae than between the same arrangement from different species. The opposite is true for collinear regions: differentiation is greater between species than within A. gambiae, due to free recombination in the latter.
Given that deeper sequencing of both chromosomal arrangements confirmed the existence of two highly differentiated clusters of genes within the rearranged part of 2L, we next asked whether there was any signature of selection in these clusters or on the 2La arrangement. The presumed recent introgression of 2La is inconsistent with a long-term balanced polymorphism. If this newly invading inversion was subject to strong directional selection in its rise to its current frequency (∼46%)—and this selection occurred in the recent enough past—a signature of selection on the level and frequency of nucleotide polymorphism should be evident. We used the sequence data collected from the 11 loci in and around the inversion to detect such a pattern.
Levels of nucleotide diversity in the rearranged region (calculated from the two separate samples of 2La and 2L+a chromosomes) were lower than in collinear regions (π: 1.11% versus 1.71%, Wilcoxon Rank-Sum test with 1-tail, p = 0.03; θ: 1.34% versus 2.14%, p < 0.02), as expected if there has been recent directional selection in the inversion. However, contrary to the expectation of the selective sweep hypothesis, levels of diversity within the inversion are highest within ∼1 Mb of the proximal breakpoint and generally decline moving distally (Table 1). The one exception to this pattern of declining heterozygosity is high levels of polymorphism in the cpr34 gene in the distal cluster.
A second prediction of the selective sweep hypothesis is that 2La chromosomes should contain less polymorphism than 2L+a chromosomes because of their more recent common ancestry. This pattern was not found: in fact, average levels of nucleotide diversity were slightly higher in 2La than in 2L+a arrangements, though the difference was only significant when diversity was estimated from the number of segregating sites (π: 1.21% and 1.02%, Wilcoxon Signed-Rank Test, p < 0.20; θ: 1.59% and 1.09%, p < 0.01). In addition, HKA tests [34] comparing A. gambiae 2La with A. quadriannulatus (2L+a), and A. gambiae 2L+a with A. arabiensis (2La) across loci from rearranged and collinear regions were not significant for either comparison (χ2 = 4.54, p < 0.96 and χ2 = 9.82, p < 0.40, respectively). These test results indicate that there is not significant heterogeneity in levels of diversity relative to divergence between rearranged and collinear regions, consistent with the absence of recent hitchhiking on 2L and the lack of major differences in mutation rate between lineages.
Two within-locus tests of deviation from the neutral-equilibrium model were conducted separately for 2La and 2L+a arrangements. Similar to previous sequence surveys of A. gambiae (e.g., [30]), Tajima's D statistic [35] was negative in most cases both inside and outside the rearranged region (Table 2), indicating an excess of low frequency SNPs (single nucleotide polymorphisms) consistent with a population expansion in A. gambiae [36]. None of the values of Tajima's D were significant, under equilibrium population histories or more realistic scenarios with expanding populations. However, four values of the R2 statistic [37], also based on the site frequency spectrum, indicated a significant excess of low frequency polymorphisms relative to the neutral-equilibrium expectation. Also evident in measures of the site frequency spectrum—and consistent with the selective sweep hypothesis—were the more extreme values of both statistics in 2La-arrangement chromosomes. At seven of nine genes within the inversion, values of Tajima's D and R2 were lower (indicating a greater excess of low frequency polymorphisms) among 2La chromosomes. This result would be expected if the 2La arrangement rose in frequency quickly, though this explanation is somewhat undermined by the fact that Tajima's D and R2 are also lower at loci in collinear regions among individuals carrying the 2La arrangement.
Although no clear footprint of a recent selective sweep or of balancing selection was found in the nucleotide sequence data, it may still be the case that selection is responsible for the maintenance of the proximal and distal clusters in association with inversion 2La. Multiple SNPs in both the proximal cluster and the distal cluster are in perfect linkage disequilibrium (D′ = 1) with the inversion (i.e., they are fixed between inversion arrangements), even though they are quite distant from the breakpoints. The alternative to a selective explanation for their maintenance is that the observed linkage disequilibrium is an historical remnant of complete association dating from the time that the inversion entered the A. gambiae gene pool. Two lines of evidence suggest that this date is quite recent. First, A. gambiae itself is considered quite recently derived. Based on its strongly anthropophilic behavior and its dependence upon anthropogenic breeding sites, Coluzzi and colleagues [20] have argued that A. gambiae is the product of a speciation process originating in the central African rainforest and driven by human impact on the environment subsequent to the Neolithic revolution ∼10,000–12,000 years ago. Second, based on the assumption of a single introduction of 2La, we can derive an estimate for the age of 2La in A. gambiae that agrees fairly well with this time frame. After removing polymorphisms shared between arrangements and between species, we used nucleotide polymorphism data from the proximal breakpoint (i.e., endp) to estimate that the E[TMRCA] of our sample of 2La chromosomes is ∼2.7 Ne generations (where Ne is the effective population size). Microsatellite-based estimates of Ne of A. gambiae are reasonably consistent across Africa [38]. Values of Ne obtained from Cameroon based on the infinite alleles or stepwise mutation models of mutation, respectively, were 11,500 and 49,000 [38]. This corresponds to the introduction of 2La into A. gambiae ∼3,000–11,000 years ago, assuming 12 generations per year.
Despite the relatively recent introduction of the 2La inversion into A. gambiae, we can distinguish between selective and neutral explanations for the maintenance of the inversion polymorphism by examining the amount of linkage disequilibrium expected between each of the clusters and their closest breakpoints given known rates of crossing-over. Using polymorphic microsatellite loci, Stump et al [30] estimated recombination rates on 2L from the backcross progeny of 2La/+a heterokaryotypes and as a control, from 2L+a homokaryotypes. They found that although recombination was at least 4× lower inside the inversion than in collinear regions, there were still appreciable levels of both gene conversion and crossing-over. From these data we estimate that the recombination fraction between the midpoint of the proximal cluster and the proximal breakpoint is r = 0.0012, and that the fraction between the midpoint of the distal cluster and the distal breakpoint is r = 0.0168. Given these estimates of recombination and Ne = 11,500–49,000 [38], the quantity 4Ner is much greater than 1 for the regions in-between both clusters and their closest respective breakpoints. With 4Ner ≫ 1, the only linkage disequilibrium expected in a population should be due to sampling variance [39]; we find that the observed value of the non-normalized linkage disequilibrium coefficient, D, is highly significantly different than 0 between polymorphic sites in either cluster and the inversion (p = 1.08 × 10−11 for the smallest sample size of any locus in the proximal and distal clusters). This result strongly supports the conclusion that some form of natural selection must be maintaining the association between the individual clusters and the inversion, and therefore the inversion polymorphism itself.
As an alternative way of considering the highly unlikely nature of the values of linkage disequilibrium observed, recall that disequilibrium declines as Dt = (1 − r)tD0, where Dt is the disequilibrium expected after t generations starting from an initial value of D0. Given values of r (see above) and a starting value of D0 = 0.25 (i.e., complete linkage disequilibrium), we would expect Dt to be less than 0.001 between the proximal cluster and the inversion in 4,600 generations and less than 0.001 between the distal cluster and the inversion in 190 generations. These numbers of generations translate to an almost complete lack of linkage disequilibrium after 380 years for the proximal cluster and 16 years for the distal cluster. If our estimates for the date of introduction of the 2La polymorphism are within even an order of magnitude of the correct time, these results suggest that more than enough time has elapsed for the decay of disequilibrium between these highly diverged regions and the inversion itself.
Use of the Affymetrix GeneChip microarray allowed us to map patterns of divergence in an inversion with unprecedented detail, leading to the discovery of two relatively small regions (the proximal and distal clusters) whose persistent association with the inversion is inconsistent with a neutral model. Below we discuss the forms of selection likely to be maintaining the inversion as a polymorphism in A. gambiae.
There have been numerous models proposed to explain the maintenance of inversion polymorphisms (reviewed in [2,40]). Perhaps the two most commonly cited are epistasis (coadaptation) among alleles within an inversion and overdominance of inversion heterokaryotypes [1]. Overdominance is an unlikely mechanism in this case. Multiple instances of stable geographic clines of 2La frequency along aridity gradients suggest that alternative arrangements are differentially adapted to dry and humid conditions, and that the cline results from a balance between migration and differential selection at opposite ends of an ecotone. This conclusion is reinforced by cyclical changes in 2La frequency associated with rainy and dry seasons each year. In addition, the overdominance hypothesis does not make clear predictions regarding linkage disequilibrium between loci within the inversion and the inversion itself, as a number of molecular mechanisms might be responsible for heterosis.
In contrast, Dobzhansky's coadapted gene hypothesis makes clear predictions about the nature of epistasis among alleles within the inverted region and linkage disequilibrium between these alleles and the inversion. The observation in Drosophila of linkage disequilibrium between inversions and genes only loosely linked to breakpoints has lead previous researchers to suggest that epistasis for fitness was maintaining these inversion polymorphisms, though epistasis was not directly evaluated in these studies [12,13,41].
An alternative selective hypothesis for the maintenance of inversions that does not require epistasis is also consistent with these findings. As in the case presented here, all of the inversion polymorphisms in which epistasis has been invoked exist along stable clines [12,13,41]. Population structure at multiple loci under selection across the cline can generate linkage disequilibrium among loci, in a multi-locus analog of the Wahlund effect [42]. This occurs because local adaptation to different environments at multiple loci can lead to parallel clines in allele frequencies, and therefore nonrandom associations among alleles. Under this model, migration-selection balance at two or more loci maintains the inversion; epistasis among the locally adapted alleles is not required and therefore the requirements of this model are less stringent than for coadaptation [8]. For A. gambiae, clines from arid to humid environments across Africa offer an ideal opportunity for local adaptation in multiple traits (see next section). Interbreeding among migrants carrying different genetic backgrounds on a collinear chromosome (e.g., humid- or arid-adapted) would create recombinants bearing fewer humid-adapted (arid-adapted) alleles, resulting in lower overall fitness under humid (arid) conditions. However, inversions that capture all humid-adapted alleles preserve their association in the face of immigrant arid-adapted genes (and vice versa). Thus, the inversion is maintained because it prevents recombination in the face of high levels of gene flow, as are observed in A. gambiae [43]. Further experiments will be needed to fully distinguish between the epistasis and local adaptation models for the maintenance of 2La.
Our data predict that the proximal and distal clusters should contain at least some candidate genes that confer resistance to aridity on 2La (and tolerance to humidity on 2L+a), though it is important to emphasize that additional candidates can occur outside of these clusters and possibly outside of the inversion itself. Within their estimated boundaries, a total of 210 genes have been predicted in both clusters. The challenge of identifying candidate genes within clusters is complicated by the fact that in many cases there is little evidence supporting gene predictions, with poor or nonexistent functional annotation (Tables 1 and 2).
The effort is further complicated by an almost complete lack of information regarding the physiological and/or behavioral traits responsible for aridity tolerance conferred by 2La, which can include both desiccation resistance and resistance to heat stress. In the only published study of desiccation resistance and water balance in A. gambiae and A. arabiensis, a laboratory colony of A. arabiensis was significantly more resistant to desiccation than a colony of A. gambiae, due to higher initial body water content [44]. Metabolic rate, respiratory pattern, rate of water loss during desiccation, and water content at death were similar. As karyotype was not investigated nor controlled for during this study, these data are difficult to interpret with respect to the contribution of 2La; both colonies are known to be polymorphic for several inversions that have been associated with aridity in the field and the A. gambiae colony used was polymorphic for 2La. The same problem applies to the sole study of heat resistance that found A. arabiensis to be more heat tolerant than A. gambiae in a behavioral assay and stress test [45]. In the absence of more detailed guidance from empirical work, the most striking observation about gene content concerns the distal cluster, which contains the largest concentration of cuticle protein genes (40) in the A. gambiae genome, as well as three hsp83 genes encoding heat shock proteins. However, the cuticle proteins are not present in the epicuticle, the layer primarily responsible for water retention [46]. Thus their role—if any—in heat or desiccation resistance remains obscure. Substantial additional effort will be required to pinpoint the important genes and to understand their contributions to adaptive phenotypes. As alluded to above, 2La is not the final story on resistance to desiccation; other inversions on 2R are also implicated in this trait [3,27]. Future progress will depend upon controlling for karyotype differences.
In the group of sister species known as the A. gambiae complex, there is a clear correlation between inversion polymorphism and involvement in malaria transmission [3]. The least polymorphic species are relatively restricted in their geographic distributions and are only locally important vectors or—in two cases—non-vectors. On the other hand, A. arabiensis and A. gambiae are counted among the most important vectors of human malaria worldwide. They carry abundant inversion polymorphism and are distributed across most of tropical Africa and its diverse landscapes. The impact of inversion 2La on the distribution of A. gambiae has been particularly profound. Once acquired from A. arabiensis, it helped A. gambiae to spread outside of the humid rainforest into arid savannas. Polymorphism for this and other inversions has enabled an already proficient malaria vector to occupy a vastly expanded species range, consequently expanding malaria transmission. Our results have laid the groundwork for the functional genomics study of 2La which will illuminate not only the genetic basis of adaptations inside inversions, but also aspects of vector behavior relevant to control.
All mosquitoes used in this study were field-collected. Collections of A. gambiae and A. arabiensis were performed between May and September of 2005 in the village of Tibati, Cameroon (6°28′N, 12°37′E) by pyrethrum spray catch. A. gambiae s.l. were identified morphologically and the ovaries of half-gravid specimens dissected and fixed in Carnoy's solution (3:1 ethanol:glacial acetic acid). Sibling species and molecular forms M and S were identified using an rDNA assay [47]. Karyotyping was performed following standard protocol [48]. Inversion status of 2La was confirmed by a PCR diagnostic [49]. A. quadriannulatus specimens were collected in 1986 from southern Zimbabwe and kindly provided by F. Collins [50]. DNA was isolated from individual mosquitoes using the DNeasy Extraction Kit (Qiagen). The concentration of eluted DNA for each specimen was determined by spectrophotometry using the Nanodrop-1000 (Nanodrop Technologies).
Fragmentation and labeling of 300 ng DNA from single specimens was achieved using random prime labeling in the presence of biotin-14-dCTP (BioPrime DNA Labeling System, Invitrogen) as described by J. Borevitz (http://naturalsystems.uchicago.edu/naturalvariation/methods/BorevitzSFPMethods.pdf). After purification by ethanol precipitation, labeled products were resuspended in 100 μl ddH20. Quality and yield (estimated at ∼10 μg) were checked by electrophoresis of a 5 μl aliquot through a 2.5% agarose gel. Most products were ∼50 bp long. The remaining 95 μl of labeled genomic DNA was hybridized to the Affymetrix Anopheles/Plasmodium GeneChip using standard protocols for eukaryotic cRNA hybridization. Hybridization and scanning of arrays was performed by the Center for Medical Genomics, Indiana University Medical School. All arrays were processed under identical experimental conditions on the same day.
Cel files containing the raw probe intensity values were imported into Bioconductor (http://www.bioconductor.org), an open source software project based on the R programming language (http://www.r-project.org). Using the “affy” package, data quality was assessed to identify aberrant chips or spatial artifacts [51]. Approaches included examination of chip images of raw probe intensities at natural and log-scales, boxplot and histogram summaries of unprocessed log scale probe intensities for each array, and MA-plots. To visualize more subtle spatial artifacts, the affyPLM package was used to examine chip pseudo-images based on the probe level model (PLM) fit. Background adjustment and quantile normalization was performed using the Robust MultiArray Average (RMA) method without summarization by probeset [52]. Probe level data were exported as a comma separated value file for importation into Excel and are available from BJW upon request.
Probes from the Anopheles/Plasmodium GeneChip have been mapped against the A. gambiae reference genome (AgamP3). To identify any probes with exact matches to multiple genomic locations or secondary one-off mismatches, a list of all probes and their genomic locations was obtained through VectorBase (www.vectorbase.org) [53] from K. Megy. A Perl script (available from BJW upon request) was used to parse probes with exact matches to unique locations; those with multiple exact matches or additional single base pair mismatches were excluded from further analysis. For each of the 151,213 probes retained a two-tailed t-test was performed to compare background-adjusted and normalized hybridization intensity values obtained from the five 2La arrays versus the five 2L+a arrays. Probes with p-values less than 0.01 were considered to contain SFPs between arrangements [14–16]. Overlapping significant probes were collapsed into one observation to control for nonindependence [16].
To test for overrepresentation of SFPs on 2L, we compared observed and expected numbers on 2L versus all other chromosomes combined, by a χ2 test. The expected number of SFPs in each category was calculated based on the genome-wide proportion of 0.89% as measured in this experiment. Similarly, overrepresentation of SFPs in the rearranged versus collinear part of 2L was tested by comparing observed and expected numbers given the 2L-specific proportion of 1.29%. An independent test of nonrandom SNP distribution on 2L that did not depend on prior information about the location of breakpoint sequences was implemented through a two-state HMM to identify differentiated versus homogenized regions along the arm. Transmission and emission probabilities of the HMM were estimated by expectation-maximization; hidden states were then inferred using the Viterbi algorithm in MATLAB (The MathWorks, http://www.mathworks.com/). To test for clustering of significant probes within the rearranged region, a sliding window analysis was performed with windows of 300 probes and a step-size of 20 probes. Each window was tested (χ2) for an excess of significant probes compared to the number expected by chance. A Bonferroni correction for multiple tests was conducted using the effective number of independent tests according to the relationship n* = n(1 − ρ)2, where n is the nominal number of tests conducted and ρ is the autocorrelation between successive test statistics [54,55].
A. gambiae GeneBuild AgamP3.4 incorporates manual annotations of genes predicted on 2L. Based on the manual models, primers targeting exons were designed using Primer3 [56] and custom synthesized (Invitrogen). Primer sequences for each of the 11 exons studied and the corresponding VectorBase gene identifier is given in Table S1.
PCRs were carried out in a 50 μl reaction containing 200 μmol/l each dNTP, 2.5 mmol/l MgCl2, 2 mmol/l Tris-HCl (pH 8.4), 5 mmol/l KCl, 10 pmol of each primer, 5 U Taq polymerase, and ∼10 ng of template DNA. Thermocycler (MJ Research) conditions were 94 °C for 2 min; 35 cycles of 94 °C for 30 s, 58 °C for 30 s, 72 °C for 1 min; a final elongation at 72 °C for 10 min; and a 0 °C hold. All 50μl of the resulting products were separated on a 1.25% agarose gel stained with ethidium bromide. Products were excised and purified using the Geneclean Spin Kit (MP Biomedicals) or QIAquick Gel Extraction Kit (Qiagen).
PCR products were directly sequenced on both strands using an Applied Biosystems 3730xl DNA Analyzer and BigDye Terminator version 3.1 chemistry as recommended by the manufacturer. Electropherograms were trimmed and visually inspected for SNPs and heterozygous indels using Seqman II (DNASTAR, Madison, WI).
Haplotypes at each locus were reconstructed from the genotypic sequencing data using the PHASE (version 2.1) program, which implements a Bayesian statistical model for inferring haplotypes from population genotype data [57,58]. All default settings in PHASE were used except for tri-allelic and quad-allelic SNPs, for which the default assumption of stepwise mutation intended for microsatellite loci was relaxed. After two haplotypes were assigned to each specimen alignment was performed using ClustalX [59].
DnaSP version 4.10.9 was used to calculate standard polymorphism and divergence statistics and tests of neutrality [60]. Coalescent simulations of population expansion were conducted in ms [61], with populations exponentially growing starting at 2.7 Ne generations in the past. Significance of FST values was based on 10,000 permutations conducted in Arlequin 3.11 [62]; significance of other values was determined from 10,000 coalescent simulations without recombination implemented in DnaSP [60]. A multilocus version of the HKA test of natural selection was implemented using HKA software developed and distributed by J. Hey (http://lifesci.rutgers.edu/~heylab/HeylabSoftware.htm#HKA). Using maximum composite likelihood distances [63], Neighbor-Joining gene trees were reconstructed in Mega4 [64].
To estimate the time to the most recent common ancestor of the 2La arrangement in A. gambiae, we used the expectation E[TMRCA] = 4Nef(1-ni−1), which is based on the number of segregating sites unique to each of the inverted and standard classes [65]. This estimate assumes that the 2La arrangement entered the population and instantaneously reached its current frequency. Violation of this assumption makes E[TMRCA] a minimum estimate of the age of the inverted class.
All sequences mentioned in this paper have been deposited in the National Center for Biotechnology Information (NCBI) GenBank (http://www.ncbi.nlm.nih.gov/sites/gquery) under accession numbers EU097365 to EU097703. |
10.1371/journal.ppat.1006032 | The Influence of Programmed Cell Death in Myeloid Cells on Host Resilience to Infection with Legionella pneumophila or Streptococcus pyogenes | Pathogen clearance and host resilience/tolerance to infection are both important factors in surviving an infection. Cells of the myeloid lineage play important roles in both of these processes. Neutrophils, monocytes, macrophages, and dendritic cells all have important roles in initiation of the immune response and clearance of bacterial pathogens. If these cells are not properly regulated they can result in excessive inflammation and immunopathology leading to decreased host resilience. Programmed cell death (PCD) is one possible mechanism that myeloid cells may use to prevent excessive inflammation. Myeloid cell subsets play roles in tissue repair, immune response resolution, and maintenance of homeostasis, so excessive PCD may also influence host resilience in this way. In addition, myeloid cell death is one mechanism used to control pathogen replication and dissemination. Many of these functions for PCD have been well defined in vitro, but the role in vivo is less well understood. We created a mouse that constitutively expresses the pro-survival B-cell lymphoma (bcl)-2 protein in myeloid cells (CD68(bcl2tg), thus decreasing PCD specifically in myeloid cells. Using this mouse model we explored the impact that decreased cell death of these cells has on infection with two different bacterial pathogens, Legionella pneumophila and Streptococcus pyogenes. Both of these pathogens target multiple cell death pathways in myeloid cells, and the expression of bcl2 resulted in decreased PCD after infection. We examined both pathogen clearance and host resilience and found that myeloid cell death was crucial for host resilience. Surprisingly, the decreased myeloid PCD had minimal impact on pathogen clearance. These data indicate that the most important role of PCD during infection with these bacteria is to minimize inflammation and increase host resilience, not to aid in the clearance or prevent the spread of the pathogen.
| Multicellular organisms are constantly interacting with microbes. Pathogens are microbes that can cause harm to the host if not properly controlled, therefore pathogen clearance is an essential part of survival of all multi-cellular organisms. Equally important factors in survival are host resilience mechanisms, or host processes that increase survival independent of pathogen burden. Not only can pathogens themselves cause damage to the host, the immune response that eradicates pathogens can cause harm in the form of immunopathology. Controlling and repairing damage are important factors in host resilience, and depending on the site of infection the specific mechanisms vary. This study examines the multiple roles that cells of the innate immune response play in both pathogen clearance and host resilience in response to both systemic and pulmonary pathogens.
| Pathogen clearance and host resilience/tolerance are both important in surviving a given infection [1,2] [3] [4] [5]. A main purpose of the immune response is to identify and clear invading pathogens. However, highly resilient hosts can survive infection with a given pathogen, independent of the ability of the immune response to clear it. One aspect of host resilience is prevention and repair of extensive tissue damage. Both the immune response and pathogens themselves can cause damage to the infected host [2] [3] [4] [5]. So while the immune system must act to clear a pathogen, it must also be carefully controlled in order to prevent excessive damage. This study seeks to understand the role that myeloid cells, cells of the innate immune response, play in both pathogen clearance and host resilience.
Myeloid cells, including monocytes, macrophages, dendritic cells (DCs), and neutrophils, are an essential part of the innate immune system. During the early stages of the immune response they are essential in both direct phagocytosis and destruction of pathogens, and activation of other immune cells by secretion of cytokines and chemokines [6–9] [10] [8]. They also have important roles in immunoregulation and tissue repair that are crucial in surviving an infection [11–13] [14] [15] [16]. If myeloid cells are not carefully controlled they can cause excessive inflammation that can lead to immunopathology and decreased host resilience [17–20] [10] [11] [21]. The importance of myeloid cells in the innate immune response against infection often makes them a target of pathogens. Microorganisms will manipulate the cells in order to survive, proliferate, and spread to other cells [7,22] [23]. One mechanism of both controlling pathogen replication and host inflammation is programmed cell death (PCD). There are many different PCD pathways of which apoptosis and autophagic cell death are largely non-inflammatory [6,16,24,25] and pyroptosis and necrosis are considered inflammatory [15,22,26] [27]. Apoptotic cell death is controlled by a caspase cascade, of which caspase-3 and caspase-7 are central players [28]. Pro- and anti-apoptotic proteins regulate apoptotic cell death. One key protein that regulates many types of PCD, but in particular apoptosis is B-cell lymphoma (bcl)-2. Infected myeloid cells will undergo cell death in order to control pathogen dissemination and replication [27] [29] [30] [31] [32]. Once the pathogen is cleared myeloid cells will undergo apoptosis to prevent excessive inflammation and immunopathology [24,25] [33] [34] [35] [26].
We have developed a mouse model with decreased myeloid cell death in order to understand the impact it has on pathogen clearance and host resilience to infection. This mouse expresses the anti-apoptotic protein human bcl-2 under the control of the CD68 promoter (CD68(bcl2)tg). This limits ectopic bcl-2 expression to cells of the myeloid lineage including monocytes, macrophages, neutrophils, and DCs. Bcl-2 primarily prevents apoptotic and autophagic cell death [25] [36] [37,38], thus making this an ideal model for studying the role of non-inflammatory myeloid PCD in pathogen clearance and host resilience.
This study uses two bacterial pathogens, L. pneumophila and S. pyogenes, that infect distinct sites, to examine how decreasing myeloid cell death impacts pathogen clearance and host resilience. While many studies have demonstrated mechanisms of myeloid cell death in in vitro infection models of S. pyogenes and L. pneumophila [39–44] [33], it remains unclear what role myeloid cell death plays during in vivo infection. L. pneumophila infection remains confined to the lung under most circumstances where it causes a severe pneumonia [45] [46]. This bacteria is found in contaminated water supplies, such as air-conditioning systems, and infects alveolar macrophages [45,47,48] [46]. It can cause complications in people with immunosuppression or other health problems, making it an important hospital-acquired infection [49] [50]. In mice, pulmonary infection can be mimicked using an intranasal infection model of L. pneumophila. S. pyogenes is a versatile pathogen that infects many areas of the body including the upper respiratory tract and soft tissue [51]. Invasive soft tissue infections can result in the systemic spread of bacteria causing a severe toxic shock syndrome (TSS) [35] [50] [29] [52]. To mimic this type of infection, we used a cutaneous infection model that rapidly causes a systemic infection. Using these two models we examined the roles that myeloid cell death play during both pulmonary and systemic infections.
L. pneumophila primarily infects lung macrophages, and actively delays apoptosis of these cells in order to replicate [53] [54] [55] [56] [31]. Infection with L. pneumophila induces an early pyroptotic cell death under the control of caspase-1 [57,58] [59] [60] [43] [61] [40] [62] [42]. There is also a caspase-11-dependent cell death that has shown in vitro to be independent of flagellin [40,57]. The later apoptotic cell death is at least partly also under the control of caspase-3, and as such can be inhibited by bcl-2 [63] [64]. Human macrophages do not express the Naip5 inflammasome that is triggered by L. pneumophila flagellin, so to better mimic the human infection we use a strain of L. pneumophila lacking flagellin A (ΔflaA). Deletion or inhibition of the pro-survival factor BCL-XL in macrophages results in decreased L. pneumophila replication [65], indicating that delaying PCD is a strategy that L. pneumophila may have for surviving in cells. When macrophages eventually undergo apoptosis this may enable the pathogen to spread to other cells. Unlike macrophages, DCs do not support the growth of L. pneumophila as they undergo rapid cell death in response to infection. When apoptotic cell death is blocked in DCs by overexpression of bcl-2 L. pneumophila will proliferate in DCs [27]. It was hypothesized that since DCs migrate throughout the body this DC cell death may be a mechanism to prevent spread of the bacteria.
Similar to L. pneumophila, S. pyogenes is thought to cause PCD by pyroptosis and apoptosis [29] [66]. The role that this PCD plays during infection is not well understood. The severe inflammatory response caused by S. pyogenes infection may be tempered by PCD in myeloid cells such as macrophages and neutrophils [67] [35] [68] [69]. S. pyogenes causes lysis of myeloid cells in a streptolysin O-dependent manner, that is thought to increase pathogen spread [68] [29] [52]. The PCD induced by S. pyogenes could be an immune evasion technique, and strains that cause less PCD have reduced virulence [29]. Therefore myeloid PCD may impact both pathogen clearance and host resilience to S. pyogenes infection.
This study explores in vivo the role that myeloid PCD plays during infection with two distinct pathogens. While the role of PCD in response to infection is well documented in vitro, less is known about what sort of balance is struck between controlling pathogen clearance and maintaining host resilience during in vivo infections. Both of the bacterial pathogens used in this study interact with myeloid cell death pathways, and this study focuses on the role that cell death controlled by bcl-2 plays during infection. Our data demonstrates that CD68(bcl-2)tg mice infected with either pathogen have decreased host resilience that occurs largely independent of any changes in pathogen clearance. This indicates that the rate of myeloid cell death is calibrated to preserve host resilience, and manipulations of this rate are detrimental to the host.
Bone marrow derived macrophages (BMDM) were infected with L. pneumophila. Human macrophages are more permissive to infection by L. pneumophila than macrophages derived from most common mouse strains [70]. Triggering of the NAIP5 inflammasome by flagellin, results in rapid pathogen clearance in mouse macrophages. In order to better recapitulate the infection that is seen in human macrophages for our mouse model we used a ΔflaA strain of L. pneumophila. To compensate for the lack of motility the bacteria are spun briefly onto cells. Twenty-four hours after infection with L. pneumophila there was an increased number of apoptotic cells as indicated by flow cytometry staining using a stain for activated caspase-3/7. (Fig 1A). Similarly there was an increase in caspase-3/7 activation in BMDMs infected with S. pyogenes (Fig 1B). This apoptosis was dependent on the dose of bacteria given. L. pneumophila induced the most cell death at a multiplicity of infection (MOI) of 20 (Fig 1C). S. pyogenes induced cell death at an MOI as low as .001 (Fig 1D).
Macrophages were derived from mice expressing human bcl-2 under the control of the CD68 promoter. Most of these macrophages constitutively express bcl-2 (Fig 2A). When BMDMs are exposed to the DNA damaging agent etoposide there is an increase in apoptotic cells as demonstrated by staining with annexin V and propidium iodide and flow cytometry analysis. Macrophages constitutively expressing bcl-2 had significantly decreased apoptotic cell death after exposure to etoposide (Fig 2B). As further evidence that ectopic expression of bcl-2 prevents etoposide-induced apoptosis cells were stained with the cell event reagent that is activated by activated caspase-3/7, and the DNA dye sytox that indicates permeable cells. The substantial caspase-3/7 activation induced by etoposide was abrogated in BMDMs derived from CD68(bcl2)tg mice (Fig 2C). The ectopic expression of bcl2 was effective in decreasing apoptosis at both high and low doses of etoposide (Fig 2D).
Constitutive expression of bcl-2 also decreased the cell death in BMDMs infected with L. pneumophila (Fig 3A), and S. pyogenes (Fig 3B) when examined with a fixable live/dead stain 24 hours after infection. The cell death prevented by bcl-2 during infection was largely apoptotic, as indicated by caspase-3/7 activation and cell permeability. Macrophages derived from CD68(bcl2)tg mice, infected for 24 hours with L. pneumophila, had decreased caspase-3/7 activation compared to macrophages derived from their littermate controls. This is shown by flow cytometry staining (Fig 3C), and quantified (Fig 3D). Likewise flow cytometry performed on macrophages infected with S. pyogenes indicated that both apoptotic and late apoptotic/necrotic stages were inhibited by ectopic expression of bcl-2 (Fig 3E and 3F).
Bone marrow, spleen and lymph nodes were examined for expression of transgenic bcl-2. The transgene was expressed primarily in CD11b+ cells in these organs (Fig 4A).
While the transgene is expressed in all myeloid cell subsets it is expressed at the highest level in F4/80+CD11b+ macrophages and lower in Ly6G+ neutrophils (Fig 4B). It is expressed at an intermediate level in CD11c+ MHC class II+ DCs, and not expressed in cells of the lymphocyte lineage (Fig 4B). In 12 week-old mice there is a slight increase in Ly6G+Ly6Clow neutrophils and Ly6G-Ly6Chigh inflammatory monocytes, while in mice aged 6–8 weeks the myeloid compartments are comparatively normal (Fig 4C). These changes are most noticeable in the spleens of older mice (Fig 4C). The spleens of 12 week old mice were slightly larger than their littermate controls (Fig 4D), therefore the total number of inflammatory monocytes and neutrophils was also higher. The spleen cellularity was comparable between CD68(bcl2)tg mice and littermate controls at 8 weeks. There was no indication of cancer development as the mice aged. The mice were healthy and lived a normal lifespan. However, given this slight accumulation of inflammatory cells as mice aged we used mice that were between 6–8 weeks of age for the infection experiments. This enabled us to focus primarily on the impact that decreased myeloid cell death had on infection, and not on homeostatic effects of constitutive bcl-2 expression.
Since the expression of bcl2 is expressed at varying levels in the different myeloid cell types, we examined the ability of the transgene to rescue DCs, neutrophils, and resident peritoneal macrophages from etoposide-induced cell death. Bone marrow-derived DCs (BMDCs) had a large increase in caspase-3/7 activation after treatment with etoposide, but this was greatly decreased with the presence of bcl-2 (Fig 5A). Neutrophils had an increase in late apoptotic or necrotic cells after treatment with etoposide based on staining for caspase-3/7 and sytox (Fig 5B). This PCD was also prevented by the ectopic expression of bcl-2 (Fig 5B). To show the effect of bcl-2 on a different subset of macrophages we used resident peritoneal macrophages. Etoposide-induced apoptosis and necrosis was decreased in peritoneal macrophages isolated from CD68(bcl2)tg mice, compared to littermate controls (Fig 5C and 5D).
In order to determine the impact that decreased myeloid cell death had on pathogen clearance and host resilience responses during pulmonary infection, mice were infected with the bacterial pathogen L. pneumophila. Mice infected intranasally with 1X106 L. pneumophila start to steadily lose weight 2 days after infection (Fig 6A). After infection CD68(bcl2)tg mice lose more weight and have a longer recovery time when compared to littermate controls (Fig 6A).
We next examined the specific cause of this decreased health status in infected CD68(bcl2)tg mice. One possibility was that the increased survival of myeloid cells, in particular DCs aided in the systemic spread of the bacteria. However, the infection was confined to the lung and there were no detectable bacteria in the spleen, liver or kidneys of infected mice (S1A Fig).
The decreased health of the CD68(bcl2)tg mice was therefore due to activity and responses in the lung. CD68(bcl2)tg mice infected with L. pneumophila had significantly increased damage in their lungs compared to littermate controls as soon as 48 hours after infection and this damage was even greater 96 hours after infection in transgenic mice (Fig 6B and 6C). The histological damage score (HDS) measured a number of parameters including area of damage and immune cell infiltration into the alveolar space. In order to determine if this increased lung damage was due to an increased bacterial burden in the lung of CD68(bcl2)tg mice we determined the colony forming units (CFUs) in the lungs from infected mice 1 hours, 48 hours, and 96 hours after infection. There was no statistically significant difference between genotypes in bacterial uptake in the lung or early proliferation as indicated by the bacterial counts at 1 hour after infection and 48 hours after infection. Interestingly, the lung damage preceded the small increase in bacteria counts in transgenic mice observed 96 hours after infection (Fig 6D). Ten days after infection both transgenic mice and littermate controls had cleared the infection (S1B Fig).
In order to determine the cause of the increased inflammation and lung damage, the pulmonary immune response to L. pneumophila was analyzed in CD68(bcl2)tg mice and littermate controls. Both transgenic mice and littermate controls had the same amount of immune cell infiltrate in the bronchoalveolar lavage fluid (BALF) 48 hours after infection (Fig 7A).
The number of infiltrating cells remained about the same in littermates 96 hours after infection, but continued to increase in the lungs of transgenic animals (Fig 7A). The increased infiltrating immune cells in transgenic mice were mostly neutrophils as determined by cytospin analysis (Fig 7B). Neutrophils migrated into the lungs of infected animals by 48 hours after infection, and their number was increased in transgenic animals compared to littermate controls 96 hours after infection. Surprisingly, other cell types had no significant increase in the lungs of infected CD68(bcl2)tg mice compared to infected littermate controls (Fig 7B). Macrophages increased in the lungs of both CD68(bcl2)tg mice and littermate controls at the same rate (Fig 7B).
In addition to looking at the immune cell infiltrate, expression levels of several cytokines and chemokines were measured in the lungs of infected animals (Fig 8A and 8B).
The peak of expression of these inflammatory genes in the lungs of both transgenic mice and littermate controls was 48 hours after infection, but the expression of several cytokines and chemokines were elevated in transgenic animals compared to littermate controls (Fig 8A and 8B). The large increase of IL-6 indicates an increase in inflammation in the lungs of these animals [71]. IL-1 receptor antagonist (Il-1rn) expression is also increased 48 hours after infection in transgenic animals. This gene is known to be important in resolution of lung inflammation and the expression is increased during times of acute inflammation [72] Also CXCL1 and CCL7 are known chemoattractants for neutrophils [73] and the increased expression of these chemokines at 48 hours may lead to the increase recruitment of the neutrophils by 96 hours. While there is a mild increase in bacterial load in lungs of infected CD68(bcl2)tg mice 96 hours after infection, the increased inflammation precedes this increase in bacterial burden. Therefore it seems that decreased myeloid cell death impacts host resilience, but does not affect the spread of the bacterial pathogen, and only slightly delays the clearance.
Given the impact that the CD68(bcl2)tg has on in vivo infection with L. pneumophila, we explored how L. pneumophila-induced PCD is affected by expression of bcl2 in relevant myeloid cell types. DCs from CD68(bcl2)tg mice had decreased PCD after infected with L. pneumophila, compared to littermate controls, as demonstrated with caspase-3/7 activation and cell permeability assays (Fig 9A and 9B). Alveolar macrophages isolated from CD68(bcl2)tg mice also had decreased L. pneumophila-induced early and late apoptosis compared to littermate controls (Fig 9C and 9D). While there is limited detection of PCD in neutrophils infected with L. pneumophila, the small amount of apoptosis observed is rescued by ectopic expression of bcl-2 (Fig 9E and 9F).
During infection with L. pneumophila there is an increase in cytokines detected in the lung. To investigate if this is caused by an increase of cells producing cytokines, or if the transgenic myeloid cells make more cytokines, infected DCs and macrophages were stained intracellularly for IL-6 and TNFα. There was not a significant increase in the percentage of DCs making either cytokine when CD68(bcl2)tg mice were compared to littermate controls infected with L. pneumophila (Fig 10A and 10B (top)). Also the mean fluorescent intensity was the same in DCs from transgenic animals or littermate controls (Fig 10B (bottom)), indicating that on a per cell basis the cytokine production is the same. Similar results were observed for macrophages. Cytokine production was equivalent between macrophages from transgenic mice and littermate controls, on a per cell and population basis (Fig 10C and 10D). We interpret this to mean that the increase in cytokines observed in the lungs of CD68(bcl2)tg mice is due to the increased number of inflammatory cells.
In order to determine how decreased myeloid cell death would impact the response to a systemic pathogen we infected CD68(bcl2)tg mice and littermate controls with S. pyogenes. In order to mimic a common course of severe infection with S. pyogenes, a cutaneous infection that causes systemic disease, the bacteria were injected subcutaneously. The bacteria spread from the skin into the blood stream within a few hours and colonize various organs. Mice constitutively expressing bcl-2 in myeloid cells had decreased survival after infection with S. pyogenes (Fig 11A).
However, there was no statistically significant change in the bacterial load in these mice (Fig 11B). The initial site of infection, the skin, had similar bacterial loads between transgenic mice and littermate controls 2 days after infection. In addition, the systemic spread was also similar as the spleen, and the liver had similar levels between the two genotypes (Fig 11C).
The decreased survival despite the unchanged pathogen clearance rates between the genotypes indicated a decrease in host resilience, and we examined a number of factors that could contribute to this. We examined the site of infection to determine if there were changes in inflammatory immune cell infiltration. There were increased infiltrating cells in transgenic mice into the area of bacterial infection (Fig 12A and 12B).
To get a clearer understanding of the types of immune cells infiltrating into the site of infection S. pyogenes was injected intraperitoneally. There was a significant increase in infiltrating cells into the peritoneal cavity of CD68(bcl2)tg mice 24 hours after infection with S. pyogenes (Fig 12C). However, the types of cells that responded to the infection did not change between the littermate and transgenic mice (Fig 12D). The responding cells were primarily neutrophils as identified by Ly6G and Ly6C expression in both genotypes of mice (Fig 12D), however the transgenic mice had more cells (Fig 12C).
Given the increase in different myeloid cell types during infection with S. pyogenes in CD68(bcl2)tg mice, the effect of ectopic expression of bcl2 on S. pyogenes-induced cell death was examined. Neutrophils (Fig 13A), DCs (Fig 13B), and resident peritoneal macrophages (Fig 13C) from CD68(bcl2)tg mice all had significantly decreased S. pyogenes-induced caspase-3/7 activation. During in vivo infection apoptotic cells are rapidly cleared [74], also the processes involved in cell isolation often cause cell death, therefore detection of apoptosis from ex vivo samples can be challenging. However, rapid assaying of whole blood cells from mice infected with S. pyogenes allows for the detection of cells undergoing PCD. S. pyogenes induces more than half of neutrophils in the blood to undergo PCD, but in CD68(bcl2)tg mice significantly less cells were undergoing PCD (Fig 13D). The same is true for blood monocytes (Fig 13E), and DCs (Fig 13F).
Since death caused by systemic infection with S. pyogenes is due to a toxic shock syndrome induced by systemic inflammation we examined the systemic responses. There were many systemic changes during infection. Most notably there were significant changes in the inflammatory cytokines TNFα, IL-1α, and IFN-γ (Fig 14A).
Transgenic mice sustained significant liver damage after infection as evidenced by increased levels of alanine aminotransferase (ALT) in the serum (Fig 14B). The decreased myeloid cell death in this systemic infection resulted in increased systemic inflammation that exacerbated the disease progression and led to decreased host resilience.
It seems likely that the increased systemic cytokine production in transgenic mice was due to the increased cellularity of infected mice. However, it is also possible that myeloid cells from CD68(bcl2)tg mice produce more cytokines. To investigate these non-mutually exclusive possibilities DCs and macrophages were stained intracellularly for IL-6 and TNFα. There was not a significant increase in the percentage of DCs making either cytokine when CD68(bcl2)tg mice were compared to littermate controls infected with S. pyogenes (Fig 15A and 15B (top)). Also the MFI was the same in DCs from transgenic animals or littermate controls (Fig 15B (bottom)), indicating that on a per cell basis the cytokine production is the same. Similar results were observed for macrophages. Cytokine production was equivalent between macrophages from transgenic mice and littermate controls, on a per cell and population basis (Fig 15C and 15D).
Pathogen clearance and host resilience are both important in surviving a given infection. While many studies have examined different mechanisms of pathogen clearance, recent studies have highlighted the importance that host resilience plays in survival of a given infection [2] [3] [4] [5]. The fact that cells of the myeloid lineage play important roles in both of these important processes [6–9] [10] [8,11–13] [14] [15] [16], we sought to determine how manipulation of myeloid cell death influenced the response to two bacterial pathogens. To do this we developed a mouse model that has decreased myeloid PCD. Ectopic expression of bcl-2 decreased PCD in response to numerous stimuli in myeloid cells of these mice (Figs 2, 3, 5, 9 and 13) [27]. We used two pathogens that interact with myeloid cells, but infect different areas of the host. The first pathogen, L. pneumophila, infects lung macrophages and remains confined to the lung [45,47,48] [46,75], while the second pathogen S. pyogenes spreads systemically [67] [35] [68] [69]. When myeloid cell death is prevented during systemic infection with S. pyogenes there is a significant decrease in host resilience (Figs 11, 12 and 14). Infected transgenic mice have decreased survival compared to littermate controls, and increased systemic inflammation. There is not a significant increase in bacterial load in the transgenic mice. The effects of decreased cell death during a pulmonary infection with L. pneumophila on host resilience are milder, but there is also an increase in inflammation in the lung. This increased inflammation precedes the small increase in bacterial load that is seen at the later stages of infection. Unlike infection with S. pyogenes both littermate and transgenic mice survive infection with L. pneumophila, indicating that the myeloid cell death has a greater impact on the systemic infection with S. pyogenes than on the pulmonary infection with L. pneumophila. In both infection models the primary cell type that is increased are neutrophils, which are known to cause tissue damage [34]. Also in both infection models the increase in inflammatory cytokine levels appears to be linked to the increased number of immune cells, and not by increased cytokine production on a per cell basis.
The lung is a delicate and essential organ thus the response to lung infections is particularly challenging, in that pathogen clearance must be balanced with host resilience mechanisms. As L. pneumophila is confined to the lung, we used this as a model of lung infection. Decreased cell death of lung myeloid cells leads to increased inflammation and immune cells infiltrate into the lung in response to this pathogen. Interestingly, this increased inflammation precedes any increase in bacterial load in the lung. There is a small, but statistically significant increase in the bacterial load in the lungs of transgenic mice compared to littermate controls, however surprisingly the decreased myeloid cell death did not lead to systemic spread of the bacteria. Transgenic mice had a decrease in health status compared to littermate controls as indicated by rapid and sustained weight loss. However, both genotypes of mice were able to survive and eventually clear the infection. The mild effect of decreased myeloid cell death observed upon infection with L. pneumophila may be for several reasons. Many different types of cell death are caused by infection with L. pneumophila, including pyroptotic cell death. Transgenic expression of bcl-2 has limited influence on pyroptotic cell death. However, the apoptotic cell death observed in later stages of infection of macrophages and also seen in DCs is profoundly affected (Fig 3) [27]. It could also be that the lung is able to cope using alternative resilience mechanisms, such as tissue repair pathways, with a threshold level of inflammation. It is likely, given the fragility and importance of the lung that there are multiple pathways that serve to protect this organ from damage. The increase in neutrophils observed in the transgenic mice could be a cause for the decreased health status, but pro-resilience pathways are able to maintain tolerance to the infection and survival is not impacted.
In order to determine what role myeloid cell death plays in pathogen clearance and host resilience during an infection that can spread systemically we infected mice with S. pyogenes. In our model this subcutaneous infection rapidly spreads systemically. This enabled us to examine what role myeloid cell death plays at the site of infection, in spread of the infection, and in the systemic response to infection. In contrast to what was observed after lung infection with L. pneumophila, where both genotypes were able to survive the infection, transgenic mice infected with S. pyogenes had significantly decreased survival compared to littermate controls. Interestingly, the decrease in myeloid cell death in the transgenic mice did not significantly influence pathogen spread and clearance. However, it did cause a significant increase in inflammation both at the site of the infection, and systemically. There was also an increase in liver damage in transgenic mice. These data suggest host resilience to systemic S. pyogenes infection is compromised by decreased myeloid cell death due to an excessive inflammatory response. During infection with S. pyogenes the bacteria spread systemically and the increase in inflammation at multiple sites may overwhelm other host resilience mechanisms. This indicates that PCD of myeloid cells is an essential disease resilience mechanism during systemic infections with S. pyogenes.
This study provides new insight into the roles that myeloid cell death plays in response to bacterial infections. While the decreased myeloid cell death caused minor inflammation in the lung, most likely other pathways were able to compensate for the increase in inflammation and prevent a large decrease in host resilience. However, during a systemic infection decreased PCD and consequently increased inflammation overwhelms the host and leads to decreased resilience as measured by survival. Interestingly, in both infections the main cell type that is increased are neutrophils. This increase may be caused either directly by the expression of bcl-2 in these cells or by the increase in other myeloid cell types that recruits neutrophils to the sites of infection. Neutrophils are known to cause damage in many different disease models [76–79], and they are the probable cause of the decreased host resilience with these infection models. As neutrophils are also essential for pathogen clearance depletion of them could result in decreased host-mediated damage, but the increase in bacteria will most likely cause an increase in pathogen-mediated damage. These data demonstrate how tightly regulated PCD in myeloid cells is, and how important it is for host resilience. Disruption of this regulation changes an infection that is potentially survivable to one that has a high rate of lethality. These findings can be applied to models of sepsis and other infections where host resilience processes are important factors in survival [67, 80].
A transgenic construct was made using the CD68 promoter and regulatory sequences as described by Gough et al. [81] and human bcl-2 cDNA was cloned into the XbaI restriction digest sites. The transgenic animals were made using standard methods. Four initial founders were selected based on screening by Southern Blot. The line used in this study Tg535 (bcl2)rm had the highest expression based on intracellular staining for the bcl-2 protein. The mice were initially on a 129/J background, but were backcrossed greater than 20 times to C57BL/6J mice. Animals were bred and maintained in a specific pathogen free facility (SPF)
All procedural protocols involving mice were approved by the appropriate Institutional Animal Care and Use Committee at the site where the work was performed. In Vienna all of the experiments have been approved by the Vienna University of Veterinary Medicine institutional ethics committee and performed according to protocols approved by the Austrian law called BMWF 68.205/0032-WF/II/3b/2014. General condition and behavior of the animals during the experiments was controlled by FELASA B degree holding personnel. The animal protocol number approved by University of Veterinary Medicine institutional ethics committee on 2/28/11 is 535233. Brown University adheres to the “U.S. Government Principles for the Utilization and Care of Vertebrate Animals Used in Testing, Research, and Training”, “PHS Policy on Humane Care and Use of Laboratory Animals”, “USDA: Animal Welfare Act & Regulations”, and “the Guide for the Care and Use of Laboratory Animals”. The University is accredited by the Association for Assessment and Accreditation of Laboratory Animal Care International (AAALAC). Brown University’s PHS Assurance Number: A3284-01 and it expires July 1, 2018 The USDA Registration Number is 15-R-0003. Brown University IACUC approved on October 8, 2013, and the animal protocol number is 1308000011.
The L. pneumophila strain used in this study was JR32ΔflaA provided by Craig Roy (Yale University School of Medicine, New Haven USA). The S. pyogenes strain used in this study was serotype M1 strains ISS3348 [82]. For L. pneumophila infections mice were infected intranasally with 1X106 L. pneumophila in 40 μl sterile saline. Bacteria were grown from a heavy patch overnight in autolyzed yeast extract broth. For S. pyogenes infections mice were infected with 1X107 bacteria in 50 μl subcutaneously in the left flank. A single colony of S. pyogenes was grown overnight in Todd Hewitt Broth (THB). This overnight culture was diluted into 100 ml of THB and grown for approximately 6 hours until in the log phase. Bacterial amounts were estimated using OD600 readings, and confirmed by colony forming unit (CFU) quantification. To ensure proper infectivity of cells when using the non-motile strain of L. pneumophila, bacteria were inoculated using a spinfection protocol. Bacteria were added to the cells in antibiotic-free media, and plates were spin for 5 minutes at 1200 rpm (290 RCF).
To determine the inoculation load and the bacterial loads in various organs at indicated time points after infection CFU assays were done. At several time points after L. pneumophila infection lungs, spleens, and livers from infected mice were homogenized in 2 ml of sterile water using an electronic homogenizer (Polytron 2100). One-hundred μl of serially diluted homogenate was plated on charcoal yeast extract (CYE) plates. Colonies were quantified after 2–3 days of incubation at 37 degrees C. To determine bacterial counts in skin and organs from mice infected with S. pyogenes, organs were homogenized in PBS, as described for L. pneumophila infection. Serially diluted homogenate was plated on blood agar plates and incubated overnight at 37 degrees C. Bacterial colonies were quantified the next day.
To collect bronchoalveolar lavage fluid (BALF), the trachea was exposed, and a flexible tube placed on a 23-gauge cannula was inserted into the trachea. The lung was rinsed with 1ml PBS using an attached syringe. The viability of the isolated cells was determined by Trypan blue exclusion, and the cells were counted in a hemacytometer. For isolation of cells from lungs, they were perfused with 20 ml of PBS. The lung tissue was diced into small pieces and incubated for 45min at 37 degrees C in 4ml of media containing collagenase and DNAse. Afterwards, digested lung tissue was made into a single cell suspension by passage through a cell strainer. After centrifugation the cells were re-suspended in 4ml 40% Percoll/Roswell Park Memorial Institute (RPMI) and carefully layered over 4ml of 80% Percoll/PBS. The formed gradient was centrifuged at RT for 20min at 652rcf (Eppendorf 5810R centrifuge) with minimal acceleration and deceleration. Cells assembled in the interphase were collected, and washed with 10ml RPMI media containing 5% fetal calf serum.
All steps of staining were performed on ice unless mentioned otherwise. All staining was done in V-bottom 96-well plates. Isolated cells were pelleted by centrifugation and washed twice with PBS containing 1% BSA and 0.001% w/v sodium azide (FB). Cells were re-suspended in FB containing rat anti-mouse CD16/CD32 antibodies (1:100) and incubated for 10min to block the Fc receptors, followed by two washes Cells were re-suspended in FB containing the desired antibodies for the surface staining in an appropriate dilution determined by titration. After 20min incubation, cells were washed twice with FB. Antibodies used included CD11c (M1/70) Biotin, Ly6C (AL-21) V450 (BD Biosciences), Ly6G (1A8) FITC, CD11c (N418) PE, F4/80 (BM8) APC, and 570 Fc receptor block (93) (Biolegend). In addition, the fixable viability dye eFluor 780 (eBioscience) (room temperature staining) and Streptavidin Brilliant Violet were used. Apoptotic cells were detected using either Annexin V (eBioscience) and PI (Sigma), or the CellEvent reagent and sytox (Thermofisher). To identify specific cell types these stains were combined with the cell surface markers described above. Stained cells were acquired on a FACSAria III cell sorter, equipped with a 488nm (blue) laser, a 633nm (red) laser, and a 407nm (violet) laser. Flow-cytometry was also performed on the Attune NxT. Finally, collected data were analyzed by FlowJo Software (Tree Star, Inc).
For histological analysis, perfused lungs or excised skin were placed in 1% paraformaldehyde (PFA) overnight at 4 degrees C. The samples were transferred into 70% ethanol and the samples were processed in the Excelsior tissue processor. The samples were embedded in paraffin blocks and 5 μMsections were made with a microtome (Leica). Standard staining protocols were used for Hematoxylin and Eosin staining of rehydrated sections. All histological samples were assessed twice in a blinded manner. To determine the histological damage score (HDS), the following criteria were considered: the frequency (none = 0; sporadic = 0.5; few = 1; many = 2, excessive = 3) of neutrophils, macrophages and lymphocytes and their location (perivascular, peribrochial, parenchymal, subpleural, and alveolar lumen), as well as the activation of the pleura pulmonalis (none = 0; little = 0.5; medium = 1; strong = 2, excessive = 3). The points were added together and a maximum of 48 points per sample could be achieved. To assign the samples into 10 HDS groups (0–1 = 1; 1.1–2 = 2; 2.1–3 = 3; 3.1–4 = 4; 4.1–5 = 5; 5.1–6 = 6; 6.1–7 = 7; 7.1–8 = 8; 8.1–9 = 9; 9.1–10 = 10), the obtained sum was divided by 4.8. A low HDS value indicates minor tissue damage, whereas high HDS values mark increased damage of the lung. In addition, also the frequency of neutrophils, macrophages, and lymphocytes as well as the level of pleura activation was analyzed separately. Skin sections were also scored in a blinded manner.
To determine the cytokine and chemokine protein levels, the Mouse Th1/Th2/Th17/Th22 13plex FlowCytomix Multiplex assay was performed according to the manufacturer’s instructions. RNA from tissues was purified using Reliaprep RNA Miniprep System (Promega). Quantitative PCR was performed on a Roche LC96 using standard methods. Alanine transaminase in the serum was measured with the use of a colorimetric kit (Cayman Chemical) according to manufacturer’s instructions.
|
10.1371/journal.ppat.1001298 | Uropathogenic Escherichia coli P and Type 1 Fimbriae Act in Synergy in a Living Host to Facilitate Renal Colonization Leading to Nephron Obstruction | The progression of a natural bacterial infection is a dynamic process influenced by the physiological characteristics of the target organ. Recent developments in live animal imaging allow for the study of the dynamic microbe-host interplay in real-time as the infection progresses within an organ of a live host. Here we used multiphoton microscopy-based live animal imaging, combined with advanced surgical procedures, to investigate the role of uropathogenic Escherichia coli (UPEC) attachment organelles P and Type 1 fimbriae in renal bacterial infection. A GFP+ expressing variant of UPEC strain CFT073 and genetically well-defined isogenic mutants were microinfused into rat glomerulus or proximal tubules. Within 2 h bacteria colonized along the flat squamous epithelium of the Bowman's capsule despite being exposed to the primary filtrate. When facing the challenge of the filtrate flow in the proximal tubule, the P and Type 1 fimbriae appeared to act in synergy to promote colonization. P fimbriae enhanced early colonization of the tubular epithelium, while Type 1 fimbriae mediated colonization of the center of the tubule via a mechanism believed to involve inter-bacterial binding and biofilm formation. The heterogeneous bacterial community within the tubule subsequently affected renal filtration leading to total obstruction of the nephron within 8 h. Our results reveal the importance of physiological factors such as filtration in determining bacterial colonization patterns, and demonstrate that the spatial resolution of an infectious niche can be as small as the center, or periphery, of a tubule lumen. Furthermore, our data show how secondary physiological injuries such as obstruction contribute to the full pathophysiology of pyelonephritis.
| When bacteria such as uropathogenic Escherichia coli (UPEC) infect a living kidney, they face numerous physiological challenges such as the flow of urine. Bacteria need to attach themselves to the epithelial linings of the kidney to withstand this flow. In this work we use a live animal imaging model to study how UPEC colonize a living kidney despite the physiological challenges they face. We show that P and Type 1 fimbriae, two of the most well described UPEC adhesion factors, work together to promote successful bacterial colonization. P fimbriae mediate binding between the bacteria and the epithelial cells lining the tubules, while Type 1 appears to play a role in inter-bacterial binding and biofilm formation in the center parts of the lumen. The heterogeneous bacterial community which filled the tubule was subsequently shown to effect nephron filtration and resulted in a total loss of filtrate flow i.e. obstruction. This work demonstrates the interplay between the bacterial and host aspects, indicating how factors such as filtration may affect bacterial adhesion and vice versa. It also highlights the multifactorial basis of kidney infection, demonstrating how physiological injuries such as obstruction may contribute towards the full pathophysiology of pyelonephritis.
| Bacteria colonizing the mammalian host face numerous dynamic challenges. In the urinary tract, this is exemplified by the shear stress of urine flow. This stress can vary considerably; in the bladder, the flow changes dramatically upon voiding whereas in the renal tubules more subtle variations occur as the body regulates renal function. Uropathogenic E. coli (UPEC), the major causative agent of urinary tract infections (UTI) have evolved mechanisms by which to overcome these challenges. For successful colonization in this hydrodynamically challenging environment, bacterial attachment to the epithelium is essential. For UTI caused by UPEC, major roles have been ascribed to the attachment organelles Type 1 and P fimbriae [1]. While both are considered important [2], [3], [4], [5], their definitive role in the progression of kidney infection, pyelonephritis, in vivo has never been clearly defined.
Type 1 and P fimbriae bind to mono-mannose and globoseries glycosphingolipids, respectively [6], [7]. Lack of mono-mannose rich uroplakin on renal epithelia has previously implied a limited role for Type 1 fimbriae in kidney infection, whereas these fimbriae have been strongly linked to many aspects of bladder infection [2]. The P fimbrial operon has been shown to be over-represented in clinical isolates from pyelonephritic patients, yet has not been demonstrated to be essential for disease [8]. Only subtle roles for P fimbriae-mediated adherence have been described in uroepithelial cell culture models [3], and investigations of its role in ascending infection models have yielded inconsistent and often conflicting results [9], [10], [11], [12]. The conflicting data may reflect the limited spatial and temporal resolution in previously used model systems for in vivo infections, highlighting the need to address the problem using alternative techniques.
We have recently employed live animal multiphoton microscopy (MPM) to visualize tissue dynamics during renal bacterial infections [13], [14], [15]. These real-time visualization studies of the infection process can be performed under the influence of all physiological factors, including the vascular, nervous, immune and hormonal systems [14], [16], [17], [18]. Previous study of the acute pathogenesis of kidney infection using live imaging revealed a very rapid bacterial colonization process accompanied by major alterations of tissue homeostasis [19]. During the first 3–4 h, local tissue oxygen tension (PO2) plummeted to 0 mm Hg, followed by clotting and cessation of blood flow in peritubular capillaries. The ensuing ischemia was established as a local innate immune defense mechanism, protecting the organism from systemic bacterial spread and sepsis [20]. Bacterial containment at the infection site resulted in a focused immune cell infiltration, which after 24 h resulted in bacterial clearance from the injection site and localized tissue edema. The tissue morphology seen at 24 h was comparable to that seen in renal abscesses in ascending infections 4 days post-infection [19].
In the live animal model, spatial-temporal control of the infection is achieved by slowly infusing bacteria into superficial proximal tubules of surgically exposed kidneys in anesthetized rats. The micro-infusion technique and the length of anesthesia are two important factors in the choice of rats in the present experimental model. Rats have previously been used successfully in experimental ascending UTI [21], [22], [23] models and do express the P fimbriae Galα1-4Galβ receptor for UPEC attachment in the kidney [24]. In our previous work immediate visualization of the infection site revealed that only a few bacteria succeed in initially withstanding renal filtrate flow and colonize the tubule epithelium [19]. The attached bacteria multiplied extensively, filling the entire tubule lumen within 4–5 h. This implies that UPEC can, and do, express sufficient adhesion factors to withstand the mechanical stresses in vivo. Here, we investigate the bacterial adhesion mechanisms that enable bacteria to withstand the obstacles to early stage kidney colonization, and define previously unknown synergistic functions of P and Type 1 fimbriae under dynamic in vivo conditions.
The initial stages of renal filtration occur in the glomerulus, which consists of capillary tufts surrounded by the Bowman's capsule (Figure 1A). Tissue biopsies from patients with pyelonephritis demonstrate that infection rarely ascend into Bowman's capsule. The disease is therefore characterized as a tubulointerstitial rather than a glomerular disorder [25]. Colonization of the proximal tubule segments of the nephrons is promoted by efficient UPEC attachment to the microvilliated cuboidal epithelia. We hypothesized that the resistance of a functional glomerulus to infection may therefore be due to an inability of bacteria to bind to Bowman's capsule flat squamous epithelium under dynamic conditions. To address this question LT004, a derivative of the prototypic UPEC strain CFT073 [26] expressing GFP+ [27] from a single chromosomally inserted gene [19] was slowly infused directly into the Bowman's space of superficial glomeruli in Munich–Wistar rats. As with previous reports of the microinfusion model, multiphoton imaging showed that the vast majority of infused bacteria are immediately flushed out by the filtrate flow, leaving only a few to initiate colonization [19]. Distinct green fluorescence, conformally lining the epithelia of Bowman's capsule, was observed 2 h after infusion, suggestive of bacterial attachment. Extensive multiplication occurred over the following hours, with a mat-like bacterial community being formed both within Bowman's capsule and in the earliest (S1) portion of the proximal tubule (Figure 1B).
Intravenous injection of a fluorescent large molecular weight dextran (a blood plasma marker) revealed that the infection was accompanied by the anticipated decrease [20] in peritubular capillary flow (Figure 1B). The glomerular capillaries and adjacent arterioles however appeared more robust, with blood flow continuing hours after the shutdown of peritubular capillaries. Within 8 h blood flow in glomerular capillaries also did shutdown, as noted by their lack of red dextran and/or flowing erythrocytes (Figure 1B). At these later time points (7–8 h) faint red staining, originating from the 500 kDa dextran vasculature marker, was observed within the S1 segment of the proximal tubule indicating a breakdown in the glomerular capillary filtration barrier (Figure 1B, 8 h). These data indicate that UPEC express the appropriate attachment organelles to mediate colonization of the glomerulus despite the shear stress of filtrate flow. This implies that the epithelial composition is not the defining factor for lack of glomerular colonization during pyelonephritis.
Genetic analysis of the expression pattern of Type 1 and P fimbriae in carefully controlled in vitro experiments has shown that an individual E. coli bacterium express only one fimbriae type at a time due to co-regulation of the fimbriae operons [28], [29], [30]. To analyze the gene expression patterns of Type 1 and/or P fimbriae by UPEC colonizing the renal tubule, the spatial-temporally controlled micro-infusion model was used. Tissue infused with LT004 8 h previously was carefully excised and bacterial mRNA was isolated. qRT-PCR analysis revealed substantial expression of both major fimbriae structural proteins PapA_2 and FimA (Figure 2A). This suggested a bacterial population with heterogeneous expression of adhesion organelles at this early stage of infection.
We then analyzed the role of the well-known UPEC adhesion factor P fimbriae during in vivo colonization. To do this we used strain ARD41, a GFP+ expressing derivative of CFT073 containing defined mutations in both copies of papG (Table 1). ARD41 therefore lacked P fimbriae mediated attachment, but could still express functional Type 1 fimbriae. Phenotypic analysis of this strain demonstrated the expected erythrocyte and yeast agglutination pattern (Figure 2B), and a significant reduction of bacterial binding to A498 renal epithelial cells in vitro (Figure 2C). In an ascending model ARD41 was able to ascend and infect the kidneys, though a reduced number of bacteria, shown as CFU/g tissue, was demonstrated as compared to the UPEC strain LT004 (CFT073 GFP+) (Figure 2D). Real-time analysis of the renal infection process was performed using multiphoton microscopy. These dynamic in vivo imaging experiments revealed that this strain, lacking P fimbriae, colonized less efficiently as it only established infection in approximately 33% of infusions, as compared to an approximate 95% success rate for LT004. In successful ARD41 infections, the absence of P fimbriae delayed colonization of the tubule to 7–8 h post infusion (Figure 3B) in comparison to the wt strain, which showed colonization within 2 h of infection (Figure 3A). We have reported previously the shutdown of local peritubular capillaries as a response to tubular infection [20]. In ARD41 infections the vascular shutdown response, relative to bacterial load, was slower than that seen for LT004, but yet visible (Figure 3B 10 h arrow). The spatial-temporal precision of this infection model allows for the exposed kidney to be returned to the peritoneal cavity and re-analyzed on subsequent days. MPM-analysis of the infection site 24 h post-infusion showed that bacteria had been cleared, leaving cortical edema and extensive tissue destruction (Figure 3B 24 h), the same outcome as seen for LT004 infections (Figure 3A 24 h).
The role of P fimbriae in early stage epithelial colonization was further strengthened by experiments using an isogenic set of E. coli K-12 strains that did and did not express P fimbriae. By complementing a GFP+ expressing E. coli strain W3110 (ARD42) with a plasmid encoding the pap operon, we obtained isogenic strains that did (ARD43) and did not (ARD42) express P fimbriae (Table 1). Their respective P fimbriae phenotype was verified in erythrocyte agglutination assays and Type 1 fimbriae expression was confirmed with a mannose independent yeast agglutination assay (Figure 2B). These strains also lacked many other known virulence factors such as the α- hemolysin toxin (Hly). To test the relevance of these K-12 strains in urinary tract infection, 108 CFU was inoculated into bladders in an ascending model of UTI. Both strains were found to ascend to and infect the kidney, though lower bacterial numbers were observed in the tissue as compared to UPEC strains (Figure 2D).
In the live imaging model ARD42, which lack P fimbriae, showed delayed tubular colonization kinetics as it was only able to initiate colonization 6–7 h after infusion (Figure 3C,3F). The colonization kinetics were rescued in strain ARD43, which over-expresses plasmid encoded P fimbriae. ARD43 showed greater initial colonization than ARD42 (Figure 3D) and populated the tubules with similar kinetics to LT004 (Figure 3A). The vascular shutdown following infection with ARD43 was however delayed in comparison to infection with UPEC strain LT004 (Figure 3D). This difference may stem from a lack of expression of other virulence factors such as Hly, known to effect vascular shut-down kinetics [19]. At 24 h post-infusion bacterial clearance, edema formation and tissue damage was observed irrespective of P fimbriae expression, implying that the delayed colonization kinetics of ARD42, lacking P fimbriae, is overcome within 24 h (Figure 3C, D 24 h). These results corroborate the findings for ARD41 that P fimbriae enhance the early stage of tubule colonization. They also indicated that while E. coli K-12 strain does not elicit the same rapid host response as UPEC, it can cause inflammation and edema over 24 h.
Dynamic imaging of the infection revealed a striking feature only observed in the two strains lacking the PapG tip adhesin. Both ARD41 and ARD42 were observed being ‘flushed’ through the tubule by the filtrate flow. Video S1 shows a large bacterial mass moving in a tubule, indicating that bacteria lacking P fimbriae appear to be more susceptible to filtrate flow. Figure S1 shows a tracing of this video showing the approximately 70 µm path the bacterial mass moves during the 70 s duration of the video. In our studies, we observed this event 3–4 times in multiple animals during independent infections. Due to the speed and unpredictability of this event the possibilities of capturing it at a certain time point is limited and therefore these numbers are probably under-representative of the occurrence of this event. Together these data suggest that expression of P fimbriae provides a fitness advantage in vivo, aiding bacterium in withstanding the filtrate flow and enhancing colonization during the first hours of infection.
A similar analysis was performed to investigate the role for Type 1 fimbriae in early colonization. An insertion deletion was introduced into the fimH of the GFP+-expressing derivative of CFT073, strain LT004 (Table 1). Successful inactivation, demonstrated in erythrocyte and yeast agglutination experiments (Figure 2B) suggests that the resulting strain ARD40 lack the ability to bind via Type 1 mediated attachment. In motility assays, LT004 showed an average swimming diameter of 26±1.5 mm and ARD40 22±1.7 mm (p = 0.116), indicating that the absence of the FimH tip adhesin did not influence bacterial motility. ARD40 was also able to infect the kidneys in the ascending model, but again showed lower numbers of bacteria in the tissue than LT004 (Figure 2D). Dynamic in vivo imaging showed that ARD40 colonized the tubule at a level comparable to LT004, and the pathophysiology at 24 h was similar to that induced by the other strains (Figure 3E). Type 1-negative bacteria did, however, display a unique feature. While bacteria efficiently colonized along the tubular epithelium, the bacterial density in the centre of the lumen was dramatically reduced (Figure 3E, 5 h). Hollow “bacterial tubes” appeared to form (Figure 3G), suggesting that in areas where bacteria have no epithelium on which to adhere they have difficulty maintaining themselves.
In perfused environments, microbial communities are established via a process known as “self-immobilization” [31], [32]. Sessile biofilms are formed as bacteria embed themselves in an endogenously formed matrix. This compact community, consisting of organisms' adherent to each other and/or a surface, provides extraordinary resistance to hydrodynamic flow shear forces. The FimH adhesin has been shown to be instrumental in biofilm formation by E. coli K-12 under both static and hydrodynamic growth condition in vitro [33], [34]. This suggested that the Type 1 fimbriae may confer on UPEC the ability to form biofilm that opposes bacterial clearance from the central part of the tubule lumen. The biofilm forming ability of strains included in this study was visualized and quantitated in vitro using polystyrene microtiter plate assays (Figure 4A, B). The wt UPEC strain LT004 as well as the papG mutant strain ARD41 both formed low, yet notable amounts of biofilm. This was in contrast to the fimH mutant strain ARD40, whose biofilm-forming capacity was significantly reduced. As expected, the E. coli K-12 strain ARD42 (expressing Type 1 fimbriae) showed robust biofilm formation, which was unaffected by P fimbriae complementation (ARD43). Salmonella enterica serovar typhimurium, a known biofilm forming strain, was included as a positive control. A Western blot revealed that all strains expressed RpoS, the master regulator of general stress responses which has previously been shown to effect biofilm formation [35] (Figure 4B). These results suggest that FimH does play a role in UPEC biofilm formation and may imply that the lack of ARD40 colonization of the tubule center is related to this strain's inability to mediate inter-bacterial binding and biofilm formation in this perfused micro-environment.
One may envisage that formation of dense bacterial communities within the tubular lumen would influence renal filtrate flow. The effect of infection on filtrate flow was analyzed by systemic injections of small molecular weight red fluorescent dextran. After filtration by the glomerulus, dextran appears within the tubule lumen where it can be used to visualize filtrate flow [36]. Four hours after micro-infusion of LT004, a bolus of dextran was delivered intravenously until fluorescence in Bowman's space reached saturation. Within seconds of infusion, red dextran appeared within peritubular capillaries (Figure 5A, Video S2). A representative animal is shown in Figure 5. In non-infected nephrons, dextran was rapidly filtered and passed swiftly through the tubule lumens. This is visualized in Figure 5A (I), with quantification of this tubule's fluorescence shown in Figure 5B. Analysis of the infected nephron within the same field-of-view revealed lowered peak intensity, indicating a degree of obstruction and reduced glomerular filtration (Figure 5A II; Figure 5B). Repeating the experiment 8 h post-infusion, when the entire tubule lumen of the infected nephron was colonized by bacteria, showed that filtrate flow was completely obstructed and peritubular capillary blood flow was shut-down (Figure 5C II; Figure 5D, Video S3). In contrast, non-infected neighboring nephrons still displayed filtration (Figure 5C I; Figure 5D, Video S3). Similar experiments were performed using the isogenic mutant strains (data not shown). Obstruction was observed in these infections, but variability between animals prevented satisfactory statistical quantification and we were therefore unable to note any significant variation in the early phases of obstruction.
In addition to obstruction, local vascular leakage occurred as the bacterial infection progressed. Loss of epithelial membrane barrier function could be identified 4 h post-infusion, when dextran was found leaking into the epithelial cells of the infected tubule (Figure 5A, 80 s, arrow and Video S2). Careful inspection of data from the dynamic imaging (Video S2 and Figure 5E, which is an inset from Figure 5A 7 s) shows that leakage appears to start from the basolateral side of the epithelium. This is in contrast to the neighboring, non-infected nephrons, which maintain their epithelial barrier function at this stage of infection (Figure 5A, 80 s, arrowhead). At 8 h, non-infected neighboring tubules also show some signs of epithelial barrier function breakdown, likely linked to an ischemic injury [20] (Figure 5C, Video S2).
The ability to monitor real-time progression of bacterial infections in a living animal is providing a new foundation for microbial pathogenesis research. As the experiment is performed within the live organ, the roles of bacterial virulence factors can be studied in vivo in the presence of all physiological parameters. While recently reported real-time live animal infection models hold numerous advantages [37], [38], the experimental models can be very complex. One concern can be the route of delivery of the infecting agent. Experimental control over the spatial and temporal aspects of the infection is of utmost importance to allow for time-resolved studies. In the MPM model presented here bacteria are microinfused directly into the kidney nephron. This evidently differs from the natural ascending route of infection but is essential to allow for imaging of the infectious time course starting from the first host-pathogen interaction. In this, as well as our previous studies [14], [19], [20] we have shown that the majority of infused bacteria are immediately flushed from the tubule and the visualized infection stems from the very few bacteria that initially bind and adapt to the tubule microenvironment.
The findings presented here suggest that UPEC's attachment organelles, P and Type 1 fimbriae, act synergistically to facilitate bacterial colonization in the face of challenges such as renal filtrate flow (see an animated summary in Video S4). In ascending UTI models we and others [10] have shown that PapG, the tip adhesin of P fimbriae, is not essential for renal infection with CFT073.The temporal control of the MPM approach used here revealed that expression of this adhesin enhances the colonization kinetics during the very early, first hours of infection. In the living animal model P fimbriae appear to promote epithelial attachment and resistance to filtrate flow, facilitating early bacterial multiplication prior to the onset of ischemia and infiltration of immune cells [20]. These findings suggest P fimbriae may function as a ‘fitness factor’ in much the same way as siderophores. Iron sequestering sideophores have been annotated as fitness factors because while their expression is not essential to bacterial virulence, it is shown to be advantageous [39], [40]. Considering P fimbriae as a fitness factor in vivo may rationalize it's over-representation in clinical UPEC isolates as well as explaining its presence in many, but not all, pyelonephritis isolates [8].
Contrary to the widespread view that Type 1 fimbriae plays its primary role in cystitis, our data indicates it also plays a key role in kidney infections. Our data are the first to indicate that Type 1 fimbriae facilitate inter-bacterial adhesion and biofilm formation, allowing bacteria to maintain themselves within the center of the tubule lumen. A lack of FimH mediated inter-bacterial binding reduced the bacteria's ability to colonize the center of the tubule, where they have no epithelium on which to attach. This hypothesis is reinforced by previous findings which demonstrate the importance of the FimH adhesin in in vitro biofilm formation under shear stress [34]. In strain ARD41, which expresses Type 1 fimbriae but lacks PapG, bacterial colonization was at later stages of infection visualized in the center of the tubule lumen, away from the epithelium (See Figure 3B, 8 h). This pattern shows similarities to anaerobic upflow systems where microbes form dense spherical biofilm aggregates within the liquid by a process known as self-immobilization [31]. These aggregates consist solely of microorganisms, which themselves produce the matrix in which they are embedded [31], [32]. Self-immobilization was proposed by Sonnenburg and co-workers as the mechanism allowing bacterial communities to establish themselves independently of epithelial attachment in perfused environments such as the lumen of the intestine [32]. It may be that similar mechanisms operate within the tubule lumen, with FimH facilitating inter-bacterial adhesion by binding to mannose present within the LPS of neighboring bacteria [41]. This inter-bacterial binding or biofilm formation would facilitate luminal colonization, while a planktonic form of bacteria would be rapidly removed by the filtrate flow. UPEC strains have previously been shown to form biofilms on catheters as well as on and in the bladder epithelium [42], [43], whereas our data suggests a new role of UPEC biofilm in renal infection. The relevance of this biofilm formation in human pyelonephritis remains to be investigated. The synergistic action of P and Type 1 fimbriae we observe supports a previous concept that bacteria generate a heterogeneous fimbriae population to cope with the unpredictability of their environmental niche [28]. Our data further implies that these niches may not only be as large as between the kidney and bladder but as small as between the centre and periphery of a single tubule lumen.
The role of shear stress on bacterial adhesion and colonization is becoming increasingly appreciated. Recently a relationship between the E. coli FimH protein and shear stress was reported [44], [45], [46], [47], showing that the binding strength of FimH to mannosylated surfaces is enhanced by shear. This is mediated via a force-enhanced allosteric catch-bond, which functions via a finger-trap like β sheet twisting mechanism [44]. At a shear of 0.02 dynes/cm2 the binding strength of FimH was considered weak whilst as the shear increased to 0.8 dynes/cm2 this binding become stronger [47]. Shear stress in the renal proximal tubules has been reported to be 0.17 dynes/cm2, though the fluctuating nature of fluid flow as well as tubular absorption and secretion may imply a degree of variability in this rate [48]. The apparent low level of shear stress present in the renal tubule may mean that the catch-bond mechanism of FimH plays a less significant role in this niche than in the bladder where shear stress may be higher upon voiding. A low shear, coupled with the lack of uroplakin monomannose residues on renal epithelial cells, may also help explain the low level of FimH mediated epithelial binding in the early stages of infection. The same laboratory has also shown that UPEC positively select for a FimH variant that maintains attachment following a drop in shear, as compared to fecal or vaginal E. coli isolates [49]. This variation in the signal peptide of FimH, which results in expression of less, though longer fimbriae, may be more relevant under the fluctuating conditions UPEC faces in vivo. Conversely PapG has been shown to mediate binding under both shear force as well as in static conditions [50]. While the reported shear stress value for the renal tubule is low in regards to the in vitro data of Thomas et al [46], [47], the supplementary videos presented in this study, (Videos S1 and S3) do demonstrate that the niche is far from static in the early phases of infection. Further investigation is needed to draw any definitive conclusions between physiological shear in the renal tubule and the catch-bond mechanisms of E. coli adhesions. Interestingly FimH is present in all virotypes of E. coli [51], and a role for FimH in inter-bacterial binding may explain a general function for the Type 1 fimbriae in diverse perfused environmental niches.
Filtrate flow is an oft overlooked, yet crucial factor of the in vivo environment. Renal obstruction injury in itself is known to cause hemodynamic changes, epithelial damage, loss of cell membrane integrity and the expression of a number of inflammatory mediators [52], [53]. We previously showed that bacteria located in the tubule activates the clotting cascade in peritubular capillaries leading to local reduction of blood flow, PO2 and subsequent ischemic damage [20]. The present study adds obstruction to the list of contributing factors resulting in the full pathophysiology of renal infection. The complete stoppage of filtrate flow through the nephron occurring only hours after bacterial exposure is likely to result from a combination of physical obstruction by the bacteria, reduction in blood flow to the area, and the death and exfoliation of proximal tubule cells related to the ischemia. If the obstruction is related to the ischemic response it may assist in isolating the infection and preventing further bacteria spread. As with other host defense mechanisms, such as neutrophil bursts and the ischemic response, a certain level of collateral damage to the tissue is inevitable. These findings may also have important clinical consequences such as difficulties in delivery of systemically injected antibiotics to the infected nephron. Further studies are needed to define the précis signaling events occurring following infection. Collectively, our work presented here illustrates that dynamic imaging within a live setting has great potential to define new physiologically/clinically relevant outcomes of the complex microbe-host interplay.
All studies were performed in accordance with the National Institutes of Health's Guide for the Care and Use of Laboratory Animals and have been approved by the Institutional Animal Care and Use Committee at Indiana University School of Medicine Indianapolis, Indiana, USA (Study number 2453 Amendments 4, 5 and 16), Uppsala Djurfösöksetiska Nämnd, Uppsala, Sweden (Permit number: C14/6) or Stockholm's Norra Djurförsöksetiska Nämnd, Stockholm, Sweden (Permit numbers: N190/05, N347/09, N402/07).
Bacterial strains used in this study are listed in Table 1. Strains ARD41 and ARD42 were constructed by inserting the gfp+ gene into UPEC76 and W3110 as previously described [19]. Briefly, the one-step allelic recombination method was used to achieve site-specific integration of gfp+, under the control of a constitutively active tetracycline promoter PLtetO-1, into the cobS gene [54]. To generate strain ARD43, plasmid pKTH3020, carrying the pap operon, was inserted into strain ARD42 by electroporation. ARD40 was created by inserting the kanamycin resistance cassette from pKD4 into the fimH gene of LT004 deleting 321 bp between nt 5143780- 5144101, using the one step allelic knockout method [54]. Oligonucleotide sequences are listed in Table 1. All insertions were confirmed by PCR and sequencing (ABI3100, Applied Biosystems). For cloning purposes, bacteria were cultivated in aerated Luria-Bertani broth at 37°C in the presence of ampicillin (Amp, 100 µg/ml), chloramphenicol (Cm, 20 µg/ml) and kanamycin (Km, 50 µg/ml) as required. No alterations in growth rates, capsule morphology or expression of α-hemolysin were observed (data not shown). To prepare bacteria for microinfusion experiments, bacteria from aerated over-night cultures were re-inoculated (1∶100) into fresh LB with antibiotics, cultivated under shaking conditions to OD600 = 0.6, then washed and concentrated to 109 CFU/ml in PBS. Bacteria were maintained on ice (maximum 2 h) until microinfusion.
The renal infection site was dissected using a 5 mm biopsy punch, medulla tissue was removed, and total RNA extraction was performed on the resulting ∼30 mg tissue using Trizol (Invitrogen, Sweden). Experimental triplicates were performed on three separate preparations for both LT004 infected and PBS samples. cDNA was transcribed from 1 µg RNA using the SuperScript III First Strand Synthesis Supermix kit (Invitrogen, Sweden). qRT-PCR was performed using a 7500 Real Time PCR System (Applied Biosystems, Sweden) and the Power SYBR Green PCR Mastermix (Applied Biosystems, Sweden). In all experiments, gfp+ was used as an endogenous reference gene. Primer sequences are listed in Table 1.
The human kidney epithelial cell line A498 was grown on coverslips in 24-well cell culture plates in RPMI-1640 media with 10% FCS and 2 mM L-glutamate. Cells were infected with 105 CFU/ml LT004, ARD40, or ARD41 for 30 min at 37°C, 5% CO2, 95% humidity. Cells were washed 2×5 min in PBS and fixed in 4% paraformaldehyde before microscopic analysis. Image J (U. S. National Institutes of Health, MD, USA, http://rsb.info.nih.gov/ij/) was used to evaluate bacterial attachment per 40 cells. Data is pooled from a minimum of 11 tests from 2 independent experiments. Statistical significance was tested using the Student's T-test.
To detect PapG mediated agglutination, bacteria were grown overnight on a LB agar plate at 37°C. Agglutination was performed with human RBCs (O type) (8% vol/vol in PBS). To detect FimH mediated agglutination bacteria were grown overnight in a static LB culture at 37°C. Bacteria were added to a yeast suspension (5% vol/vol PBS) in the absence of mannose.
Bladders of isofluran-anesthetized female Sprauge-Dawly rats (200 g) (B and K Universal AB) were catheterized and 108 CFU of the respective bacterial strains in 1 ml PBS or control PBS were slowly infused. All strains were introduced into 5 separate animals (n = 5). Four-days post infection animals were sacrificed and kidneys removed. Kidneys were homogenized and CFU counts were obtained by plating on selective LB agar plates containing appropriate antibiotics.
Microinfusion infection was carried out as previously described [19]. Bacteria (109 CFU/ml in PBS) were mixed with 1 mg/ml Fast Green FCF (Fisher, Fair Lawn, NJ, USA) and 0.2 mg/ml Cascade blue-conjugated 10 kDa dextran. Bacteria or PBS control suspensions were aspirated into sharpened micropipettes filled with heavy mineral oil. Male Sprague-Dawley (269±30 g) or Munich-Wistar (240±80 g) rats were anesthetized by intra-peritoneal injection of 40–50 mg/kg sodium pentobarbital or 130–150 mg/kg thiobutabarbital (Inactin) (Sigma, St. Louis, MO). Munich-Wistar rats were used to allow for Bowman's capsule injections, and glomerular imaging, due to their surface glomeruli. Surgical procedures performed included a tracheotomy and cannulation of femoral artery, femoral vein and the jugular vein. The left kidney was exposed via a subcostal flank incision, and gently exteriorized. The kidney was supported by a shaped cup and using a Leitz micromanipulator and mercury leveling bulb under stereoscopic microscope observation (96×), the bacterial suspension was infused over a period of 10 min. To allow localization of injection sites Sudan black-stained castor oil was injected into nearby tubules. Injections were performed into either the proximal tubules (LT004 n = 15, ARD40 n = 7, ARD41 n = 12, ARD42 n = 5, ARD43 n = 4, and PBS n = 20) or Bowman's space (LT004 n = 5 and PBS n = 3). Bacteria were infused at an average rate of 43 nl/min corresponding to delivery of 3–6×105 CFU per injection.
All multiphoton imaging was performed using the set-up previously optimized and described [19]. Images were collected using a Bio-Rad MRC 1024 confocal/2-photon system (Bio-Rad, Hercules, CA) attached to a Nikon Diaphot inverted microscope (Fryer Co, Huntley, IL) with either a Nikon ×60 1.2-NA water-immersion or a 20x objective. Fluorescence excitation was provided by a Tsunami Lite titanium-sapphire laser (Spectraphysics, Mountain View, CA). Image stacks were collected in 1 µm optical steps into the tissue at a depth of approximately 30-100 µm using an excitation wavelength of 810 nm and neutral density filters set to 25–40%. Fluorescent probes were injected as a single bolus via a jugular vein access line. Tetramethylrhodamine-conjugated 500 kDa dextran (∼2.5 mg/400 µl 0.9% saline, Molecular Probes, Eugene, OR) was used to visualize blood flow and Hoechst 33342 (∼600 µg/0.4 ml of 0.9% saline, Molecular Probes, Eugene, OR) to stain cell nuclei. To image filtrate flow 10,000 kDa texas-red dextran was infused via the jugular vein access line until the concentration in Bowman's space reached saturation.
Anesthetized rats were placed on the microscope stage with the exposed kidney positioned in a 50 mm-diameter coverslip-bottomed cell culture dish (Warner Instruments, CT, USA) containing isotonic saline. Body temperature was monitored rectally and maintained using a heating pad covering the rat. Arterial blood pressure was regularly monitored and the rat continuously infused, via a femoral venous line, with normal saline (0.9%, 1.5 ml/h) using a syringe pump (Sage Instruments, Freedom CA). During the experiments control regions of either PBS infusion or naïve cortex were routinely checked to verify both fluorescent signal and healthy renal function.
Images and data-volumes were processed using Metamorph Image Processing Software (Universal Imaging-Molecular Devices, PA, USA) and Image J (U. S. National Institutes of Health, MD, USA, http://rsb.info.nih.gov/ij/). Final figures were prepared with Adobe Photoshop (Adobe, CA, USA). All figures presented are representative images from each experimental set (n numbers listed above).
The bacterial strains were diluted 1∶10 from an LB overnight culture (37°C) to LB medium without NaCl. 0.2 ml was added into the wells of a 96-well microtitre plate, which was incubated at 28°C for 24 h. Following incubation medium containing the planktonic bacteria was decanted and wells were washed three times with PBS. Bacteria attached to the walls of the wells were stained by adding 250 µl/well of crystal violet and incubated 10 min (room temperature) before decanting and drying. Biofilm was imaged using a digital camera. Quantification was performed by dissolving the attached bacteria with 70% ethanol and measuring the optical density at 600 nm. All samples were analyzed in triplicate from three independent experiments, using Student's t-test.
Bacteria were grown on LB agar for Western blot analysis. Bacteria (5 mg wet weight) were harvested, re-suspended in sample buffer (0.5 M Tris-HCl, pH 6.8, 87% glycerol, 4% SDS, 0.2% bromphenol blue) and heated at 95°C, 10 min. To equalize the samples, protein content was adjusted using Coomassie blue staining (20% methanol, 10% acetic acid, 0.1% Coomassie brilliant blue G). Proteins were separated by sodium dodecyl sulphate-polyacrylamide gel electrophoresis (12% resolving gel with 4% stacking gel), and transferred to a polyvinylidene difluoride membrane (Hybond-P, Amersham Biosciences). Detection of RpoS was performed according to the manufacturer instruction using a primary anti-mouse monoclonal antibody (2G10, dilution 1∶10 000, NeoClone Biotechnology, Madison) and anti-mouse immunoglobulin G conjugated with horseradish peroxidase (1∶5000, DAKO A/S Denmark). Peroxidase activity on the Hyperfilm ECL (Amersham Biosciences) was recorded using LAS-1000 system (FUJIFILM).
LT004 and ARD40 grown on LB agar plates overnight were re-suspended in water to OD600 = 0.6. 7 µl of the suspensions were inoculated into the swimming media (0.3% LB agar) and plates were incubated at 37°C for 16 h. The diameter of the swimming zone was then measured. Three experiments were performed with independent cultures in triplicate, and analyzed using Student's t-test.
CFT073 – Genebank AE014075, ref seq. NC-004431; E. coli K-12 W3110 AC_000091; fimH Gene ID1037233; pap operon Gene ID 1039518, CobS Protein ID AE016762_190.
|
10.1371/journal.ppat.1007769 | The HPV-18 E7 CKII phospho acceptor site is required for maintaining the transformed phenotype of cervical tumour-derived cells | The Human Papillomavirus E7 oncoprotein plays an essential role in the development and maintenance of malignancy, which it achieves through targeting a number of critical cell control pathways. An important element in the ability of E7 to contribute towards cell transformation is the presence of a Casein Kinase II phospho-acceptor site within the CR2 domain of the protein. Phosphorylation is believed to enhance E7 interaction with a number of different cellular target proteins, and thereby increase the ability of E7 to enhance cell proliferation and induce malignancy. However, there is little information on how important this site in E7 is, once the tumour cells have become fully transformed. In this study, we have performed genome editing of the HPV-18 E7 CKII recognition site in C4-1 cervical tumour-derived cells. We first show that mutation of HPV18 E7 S32/S34 to A32/A34 abolishes CKII phosphorylation of E7, and subsequently we have isolated C4-1 clones containing these mutations in E7. The cells continue to proliferate, but are somewhat more slow-growing than wild type cells, reach lower saturation densities, and are also more susceptible to low nutrient conditions. These cells are severely defective in matrigel invasion assays, partly due to downregulation of matrix metalloproteases (MMPs). Mechanistically, we find that phosphorylation of E7 plays a direct role in the ability of E7 to activate AKT signaling, which in turn is required for optimal levels of MMP secretion. These results demonstrate that the E7 CKII phospho-acceptor site thus continues to play an important role for E7’s activity in cells derived from cervical cancers, and suggests that blocking this activity of E7 could be expected to have therapeutic potential.
| In this study we have used genome editing to mutate the HPV-18 E7 CKII phospho-acceptor site in cells derived from a cervical cancer. We demonstrate that this results in a decrease in cell proliferation and renders the cells particularly susceptible to low nutrient conditions. Furthermore these cells are defective in invasive potential and this appears linked to a decrease in the levels of secreted MMPs. Mechanistically this is linked directly to a role of the E7 CKII phospho-acceptor site in upregulating AKT signaling. These studies demonstrate that the E7 CKII site plays a direct role in maintaining a fully transformed phenotype, and indicates a novel function for this region of E7 in regulating AKT and the levels of secreted MMPs.
| Human papillomaviruses (HPVs) are major causes of human cancer, with cervical cancer being the most important. Whilst there are over 200 different HPV types, only a small subset are responsible for the development of human cancers and, of these, HPV-16 and HPV-18 are the most common [1]. HPVs replicate in differentiating epithelia, in cells that would normally have exited the cell cycle. Since HPVs do not encode any proteins that can be used to replicate DNA, they need to drive these non-dividing cells back into cell cycle, so that the viral DNA can be amplified. This is brought about by the action of the two viral oncoproteins, E6 and E7, which together create an environment favourable for viral DNA replication [2]. This is achieved primarily through interfering with cellular growth control pathways, with E7 targeting many elements involved in the control of cell cycle, whilst E6 inhibits the pro-apoptotic response of the cell to this unscheduled DNA replication [3, 4]. In rare instances, the viral life cycle is perturbed and the events that, ultimately, give rise to malignancy are initiated. In these tumour-derived cell lines, E6 and E7 continue to be expressed, and loss of expression of either brings about cessation of cell growth and the induction of apoptosis [5–7]. Therefore, both proteins are excellent targets for therapeutic intervention in HPV-induced malignancy.
HPV E7 is a highly multifunctional protein. Major targets include the pRb family of tumour suppressors, which E7, by recruitment of cullin ubiquitin ligases, can target for proteasome-mediated degradation [8]. E7 also interacts with a large number of transcriptional regulators and other potential tumour suppressor proteins, such as PTPN14 [9, 10], and it has been shown to also recruit additional components of the ubiquitin proteasome pathway, including the UBR4/p600 ubiquitin ligase, amongst others [11, 12]. HPV E7 is also subject to phosphorylation by Casein Kinase II (CKII) [13, 14], and this appears to play an important part in the ability of E7 to bring about cell transformation [15]. At a molecular level this phosphorylation event is also thought to increase the interaction of E7 with a number of cellular target proteins, including pRb and TBP [16–19]. Indeed, recent studies identified a variant HPV-16 E7 with an extra CKII phospho-acceptor site, and this appeared to correlate with increased transforming potential [20]. Finally, recent structural studies have also begun to shed light on how phosphorylation of E7 might affect its overall structure, and thereby also affect target recognition [21].
Whilst it is clear that CKII phosphorylation of E7 plays an important role in the transformation of cells in vitro, there is little information as to how important this phospho-acceptor site is, once the cells have become fully transformed, and whether it plays any role in the maintenance of the transformed phenotype. Complementation assays using various E7 mutants, in cells where E6/E7 expression is suppressed by E2 overexpression, point to a role for CKII phosphorylation in optimal inactivation of the pRb pathway [22]. However, we have been interested in determining whether specifically blocking E7 phosphorylation by CKII in its true physiological setting could be considered as a potential option for therapeutic intervention in HPV-induced cancer. Therefore, in this study we have used a genome-editing approach to mutate the CKII phospho-acceptor site in HPV-18 E7 in cells derived from a cervical cancer. We show that these cells have profound defects in cell proliferation and invasion, which is accompanied by a marked decrease in the levels of expression of certain MMPs. These results demonstrate that phosphorylation of E7 by CKII is required for maintaining a fully-transformed phenotype, and identifies this as a potential route for therapeutic intervention in HPV-induced malignancy.
Previous studies have shown that HPV-16 E7 is phosphorylated at residues S31 and S32 by CKII, and protein alignments would also suggest that the equivalent residues S32 and S34 in HPV-18 E7 would likewise be similarly phosphorylated by CKII [14]. In order to confirm this, we first performed a series of in vitro phosphorylation assays on purified GST-18 E7, using commercially available purified CKII and 32P-γATP. To verify the identity of the phospho-acceptor site, we also generated single and double amino acid substitutions within the putative CKII phospho-acceptor site of HPV-18 E7. The results obtained are shown in Fig 1 and demonstrate that both S32 and S34 are phosphorylated by CKII. Interestingly, the single S32A mutation appears to increase the overall level of E7 phosphorylation, although the reasons for this are unclear. Mutation of both of these residues completely abolishes CKII phosphorylation of E7, confirming that these are the only CKII phospho-acceptor sites in HPV-18 E7. Taken together, these results demonstrate that both S32 and S34 are phosphorylated by CKII.
We then proceeded to design strategies for mutating these residues in cells derived from a cervical cancer. To do this we chose C4-1 cells, which only have one copy of HPV-18 DNA integrated into the host genome, and are therefore more amenable to genome editing than a line containing multiple copies of integrated HPV genomes [23]. Prior to performing genome editing it was first necessary to confirm that C4-1 cells were indeed still dependent upon the continued expression of E6 and E7 for maintenance of the transformed phenotype. To do this, the cells were transfected with siRNAs targeting HPV-18 E6 and 18 E6/E7, and after 72hrs the numbers of cells remaining were counted. The siRNA-transfected cells were also analysed for the levels of p53 and pRB using Western blotting. The results, together with the microscopic appearance of the cells, is shown in Fig 2, where it can be seen that C4-1 cells are indeed still dependent upon continued E6/E7 expression for maintenance of cell survival, in agreement with similar studies done in other HPV positive cervical tumour-derived cell lines [5–7].
To mutate the E7 CKII phospho-acceptor site in C4-1 cells, two guide RNAs were designed, aiming to generate the S32A and S34A double mutant. These are depicted in Fig 3A, together with the donor DNA sequence shown in Fig 3C, which was designed to create a new HgaI restriction site in E7 to facilitate the screening of the clones. The efficiency of the guide RNAs was first confirmed using the surveyor nuclease assay in C4-1 cells, to analyze their ability to cut the target DNA sequence. As can be seen from Fig 3B the guide RNAs efficiently targeted the relevant region of the E7 gene.
C4-1 cells were then transfected with the guide RNAs and the single-stranded donor DNA and, following an extended period of selection and single cell cloning, the cells were analysed for the S32A/S34A double mutation in the CKII phospho-acceptor site, by HgaI digestion of the 18E7 PCR amplicon and then by DNA sequencing, whereby two such clones were ultimately identified (Fig 3D and 3E).
Having obtained cell lines in which the CKII HPV-18 E7 phospho-acceptor sites were mutated, we were next interested in more fully characterizing them. We first analysed their growth rates in normal and low nutrient conditions. As can be seen from Fig 4, both the mutant clones grow somewhat more slowly than the wild type cells and, furthermore, the saturation densities are also consistently lower with the mutant cell lines. This is even more apparent when the cells are grown under conditions of low serum, where in 0.2% serum there is a dramatic decrease in the growth rate in the two mutant cell lines (Fig 4C and 4D). The growth curve in 0.2% serum (Fig 4C) reaches saturation density and drops after 4 days, possibly due to nutrient deprivation and cell death. When the assay is performed replacing the medium after the third day (Fig 4D), it is clear that the wild type cells can continue to proliferate while the mutant cells have greatly reduced rates of cell proliferation. These results suggest that mutation of the CKII phospho-acceptor sites in HPV-18 E7 in C4-1 cells has a markedly deleterious effect upon the ability of the cells to proliferate.
Having shown that the mutant cells are less proliferative than the parental cells, we then analysed the levels of pocket proteins (pRB and p130) in these cell lines in comparison with the parental cell line. As can be seen from Fig 5, both the double mutant cell lines have slightly higher levels of pRB and p130 compared with the wild type line. Since previous studies had indicated that an intact CKII phospho-acceptor site can aid pocket protein recognition [24, 25], we also performed co-immunoprecipitation analysis between HPV-18 E7 and pRb in the wild type and mutant E7 cell lines. The results obtained are shown in Fig 5C and 5D where it can be seen that lower levels of pRb are complexed with E7 in the two mutant lines when compared with the wild type C4-1 cells. We also analysed the total levels of expression of the E7 oncoprotein and, as can also be seen from Fig 5A and 5C, mutation of the CKII phospho-acceptor site had no deleterious effects upon the levels of E7 protein. In addition, the levels of p53 were also unchanged, suggesting that the levels of E6 expression were similarly unaffected. These results demonstrate that mutation of the E7 CKII phospho-acceptor site has a modest effect upon overall levels of pocket protein expression, as would be expected from previous structural studies [17–19, 24, 25], however whether such changes are sufficient to account for the defects in cell proliferation remains to be determined.
We then proceeded to investigate the invasive capacity of the mutant clones using a matrigel invasion chamber assay. The cells were plated on the upper chamber without serum and a chemo-attractant stimulus was provided in the lower chamber in the form of 2% calf serum. After 22hrs the lower chamber was fixed and stained and the number of migrating cells were counted. As can be seen from Fig 6, wild type C4-1 cells are very invasive, and migrate readily through the collagen matrix. In contrast, both mutant lines show defects in their invasive potential. Thus, the mutation within the HPV-18 E7 CKII phospho-acceptor site greatly decreases the ability of the cells to invade through a collagen matrix.
To investigate whether the mutant cells’ defective invasion potential was related to any change in their ability to hydrolyse the extracellular matrix (ECM), we concentrated conditioned media from the mutant cells and analysed their activity by Gelatin Zymography. As can be seen in Fig 7A, conditioned media from the mutant lines are unable to hydrolyse the gelatin gel, whereas that from the wildtype cells shows significant bands of gelatin hydrolysis. We then performed a series of western blot analyses on the conditioned medium to try and determine which matrix metalloproteases might be involved. As can be seen from Fig 7B and 7D, both MMP1 and MMP13 are significantly downregulated in the two mutant clones when compared with the wild type C4-1 cells. This appears to be highly specific, since there is no change in the levels of expression of MMP8 (Fig 7E). To determine whether the defect in MMP secretion was a reflection of a slower growth rate in the clones, we repeated the assay but growth arrested the wild type C4-1 cells with mitomycin C treatment for 48hrs. Secreted MMP levels were then analysed by western blotting, and as can be seen from Supplementary S1 Fig there is no apparent difference in MMP levels in the growth arrested cells. These results indicate that the differences in MMP levels between the wild type and mutant C4-1 cells, is not due to differences in proliferation.
In order to determine whether the high levels of MMP1 and MMP13 expression are dependent upon CKII activity, we treated the wild type C4-1 cells with the highly specific CKII inhibitor CX-4945 and analysed the effects upon the protein levels of MMP1 and MMP13. As can be seen from Fig 7C and 7D, CX-4945 treatment of the wild type cells reduces the levels of MMP1 and MMP13 expression to levels similar to those seen in the two mutant lines. This suggests that active CKII phosphorylation of HPV-18 E7 is required for maintaining high levels of MMP protein expression in C4-1 cells.
All the above analyses were done on the C4-1 cells, but we wished to determine whether high levels of MMP secretion were specific to active CKII in HPV-positive cells. To do this we analysed the levels of secreted MMPs in HPV-16-positive CaSki cells and HPV-negative C33a cells, in the presence and absence of CKII inhibitor. The results obtained are shown in Fig 7F and 7G and demonstrate that high levels of secreted MMP1 and MMP13 in CaSki cells are also dependent upon active CKII, whilst in C33a cells inhibition of CKII has no effect upon the levels of secreted MMPs. These results indicate a specific role for active CKII in regulating MMP levels in HPV-positive cells. To verify this, we performed a rescue experiment where we overexpressed FLAG-tagged wild type HPV-18 E7 in the two mutant lines and examined the effects on MMP levels secreted by the cells. Fig 8A shows the levels of endogenous and overexpressed HPV-18 E7 whilst, as can be seen from Fig 8B, in cells that expressed high levels of wildtype FLAG-tagged HPV-18 E7 the amounts of secreted MMP1 and MMP13 were increased, while no effect was seen on the levels of MMP8 (Fig 8B). In order to determine whether E7 affected the total levels of MMP1 expression, we also performed western blot analysis on total cell extracts derived from the wild type, mutant and E7 overexpressing cells. As can be seen from Fig 8C, the intracellular levels of MMP1 are unaffected by mutation of the E7 CKII phospho-acceptor site and remain unchanged following overexpression of wild type E7.
Since the rescue of the levels of secreted MMPs was only partial, which was most likely due to a low level (approx. 20%) of transfection efficiency, we proceeded to generate cells stably expressing wild type HPV-18 E7. To do this, the CKII mutant cells were transfected with GFP 18E7 and selected by FACS sorting and subsequent G418 selection. The pooled cells were then analysed for the levels of secreted MMPs. As can be seen from Fig 8D, both of the CKII mutant polyclonal cell lines stably express GFP 18E7. Furthermore, in these rescued cells the conditioned medium has very similar levels of secreted MMP1 to that seen in the wild type C4-1 cells. To further verify that this rescue is due to phosphorylation of stably expressed wild type HPV-18 E7, cells stably expressing a C-terminal FLAG tagged wild type HPV-18 E7 and the HPV-18 E7 S32A/S34A mutant were also obtained and analysed for MMP secretion. As can be seen from Fig 8E, despite similar levels of wild type and mutant HPV-18 E7 S32A/S34A expression, wild type levels of MMP1 secretion was only obtained in the cells expressing the wild type HPV-18 E7 protein. These results demonstrate that an intact E7 CKII phospho-acceptor site plays an essential role in ensuring high levels of MMP secretion in a HPV-18-positive cervical tumour-derived cell line.
Having identified MMP1 and MMP13 as being upregulated in a CKII dependent manner in HPV-positive cells, we then asked if knockdown of MMP1 and MMP13 would have any deleterious affect upon the invasive ability of the wild type cells. To do this, we transfected siRNA against luciferase as control, or MMP1 or MMP13, or a combination of the two into wild type C4-1 cells, and after 48 hours pooled equal number of cells and performed a Matrigel invasion assay. As can be seen from Fig 9A–9C, there is marked decrease in invasion upon knockdown of MMP13, little effect with MMP1 knockdown alone, but a marked synergistic effect when both MMP1 and MMP13 were removed. Similar results were obtained when the assay was performed in HPV-16-positive CaSki cells (Fig 9D and 9E).
Taken together, these results demonstrate that loss of CKII phosphorylation of HPV-18 E7 in cells derived from a cervical cancer, has diverse inhibitory effects upon the ability of the cells to proliferate and to invade a collagen matrix, and the defect in the ability to invade is at least in part due to downregulation of MMP1/13.
We were next interested in understanding the molecular basis for this defect. Previous studies had implicated the ability of E7 to activate AKT signaling as being in part responsible for promoting invasive behavior [26]. However no studies have been done to investigate whether CKII phosphorylation of E7 can lead to activation of AKT, and whether this in turn might be responsible for the increased levels of MMP secretion. To investigate these possibilities we first analysed the ability of wild type and CKII site mutants of HPV-16 and HPV-18 E7 to activate AKT in a transient assay in HEK293 cells. In order to first verify the levels of E7 phosphorylation in these assays, we generated a phospho-specific antibody directed against the HPV-16 E7 CKII phospho-acceptor site. The characterization of this antibody is shown in Supplementary S2 Fig, where it can be seen that it only reacts against the phosphorylated form of HPV-16 E7. After 24 hours, the transfected 293 cells were serum starved for 16 hours and harvested and analysed by western blotting. As can be seen from Fig 10A, only the wild type E7 is capable of inducing AKT phosphorylation. Interestingly, the phospho-specific antibody detects the wild type E7 but fails to react against the CKII phospho-site mutant. These results confirm a significant level of E7 phosphorylation with the wild type protein, but no phosphorylation in the case of the double CKII phospho-site mutant of E7.
Having found that an intact CKII phospho-acceptor site was required for the ability of E7 to increase AKT phosphorylation, we then analysed the levels of AKT phosphorylation in the wild type and CKII phospho-site mutant C4-1 cells following serum starvation for 48 hours. The results in Fig 10B also confirm the requirement for an intact CKII phospho-acceptor site to obtain high levels of AKT phosphorylation.
These results implicate AKT signaling in HPV-positive cells as being responsible for the high levels of MMP secretion. To investigate this further we analysed the levels of MMP1 and MMP13 secretion in HPV-18-positive C4-1 cells, HPV-16-positive CaSki cells and HPV-negative C33a cells following inhibition of AKT and PI3K using the specific AKT inhibitor (124005-Calbiochem) and PI3K inhibitor (LY294002). After 48 hours the supernatants were harvested and analysed by western blotting. As can be seen from Fig 10C–10E, blocking AKT signaling in HPV-positive cells results in a marked decrease in the levels of secreted MMP1 and MMP13, whilst this has no effect in the HPV-negative cells. In agreement with results in Figs 7 and 8, which showed that CKII phosphorylation of E7 had no effect on MMP8, likewise blocking AKT signaling also had no effect on MMP8 secretion in HPV-positive cells. Taken together these results demonstrate that CKII phosphorylation of E7 promotes AKT activation, which, in turn, is responsible for increased MMP secretion and subsequent enhancement of invasive potential in HPV-positive cervical tumour-derived cells.
Despite extensive characterization of the different biochemical functions of the HPV E6 and E7 oncoproteins, and their respective contributions towards the development of cell transformation and malignancy, there is still very little information on the continued requirement for specific activities of E6 and E7 in cells derived from a cervical cancer. This is important, since we know there is an absolute requirement for both oncoproteins for maintaining the transformed phenotype [5–7], but which specific biochemical functions of E6 and E7 contribute towards this remains obscure. In this study we have begun to address this question by performing genome editing of the HPV-18 E7 CKII phospho-acceptor site in cervical cancer-derived C4-1 cells. We find that an intact CKII phospho-acceptor site in E7 remains important, both for optimal levels of cell proliferation and for maintaining high levels of invasive potential.
Previous studies have shown that the CKII phospho-acceptor site in HPV-16 E7 plays an important role in the HPV life cycle, and in the ability of E7 to bring about cell transformation in a variety of different experimental settings [15, 27]. Furthermore, recent studies showed that a variant HPV-16 E7, in which an extra CKII phospho-site was present at N29S, exhibited a marked increase in transforming potential [20]. Many of these activities have been linked to the effects of phosphorylation on the ability of E7 to interact with some of its cellular substrates. Indeed the acidic pocket provided by the CKII consensus motif, and the subsequent increase in negative charge following phosphorylation, have been shown to increase interaction with pocket protein family members [17–19]. One consequence of which is the enhancement of HPV-16 E7’s ability to overcome pRb-associated cell senescence [22]. Phosphorylation of HPV-16 E7 has also been shown to increase interaction with TBP [16]. HPV-18 E7 is also widely assumed to be similarly phosphorylated, however the specific phospho-acceptor site had never been formally verified [14]. CKII substrate specificity is known to be determined by multiple acidic residues located between -2 and +5 positions relative to the S/T phospho-acceptor site [28, 29]. Based on our mutational analysis we confirmed that S32/S34 are the only CKII phospho-acceptor sites in HPV-18 E7. However, S34 seems to be a more preferred CKII recognition site, although the S32A mutation appears to further enhance phosphorylation of E7 by CKII at S34.
To perform the genome editing of the CKII phospho-acceptor site, we chose C4-1 cells as these cells have only a single integrated copy of HPV-18 DNA [23]. We first confirmed previous studies demonstrating a continued requirement for E6/E7 expression for maintenance of the transformed phenotype in cells derived from a cervical cancer [5–7], as this had not formally been demonstrated in C4-1 cells. Following the design of gRNAs for targeting the two serines in E7 and subsequent transfection and selection, two mutant cell lines were eventually obtained. This demonstrates that mutation of the CKII site per se is not incompatible with continued cell growth and survival. However, upon more detailed analysis, it is clear that mutation of the CKII phospho-acceptor site results in quite marked effects upon the degree of cell transformation. Thus cells with a mutant CKII site in E7 grow more slowly and attain lower saturation densities than wild type E7 cells. This is even more marked when cell growth is assayed in low concentrations of calf serum, where there is a dramatic reduction in the rates of cell proliferation. None of these defects appear related to significant changes in the levels of E7 protein expression, as this was unchanged in the mutant lines. It was also unrelated to altered levels of E6 expression, as gauged by the levels of p53. One possible explanation for these slower growth rates under nutrient deprivation could be related to the lower levels of mutant E7 interaction with pocket protein family members, an effect that confirms previous studies showing a role for the acidic patch of the CKII phospho-acceptor site in pocket protein recognition [24, 25]. Indeed, these studies in C4-1 cells are the first to demonstrate unequivocally that an intact CKII phospho-acceptor site in endogenously expressed E7 does influence interaction with pocket proteins. However, whilst this may influence the rates of proliferation, it seems unlikely to be the major cause, as changes in pocket protein levels are quite small. It is tempting to suggest that it is the defect in AKT activation (see below) in the E7 CKII phospho-site mutant lines, especially under low nutrient conditions, which is primarily responsible for the slower growth rates [30].
Further analysis of the effects of the CKII mutation upon the transformed properties of the cells revealed marked alterations in the invasive potential of the mutant cell lines. Using matrigel invasion assays, we found that the mutant cells were much less invasive than the parental cells, suggesting that mutation of the E7 CKII phospho-acceptor site reduces the capacity of the cells to invade through a collagen-containing 3D matrix. This defect was not due to a generalized defect in migratory capacity, since the wild type and mutant cells showed no differences in their respective abilities to migrate in a scratch wound healing assay (S3 Fig). Instead, this points to a specific defect in migration through a collagen matrix. This therefore led us to examine any potential effects upon MMPs. Using conditioned medium from the different cell lines, we first analysed potential differences in a collagen zymograph assay. To our surprise, we found marked differences between the wild type and mutant cells lines, with the cells harbouring the CKII site mutation exhibiting much lower levels of collegenase activity. This focused our attention on a subset of MMPs, which might be responsible for this defect, and we found that two MMPs, MMP1 and MMP13 were particularly highly secreted by the wild type C4-1 cells, but were largely absent from the conditioned medium of the mutant cell lines. This defect appears to be related to CKII activity, since treatment of the wild type cells with a CKII inhibitor also resulted in a decrease in the amount of secreted MMP1 and MMP13. Similar results were also obtained in HPV-16-positive CaSki cells, where again high levels of MMP1 and MMP13 secretion were dependent upon active CKII, whilst in HPV-negative C33a cells inhibition of CKII had no effect upon MMP secretion. Obviously, confirming the link between E7 phosphorylation and MMP secretion through further genome editing in other HPV-positive cells would be a valuable confirmation, but the technical difficulties in achieving this in cells with multiple HPV copies could render such approaches inconclusive. Finally, overexpression of wild type HPV-18 E7 in the two mutant lines effectively restored the levels of secreted MMP1 to levels similar to that seen in the wild type C4-1 cells, although restoration of MMP13 was much lower. Furthermore, this rescue of MMP1 secretion was dependent upon an intact CKII phospho-acceptor site in E7, since stable expression of the S32A/S34A double mutant failed to restore the levels of MMP1 secretion. It should be emphasized that when total cell extracts were analysed for the different MMPs, there was very little difference between the different cells, suggesting that the defect was not in expression, but rather in secretion. We also confirmed that MMP13 in particular also played an essential role in promoting invasion in the matrigel invasion assay, where ablation of expression resulted in a marked decrease in invasive potential, and, where interestingly, the double knockdown of both MMP1 and MMP13 gave the most severe defect in invasive capacity.
A major question at this point was the underlying mechanism. Previous studies have shown that E7 can induce upregulation of diverse MMPs at the transcriptional level [26, 31–34], although that did not seem to be the case in the case of the mutant C4-1 cells, where total levels of MMP expression were similar to those seen in the wild type cells. Other studies had shown that E7 could also upregulate AKT activity, which in turn could affect invasive potential, [26, 35], although the potential role of the CKII phospho-acceptor site of E7 was not investigated in any of these or earlier analyses [30]. We therefore proceeded to determine whether CKII phosphorylation of E7 might impact upon the levels of AKT activation, and in turn whether this might be responsible for the high levels of MMP secretion. Using both transient overexpression and the mutant C4-1 cells, it is quite clear that an intact CKII phospho-acceptor site on both HPV-16 and HPV-18 E7 plays an important role in the activation of AKT. Most importantly however, this activation of the AKT pathway in HPV-positive cells appears to be directly involved in the increase in secretion of MMP1 and MMP13, but not of MMP8, which is unaffected by either mutation of the E7 CKII phospho-acceptor site, or by CKII or AKT inhibition. In contrast, in HPV-negative C33a cells, inhibition of AKT signaling has no effect upon the levels of secreted MMP1 and MMP13. These studies demonstrate that phosphorylation of E7 enhances AKT signaling, which in turn can increase levels of MMP secretion and invasiveness, and which might also contribute towards increased rates of cell proliferation in low nutrient conditions.
It is intriguing that these studies show a clear difference in MMP1 and 13 upregulation, but no change in MMP8. Whilst the molecular basis for this requires further experimentation it is interesting that MMP1 and MMP13 have been found to be upregulated in a variety of different tumours, including cervical, whilst MMP8 on the other hand has been associated with an anti-invasive effect in certain settings [36–41]. Our current studies are now aimed at understanding how phosphorylation of E7 might modulate MMP secretion, and our focus is upon potential interaction with endocytic transport pathways.
In summary, these studies demonstrate that the activity of the E7 CKII phospho-acceptor site remains functional and relevant in cells derived from cervical cancers and indicates that the CKII phospho-acceptor site remains a very attractive candidate target for therapeutic intervention in HPV-induced malignancy.
The C4-1 cell line was obtained from the American Type Culture Collection (ATCC) and maintained in Dulbecco’s modified Eagle’s medium (DMEM), supplemented with 10% fetal bovine serum, glutamine (300 μg/ml), and penicillin-streptomycin (100 U/ml).
The inhibitors and chemicals used for the experiments are as follows: 2.5 μM Silmitasertib (CX-4945), 5μM AKT inhibitor (124005-Calbiochem), 15μg/mL PI3K inhibitor (LY294002) and 10μg/mL Mitomycin C.
The GST 18 E7 S32A/S34A was synthesized by the Gene Art Gene Synthesis protocol (Invitrogen) and cloned into the BamHI/EcoRI sites of the pGEX2T for GST fusion protein expression. The wild-type pGEX 18E7 plasmid was a kind gift from Karl Münger. The same plasmid was used to generate pGEX 18 E7 S32A and pGEX 18E7 S34A using a modification of the QuickChange site-directed mutagenesis system (Stratagene), according to the manufacturer’s instructions, with the following primers: S32A forward primer 5’-TGTCACGAGCAATTAGCGGACTCAGAGGAAGAA-3’ and S32A reverse primer 5’- TTCTTCCTCTGAGTCCGCTAATTGCTCGTGACA-3’; S34A forward primer 5’-GAGCAATTAAGCGACGCAGAGGAAGAAAACGAT-3’ and S34A reverse primer 5’-ATCGTTTTCTTCCTCTGCGTCGCTTAATTGCTC-3’. pSpCas9(BB)-2A-GFP (PX458) was purchased from Addgene.
The C-terminal FLAG-tagged HPV-18 E7 wild type and S32A/S34A CMV plasmid construct was generated by amplification and sub-cloning of the E7 coding sequences from the wild-type pGEX 18 E7 and pGEX 18 E7 S32A/S34A constructs as templates respectively into the pCMV Neo Bam (XhoI) empty vector (kindly provided by J. Mymryk). The primers used were the following: HPV-18 E7 forward primer 5’-CGACGGATCCGATTCGAGACCATGCATGGACCT-3’ and reverse primer 5’-GCATCTCGAGCTACTTGTCATCGTCGTCCTTGTAGTCCTGCTGGGATGC-3’. The amplified sequences were visualized by agarose gel electrophoresis, purified by using a QIAquick gel extraction kit (Qiagen), digested with BamHI and XhoI restriction enzymes, and ligated into pCMV Neo Bam (XhoI).
The C-terminal FLAG/HA-tagged pCMV HPV-16 E7 plasmid was a kind gift from Karl Münger [42]. The same plasmid was used to generate pCMV HPV-16 E7 S31A/S32A using the forward primer 5’-CTCTACTGTTATGAGCAATTAAATGACGCCGCAGAGGAGGA-3’ and the reverse primer 5’-TCATCCTCCTCCTCTGCGGCGTCATTTAATTGCTCATAACA-3’ using the QuickChange site-directed mutagenesis system (Stratagene) as described previously. The pEGFP 18 E7 was generated by PCR amplification and sub-cloning of E7 coding sequences from wildtype CMV 18E7 FLAG plasmid as template using the forward primer 5’- ATGCGAATTCATGCATGGA-3’ and the reverse primer 5’- GCATTCTAGATT ACTGCTG-3’ into EcoRI and XbaI restriction sites of the pCANmyc-EGFP 16E7 (JMB-04093) plasmid (kindly provided by J. Mymryk). pEGFP-C1 empty plasmid was from Invitrogen. The HA-PKB expression plasmid has been described previously [30].
Purified GST fusion proteins were incubated with CKII enzyme (NEB) in 20 μl kinase buffer (20 mM Tris-HCI [pH 7.5], 5 mM MnCl2) in the presence of 50 μCi [γ-32P] ATP (2,000 Ci/mmol)) for 15 min at 30°C. After extensive washing with kinase wash buffer (20 mM Tris-HCI [pH 7.5], 5 mM MnCl2, 0.1% NP-40), GST fusion proteins were subjected to SDS-PAGE and autoradiographic analysis [16].
For transfection of siRNAs, C4-1 cells were seeded at 2 x 105 cells in 60 mm petri plates and grown overnight in a humidified CO2 incubator. Lipofectamine RNAiMAX (Invitrogen) was used to transfect siRNAs against luciferase, HPV-18 E6 (5’-CUCUGUGUAUGGAGACACAT-3’), HPV-18 E6/E7 (5’-CAUUUACCAGCCCGACGAG-3’), MMP1 (Dharmacon) and MMP13 (Dharmacon) according to the manufacturer’s instructions.
A couple of gRNAs targeting CKII phosphorylation site in HPV 18 E7 genomic region was designed using online gRNA design platform–CRISPR MultiTargeter (http://www.multicrispr.net/). gRNA oligos (gRNA1–5’-CACCGCGAGCAATTAAGCGACTCAG-3’ and 5’-AAACCTGAGTCGCTTAATTGCTCGC-3’; gRNA2–5’-CACCGTTAATTGCTCGTGACATAGA-3’ and 5’-AAACTCTATGTCACGAGCAATTAAC-3’) were then annealed and cloned into the BbsI restriction site in pSpCas9(BB)-2A-GFP (PX458).
Single stranded DNA oligonucleotide (ssODN) as donor template for homology-directed repair (HDR) was designed as 100bp homology arm flanking the predicted double strand break site to abolish phosphorylation, substituting serine 32 with alanine (agc to gcc) and serine 34 with alanine (tca to gca). The designed mutagenesis also included a unique Hga I restriction site, to allow screening of the edited clones by genomic DNA isolation, PCR amplification of edited region and Hga I restriction digestion.
C4-1 cells were transfected with pSpCas9(BB)-2A-GFP or with the same construct with gRNA1 or gRNA2 using the Amaxa Cell Line Nucleofector Kit C according to the manufacturer’s instructions. Transfected cells were incubated for 72 hrs in a humidified CO2 incubator and then harvested with trypsin, washed once with PBS and resuspended in 5mM EDTA [pH 8.0] in PBS for FACS (BD FACSAria II) sorting. GFP-positive cells were collected and genomic DNA was isolated using the Wizard Genomic DNA Purification Kit (Promega) according to the manufacturer’s instructions. HPV 18 E7 ORF was then amplified (414 bp) using forward primer 5’-CCAACGACGCAGAGAAACAC-3’ and reverse primer 5’-AAACCAGCCGTTACAACCCG-3’. Amplicons were then denatured and reannealed as follows: 95°C for 10 min, 95°C to 85°C for 1 min, 85°C to 75°C for 1 min and so on for every 10°C decrease in temperature for 1 min and finally a hold at 25°C for 1 min using a thermal cycler to allow DNA heteroduplex formation. The heteroduplexes were then treated with Surveyor nuclease and Surveyor enhancer (IDT) and analyzed on a 2% agarose gel. The cleavage intensity of the gRNAs was calculated by analyzing band intensities using the ImageJ program. The indel percentage caused by respective gRNA was then calculated as described by Ran et al. [43].
C4-1 cell lines were transfected as described above with pSpCas9(BB)-2A-GFP gRNA1 together with ssODN HDR template and incubated in a humidified CO2 incubator. Seventy-two hours after transfection, the top 10% GFP-positive cells were sorted as described above and seeded in limiting dilution in 100 mm diameter petri dishes for isolation of single cell clones using cloning chambers or 96 well tissue culture plates. Genomic DNA extraction of the single cell colonies and PCR amplification of the E7 ORF was performed as before. Then, individual PCR amplicons were digested with Hga I restriction enzyme and analyzed on a 2% agarose gel. Clones positive for Hga I restriction digestion were further verified by Sanger sequencing for genome editing at the CK II site.
C4-1 wild type and mutant lines were seeded at 3 x 105 cells in a 60 mm petri dishes in DMEM with 10% fetal bovine serum. Cells were harvested with trypsin, dispersed and counted using a hemocytometer every single day or two. For determination of growth in low serum, the experiment was performed independently in DMEM with 1% and 0.2% fetal bovine serum. Cell counts were analyzed using Graphpad Prism 7.0 to generate a growth curve.
Following antibodies were used for western blotting. Mouse monoclonal anti-p53 antibody (DO-1), rabbit polyclonal anti-p130 antibody (C-20), mouse monoclonal anti-HPV18 E7 antibody (F-7), mouse monoclonal anti-MMP-1 antibody (SB12e), mouse monoclonal anti-MMP-8 (B-1), mouse monoclonal anti-MMP13 (C-3), mouse monoclonal anti-α-actinin antibody (H-2), mouse monoclonal anti-GFP Antibody (B-2) from Santa Cruz Biotechnology; mouse monoclonal anti-Rb (G3-245; BD Pharmingen), mouse monoclonal anti-α-tubulin, mouse monoclonal anti-HA-peroxidase (clone HA-7), mouse monoclonal anti-FLAG-M2-peroxidase from Sigma, mouse monoclonal anti-18E6 antibody (N-terminus #399; Arbor Vita Corporation), rabbit Akt antibody #9272, rabbit anti phospho-Akt (Ser473) antibody #9271 from Cell Signaling. HPV-16 E7 pS31/S32 peptide antibody was generated by Eurogentec (peptide sequence: C+EQLND-S(PO3H2)S(PO3H2)-EEED and validated by ELISA and Western blotting. Secondary anti-rabbit HRP and anti-mouse HRP antibodies were obtained from Dako.
For Western blotting, total cell extracts were obtained by lysing the cells directly in 2X SDS-PAGE sample buffer, and were then separated by SDS-PAGE and blotted on 0.22-μm nitrocellulose membrane. Membranes were blocked in 5% non-fat dry milk in TBST (20mM Tris-HCl pH 7.5, 150 mM NaCl, 0.1% Tween-20) for 1 h and probed with appropriate primary and secondary antibodies. The blots were then developed using the ECL Western blotting detection reagent (GE Healthcare) according to the manufacturer’s instructions.
For detection of extracellular MMPs, confluent cells were cultured in serum free medium for 48 hours and conditioned media from cell lines were concentrated 10 times using Amicon Ultra-4 centrifugal devices and equal amounts of proteins were mixed with 2X SDS-PAGE sample buffer and processed as described previously for Western blotting.
For co-immunoprecipitation, wild type and CKII phospho-acceptor site mutant C4-1 cells were harvested using lysis buffer (50mM HEPES pH7.4, 150mM NaCl, 1mMMgCl2, 1mM NaF, 1% Triton-x-100, protease inhibitor cocktail I [Calbiochem]) and incubated with 1μg of mouse monoclonal HPV-18 E7 antibody overnight at 4°C. Mouse monoclonal IgG antibody against GFP was used as a control. After incubation, immune complexes were incubated further with Protein A beads (Immobilized protein A 300, Repligen) for 90 minutes at 4°C, followed by four washes in the lysis buffer. Immunoprecipitates were then run on SDS PAGE and analysed by Western blotting.
Matrigel Invasion Assays were performed as described previously [44]. Briefly, Matrigel invasion chambers (Corning BioCoat Matrigel Invasion Chamber) were brought to room temperature and rehydrated with DMEM without serum for 2 hours in a humidified CO2 incubator. Wild type and mutant C4-1 cells were seeded at 1×105 cells in 200 μl of growth medium into the upper chamber. After allowing the cells to attach for 1 h, the medium in the upper chamber was replaced with DMEM without serum, while DMEM with 2% serum was added to the lower chamber as chemoattractant. After 22 hours, DMEM and any cells remaining in the upper chamber were removed by wiping with a cotton swab. Cells that had invaded the lower chamber were then fixed and stained with 0.5% Crystal Violet in 5% glutaraldehyde for 10 mins. After washing the excess stain with distilled water, the membrane was removed from the insert housing and placed on a microscope slide for imaging and analysis using a transmitted light microscope at 20X magnification. At least three fields per membrane were counted for each cell line. Invaded cell counts were analyzed using Graphpad Prism 7.0.
Wild type and mutant C4-1 cell lines were incubated in a serum-free medium for 48 hours and conditioned media from the cell lines were concentrated 10 times using Amicon Ultra-4 centrifugal devices. Equal amounts of proteins were mixed with non-reducing sample buffer (125mM Tris-HCl pH 6.8, 4% SDS, 20% glycerol and 0.01% bromophenol blue) and run on a 10% SDS-PAGE containing 0.1% gelatin. The gel was renatured by washing with 2.5% Triton X-100 in 50mM Tris pH 7.4, 5mM CaCl2 and 1 μM ZnCl2 for 1 hr. After rinsing briefly with deionized water, the gel was incubated overnight at 37°C in 1% Triton X-100, 50mM Tris pH 7.4, 5mM CaCl2 and 1 μM ZnCl2 and stained with Coomassie staining solution (0.5% Coomassie G250, 40% Methanol, 10% Acetic acid and 50% deionized water) for 1 hr and de-stained in 40% Methanol, 10% Acetic acid and 50% deionized water, until clear bands of hydrolyzed substrate were visualized.
|
10.1371/journal.pbio.0060122 | Resource Heterogeneity Moderates the Biodiversity-Function Relationship in Real World Ecosystems | Numerous recent studies have tested the effects of plant, pollinator, and predator diversity on primary productivity, pollination, and consumption, respectively. Many have shown a positive relationship, particularly in controlled experiments, but variability in results has emphasized the context-dependency of these relationships. Complementary resource use may lead to a positive relationship between diversity and these processes, but only when a diverse array of niches is available to be partitioned among species. Therefore, the slope of the diversity-function relationship may change across differing levels of heterogeneity, but empirical evaluations of this pattern are lacking. Here we examine three important functions/properties in different real world (i.e., nonexperimental) ecosystems: plant biomass in German grasslands, parasitism rates across five habitat types in coastal Ecuador, and coffee pollination in agroforestry systems in Indonesia. We use general linear and structural equation modeling to demonstrate that the effect of diversity on these processes is context dependent, such that the slope of this relationship increases in environments where limiting resources (soil nutrients, host insects, and coffee flowers, respectively) are spatially heterogeneous. These real world patterns, combined with previous experiments, suggest that biodiversity may have its greatest impact on the functioning of diverse, naturally heterogeneous ecosystems.
| The world is currently experiencing a rapid loss of species, prompting investigation into the role of biodiversity on the functioning of ecosystems. Many recent studies have shown that high diversity of plants, pollinators, and predators is related to high plant growth, pollination, and predation, respectively. Many of these studies involved controlled experiments, yet results were highly variable, indicating that the environment may affect the relationship between species diversity and these ecosystem functions. In a heterogeneous environment, different species can occupy different microhabitats, or use different resources. This reduces competition between species, and can mean that diverse assemblages perform their ecosystem functions at elevated rates. Here we examine rates of three important functions in different natural, nonexperimental ecosystems: plant biomass production in German grasslands, parasitism rates across five habitat types in coastal Ecuador, and coffee pollination in agroforestry systems in Indonesia. We demonstrate that the effect of diversity on these processes increases in environments where limiting resources (soil nutrients, host insects, and coffee flowers, respectively) are spatially heterogeneous. These real world patterns, combined with previous experiments suggest that biodiversity may have its greatest impact on the functioning of diverse, naturally heterogeneous ecosystems.
| Global biodiversity decline has prompted a recent explosion of experimental studies addressing the relationship between biodiversity and ecosystem functioning (BDEF) [1–4]. Most of this research has focused on the relationship between diversity and productivity in plant communities (e.g., [5,6]), as reduced primary production due to lost biodiversity may have critical consequences for food production, carbon sequestration, and ecosystem functioning. However, impacts of biodiversity change on function may also be trophically mediated by mobile consumers [7–9]. Therefore, a critical new direction in BDEF research addresses predator-prey interactions [10,11], with important implications for biological pest control, such as the effect of natural enemy diversity on rates of prey consumption [12,13]. Pollination has also been examined within a BDEF framework, as loss of pollinator diversity can reduce pollination rates and may threaten crop production [14–16].
Recent meta-analytical syntheses have shown that on average, reductions in species richness result in a decrease in within-trophic-level abundance or biomass production, less complete resource depletion [17], and can negatively affect ecosystem services [18]. However, results of experiments are notoriously variable [19], with many studies also finding negative [20,21] and no effects [22] of diversity on function. An important development in the BDEF literature has thus been the recognition that the shape and direction of the BDEF relationship can depend critically on environmental context [23–28], which may partly account for the variability in observed outcomes across studies. For example, resource enrichment often strengthens the relationship between plant diversity and productivity [25,29,30], and the relationship between natural enemy diversity and prey consumption rates can vary as a function of prey density [31,32], identity [12], or relative abundance in the environment [33]. Finally, the effects of pollinator diversity on pollination depend on the plant's breeding system and the functioning of key pollinator species, which can change in different habitats or landscape contexts offering different resources [34].
Theoretical studies suggest that spatial heterogeneity in structure or resources can modulate the strength of the diversity-productivity relationship [23], and such heterogeneity has long been known to promote coexistence through resource partitioning among species [35]. As biotope availability (the physical space associated with a species niche) increases along spatial gradients, species may complement each other by occupying slightly different sections of the niche space [27,36]. It may therefore be expected that complementarity effects, whereby resource partitioning among species leads to increased total resource use, would be most strongly expressed in heterogeneous habitats with varied niches, thereby magnifying the effects of biodiversity on functioning (Figure 1). A rare empirical study to examine this possibility [37] found that the influence of species composition (rather than richness) on below ground biomass depended on small-scale heterogeneity in soil resources, suggesting that heterogeneity may modulate the diversity-productivity relationship in plant communities. Surprisingly, we know of no studies that have examined similar issues in a trophic context, despite the fact that natural enemies and pollinators are both known to be strongly influenced by spatial patterns of abundance and diversity of their respective resources [38,39].
The dearth of empirical studies on heterogeneity effects may partly reflect the broader paucity of field studies examining the effects of diversity on productivity or consumption in natural plant or animal assemblages, which are likely to vary in underlying environmental conditions [40] (but see [11,22,41]). This hinders general extrapolation of BDEF results from rigorous experimental studies with low spatial heterogeneity to natural and more complex ecosystems, within which biodiversity loss initially prompted the concern [7,17,40,42]. In fact it is quite likely that the full functional significance of biodiversity will only appear at larger spatial and temporal scales [36,43], and a critical future challenge is to unravel the ecological conditions under which biodiversity has its greatest effect on functional rates [40,42].
Here we show that within-habitat resource heterogeneity strongly moderates the recently demonstrated BDEF relationship in semi-natural and natural ecosystems [11,15,41]. The robustness of this result is underscored by testing diverse functions (belowground standing plant biomass, rates of parasitism, and crop pollination) in different ecosystems (temperate grasslands and a range of tropical habitats on two continents).
In temperate grasslands, plant diversity (i.e., Shannon index; see Methods) was significantly positively correlated with belowground standing biomass (F1,15 = 8.98, p = 0.009, see Text S1A for full parameter values), although no significant predictors remained in the model for aboveground standing biomass (Text S1B). In the minimal adequate regression model for belowground biomass, one of the five estimates of soil heterogeneity, Factor 5 (from a principle component analysis [PCA] to reduce 12 soil heterogeneity variables to five orthogonal factors, see Methods, Table S1), was also significantly positively correlated with belowground standing biomass (F1,15 = 6.86, p = 0.019), and most importantly, the interaction between the effects of plant diversity and soil heterogeneity (Factor 5) was significant (F1,15 = 8.50, p = 0.011) (Figure 2A). This positive interaction was due to the relationship between plant diversity and belowground standing biomass becoming more positive in heterogeneous soils, with the minimum adequate regression model (containing these two variables and their interaction) explaining 41.6% of the variance in belowground standing biomass. It may be argued that two models (one for aboveground, one for belowground biomass), each with 11 predictor variables (plant diversity, five soil factors, and their interactions) may lead to an inflated Type I error rate. Although this risk is often ignored in multifactor models, the probability of obtaining multiple significant effects can be easily calculated using a Bernoulli process [44]: p = [N!/(N − K)!K!] × αK(1 − α)N−K, where N = the number of “tests” and K = the number of tests below α. Therefore, given 22 tested predictors including interactions (11 per model for above- and belowground biomass), the probability of obtaining three predictors significant at p < 0.02 (as we did) by chance is a very low p = 0.00839, giving us confidence that these effects were not spurious. The congruence between the variables for which we found significant results and previous studies (see below), gives us further confidence that these relationships are not due to chance alone.
The significant soil heterogeneity factor (Factor 5) was largely driven by heterogeneity (coefficient of variation [CV]) in NO3− concentrations (Pearson's r = 0.68), and also by soil carbon and phosphorus concentrations (see factor loadings, Table S2). Importantly, it has been found recently that complementary strategies of inorganic soil nitrogen use among different plant species may lead to a positive biodiversity-productivity relationship in the same N-limited grasslands investigated here [45]. Effects of heterogeneity were not confounded by higher plant diversity in heterogeneous soils, as plant diversity weakly decreased with increasing soil heterogeneity (simple regression: F1,17 = 5.38, p = 0.033, r2 = 0.240; Tables S6 and S7). Nutrient availability is known to affect the BDEF relationship in plants [25,28], and it may be argued that nutrient concentrations co-vary with heterogeneity. However, our significant soil heterogeneity factor was not significantly correlated with the concentration of any of the soil variables examined (Figure 3A, Text S2, and Table S3). In the structural equation model (SEM) (Figure 3A), soil heterogeneity had the largest standardized total effect on belowground biomass (Table S7), with the diversity × heterogeneity interaction having the second largest effect. Despite being highly significant (Figure 3A), the size of the effect of plant diversity on biomass was relatively small compared to that of the other parameters (Tables S6 and S7). Although some paths leading to diversity (from three of the soil nutrient availability factors, the heterogeneity factor, and one of the composition axes) remained in the model (Figure 3A), none of these were significant at α = 0.05. Soil nutrient availability and plant species composition significantly affected biomass, although the effects of plant composition axis 1 and soil nutrient availability Factor 2 were entirely indirect, mediated via plant diversity. The SEM for aboveground biomass was unstable, so interpretation of the best fitting model (Figure 3B) must be made with caution. The only significant direct effects in the model were the effects of plant diversity on soil heterogeneity (which had the lowest standardized total effect on soil heterogeneity, Table S9), plant species composition axis 2 on the diversity × heterogeneity interaction, plant diversity on composition axis 1, and the effect of plant composition axis 1 on aboveground biomass. However, the total and standardized total effect of composition on biomass was lower than the effects of each of the soil variables that remained in the model, indicating that availability and heterogeneity of soil nutrients were more important predictors of aboveground biomass than were diversity or composition of the plant community.
For the ecosystems investigated in coastal Ecuador, we also found that rates of parasitism were significantly higher in plots with high natural enemy diversity compared with low diversity plots (Analysis of Covariance [ANCOVA]: F1,29 = 15.35, p < 0.001, Text S1C), congruent with previous work in our study region using a diverse host guild [11]. Parasitism rates did not vary significantly across habitat types (ANCOVA: F4,29 = 2.41, p = 0.072). Most importantly, however, the strength of the positive effect of parasitoid diversity on parasitism rates increased as within-plot spatial heterogeneity in host abundance increased (ANCOVA, interaction effect: F1,29 = 5.20, p = 0.030) (Figure 2B). This correlation between the BDEF relationship and heterogeneity did not vary significantly across habitat types (ANCOVA, three-way interaction: F4,29 = 0.99, p = 0.427). Host heterogeneity did not affect rates of parasitism directly in the ANCOVA (F129 = 1.40, p = 0.247), instead the effect was mediated indirectly via its interaction with the diversity-parasitism relationship. The overall model explained 55% of the variance in naturally occurring rates of parasitism. The effects of parasitoid diversity were not confounded by effects of host heterogeneity on diversity, as this link was not included in the best fitting structural equation model (Figure 3C; see also Figure 2B). Heterogeneity (CV) in host abundance was also not affected by habitat type, as habitat was not included in the best fitting SEM (Figure 3C). The SEM showed that parasitoid species composition (nonmetric multidimensional scaling [NMDS] axes 1 and 2, see Text S3) had positive effects on parasitism rates that were comparable in magnitude to those of species diversity (Tables S10 and S11), although diversity did have a strong effect on composition axis 1. This indicates that the effect of changing parasitoid diversity on parasitism may be at least partly mediated by changes in parasitoid community composition. The diversity × heterogeneity interaction also had a significant effect on parasitism rates (Figure 3C, Tables S10 and S11). Therefore, even when host abundance and parasitoid composition were considered in the model, parasitoid diversity still positively affected parasitism rates, and this effect became stronger when hosts were heterogeneously distributed.
In Indonesia, agroforests with high bee diversity had higher rates of coffee pollination (Multiple regression: F1,20 = 8.95, p = 0.007, Text S1D) [15], and again, the slope of this relationship increased with increasing spatial heterogeneity in the density of coffee flowers within plots (multiple regression, interaction effect: F1,20 = 9.42, p = 0.006), even though flower heterogeneity had no significant main effect on pollination (multiple regression: F1,20 = 1.02, p = 0.325) (Figure 2C, see also [15]). The overall model explained 49.2% of the variance in enhanced fruit set. These results were reiterated by the SEM (Figure 3D), where the largest standardized total effect on pollination benefit was the interaction between pollinator diversity and flower heterogeneity, with the second largest effect attributed to pollinator diversity (Table S13). Pollinator species composition (NMDS axis 3) also had an indirect effect on pollination, mediated through diversity (Tables S12 and S13). Heterogeneity in coffee flowers had no significant direct effects on pollination, although it was affected by floral abundance (Figure 3D, Tables S12 and S13).
In three different systems (temperate grasslands, host-parasitoid assemblages, and coffee pollinators) across three continents, the relationship between diversity and an ecosystem function became steeper in habitats with spatially heterogeneous distributions of an essential resource. This striking convergence across varied systems and functions indicates a highly robust pattern.
Working at large scales with natural diversity gradients has several limitations, so we needed to consider potentially confounding factors [46]. Because spatial heterogeneity can be an important factor driving patterns of diversity, one issue is that the two are often inextricably correlated, complicating attempts to treat them as independent predictors. In addition, previous models have shown that the direction of the BDEF relationship can depend on whether local heterogeneity or regional species pool differences underlie diversity differences, with positive BDEF relationships only expected in the latter case [26]. In no case did we find a significant positive relationship between local resource heterogeneity and species diversity, suggesting that heterogeneity did not underlie the effects of diversity in the three systems examined. This allowed a valid examination of how broader scale (site-to-site) differences in diversity interact with smaller scale variation in resource abundance (within-site heterogeneity) to determine functional processes in each focal system. Our finding that the slope of the BDEF relationship increased with resource heterogeneity could also not be attributed to the potentially confounding, yet commonly observed, influence of resource abundance on function (e.g., [25,31]), because the standardized total effect (the sum of direct and indirect effects, while controlling for all other variables in the model, standardized for differences in the magnitude of different units) of all of the resource abundance measures on function was lower than the diversity × heterogeneity interaction in the SEMs from all three systems (Tables S7, S11, and S13). Finally, it may be argued that in real world ecosystems, species composition will co-vary with diversity, potentially confounding BDEF relationships. We tested changes in community composition (Text S3), and found that this was not confounding the effects of diversity in our system (Figure 3, Text S4). Nevertheless, nonrandom extinctions and altered community composition may be an additional mechanism through which biodiversity loss affects ecosystem functioning [7,40], particularly if sampling effects are also important in real systems [17]. This was supported by the composition-mediated effects of diversity in the parasitoid system, and species composition additionally contributed to differences in belowground biomass for the plant communities (Figure 3A, Tables S6 and S7).
Thus, after accounting for likely confounding factors, the data still strongly support a positive diversity-function relationship, which increases when resources are heterogeneously distributed. Distinguishing between the mechanisms underlying positive diversity effects (e.g., selection, facilitation, or complementarity) is difficult in this kind of field study, and requires future research to experimentally manipulate diversity and heterogeneity. However, given the strong influence of resource heterogeneity in mediating the BDEF relationship, complementarity seems to be a likely candidate, meriting discussion in the context of each of the three systems examined.
In the temperate grassland system, the effects of plant diversity on belowground standing biomass became most evident in ecosystems where essential nutrients (particularly NO3−, P, and C) were heterogeneously distributed, which is the norm in most ecosystems [47], and allows complementary exploitation of the resulting niche spaces [45]. The contrasting patterns between above- and belowground biomass may be explained by root foraging patterns that make belowground biomass particularly susceptible to nutrient heterogeneity [47]. In contrast, aboveground biomass may respond more to absolute nutrient concentrations [25,47] than to heterogeneity, and this was supported by the best fitting SEM for aboveground biomass (Tables S8 and S9), where each of the soil nutrient availability factors had greater total and standardized total effects on biomass than did soil heterogeneity (or the diversity × heterogeneity interaction, which had no effect). Therefore, our finding that belowground biomass responded more strongly to diversity under high heterogeneity in NO3− is congruent with previous experiments.
Parasitoids have long been known to respond strongly to heterogeneity in the availability of underlying host resources (e.g., [48,49]), and here too, responses are often highly species specific. For example, different parasitoid species attacking a shared host can differ in their density dependence, with some species parasitizing a greater proportion of hosts at low densities and others showing the opposite pattern [39,50]. These different behavioral responses of parasitoid species to host density likely reflect differences in species characteristics such as dispersal ability, degree of specialization and search behavior [39,50]. In the host-parasitoid system of coastal Ecuador, we found that the positive effect of diversity on parasitism rates increased with increasing within-plot spatial heterogeneity in host abundance. Again this is consistent with the idea that different species were more effective at exploiting subpopulations at different densities resulting in complementary resource use in heterogeneous systems.
Different pollinator groups have been shown to be attracted by different floral designs, forms, abundances, and densities, and this can depend largely on their foraging behavior [38,51]. For example, Thompson found that butterflies and hover flies responded strongly to the number of open flowers in a patch (i.e., resource availability) [38], whereas other pollinators such as short-tongued bees responded more strongly to other floral characteristics. Similarly, mass recruiting species such as honeybees (Apis) are known to respond quickly to changes in floral resource availability through communication between workers, and tend to be attracted to patches of high resource abundance or density [52,53], while bumblebee (Bombus spp.) workers forage more independently and may be more likely to visit rarer or more sparsely distributed resources. Although coffee plants attract a diversity of solitary species, social bees such as honeybees and stingless bees are almost absent when flowering is scarce [34]. Displacement by aggressive social bees of inferior competitors from large floral resource patches may also cause rarer solitary bees to utilize smaller patches that are less attractive to their competitors [54]. Given that different pollinator species can vary in their foraging behavior [55] and responses to patterns in the floral community, increasing pollinator diversity would be expected to result in more complete utilization of host plant patches and thus greater function in heterogeneous systems. This is consistent with results from our coffee agroforest system, in which the slope of the relationship between bee diversity and rates of coffee pollination increased with increasing spatial heterogeneity in the abundance of coffee flowers.
Alternative approaches to studying the relationships between biodiversity and processes or functions have often led to contrasting results, with a positive relationship generally found in smaller scale manipulative experiments and more variable or negative relationships found in larger scale observational studies [56–58]. Thus, while experimentalists argue that effects of biodiversity can only be understood via direct manipulations of diversity (indeed such experiments are obviously critical for testing certain causal mechanisms, such as facilitation or complementarity), others have begun to question the relevance of such experiments for assessing the importance of biodiversity in real world systems [40,56]. Ultimately, such polarization is inimical to ecological progress, and full understanding of real world systems and their underlying mechanisms requires a suite of theoretical, empirical, and observational techniques [59].
Our results are congruent with predictions from theory, previous experiments, and recent field studies, showing that biodiversity can have strong effects on ecosystem functioning even in real, nonexperimental habitats. Our study presents statistical correlations and can therefore not be used to infer causation or effects of biodiversity loss per se (differences in diversity among our sites may reflect factors other than species losses). Nevertheless, our results provide strong empirical evidence from different ecosystems that the BDEF relationship can depend on the habitat context, requiring resource heterogeneity for a positive BDEF relationship to occur. Previous controlled experimental studies using random species assemblages without consideration of environmental heterogeneity, may not have contained the varied niche space required for full expression of diversity effects [36]. Therefore the positive relationships observed in experiments may have even been underestimates of the strength of the BDEF relationship, and its importance to real world ecosystems, and future experiments that explicitly consider heterogeneity are needed.
Recent years have seen enormous strides in understanding of the BDEF relationship. Our results provide generality, by elucidating the real world conditions under which previously established patterns are likely to occur. Combined with calls from theoretical work, our results suggest that expanding research to include more realistic conditions, such as heterogeneous habitats or resources, will be critical towards further understanding the BDEF relationship. Only a synergy between theoretical, experimental, and observational approaches will be able to untangle the real world importance of diversity for ecosystem functioning, and allow us to fully understand the perils of biodiversity loss.
The study was conducted in 19 semi-natural grasslands in the Thüringer Schiefergebirge/Frankenwald, in central Germany [41]. In each grassland, four soil samples were collected in mid May 2002 to characterize soil spatial heterogeneity. Part of the sample (∼10 g) was extracted with 50 ml 1 M KCl for 60 min on the day of sampling (May 15, 2002). KCl extracts were filtered and frozen at −20 °C and later analyzed using a continuous flow analyzer (SAN Plus, Skalar) for NH4+ and NO3−, and an ICP-AES (Optima 3300 DV, Perkin-Elmer) for Ca2+. The remaining soil was dried at 35 °C and extracted for 1 h using a 1 M calcium acetate lactate (CAL) solution. CAL extracts were analyzed with ICP-AES for P, K+, Mg2+, and SO42−. Soil pH was measured in a water extract. To measure soil C:N ratio, total N and total C concentrations, dry soil was ground and analyzed with an elemental analyzer (Vario EL II, Elementar).
A 3 × 3-m area was sampled in each grassland for plant biodiversity [41]. We followed Kahmen et al. [41] by calculating plant diversity (exponential Shannon-Wiener) as H′ = −Σ(pi)(ln pi), where pi = species cover/sum of cover for all species. Aboveground standing biomass was determined in eight 25 × 50-cm plots per grassland. Samples were collected in late June and early September 2003, following the local management regime. Biomass was clipped 2 cm above the ground. Belowground standing biomass was determined using soil cores (4.3 cm diameter, 10 cm length). Four cores were collected in the central sampling area of each grassland, at the end of the growing season (mid September). The roots were removed from the soil, dried, and weighed.
Soil heterogeneity was characterized by the within-plot CV (the standard deviation as a proportion of the mean) of each of the soil variables (NH4+, NO3−, Ca2+, P, K+, Mg2+, Na, C:N ratio, total N, Nmin, total C, and pH(H2O)). As there were a large number of intercorrelated soil variables relative to our number of replicates, and to alleviate the need for subjective decisions about which soil characteristics may be important, we used a PCA in Statistica 6.1 (Statsoft) to reduce these 12 variables to five factors with an eigenvalue greater than 1, which cumulatively explained over 80% of the variation in soil heterogeneity (Tables S1 and S2). PCA is a technique for simplifying a multidimensional dataset by reducing it to fewer orthogonal dimensions for analysis, while retaining those characteristics that contribute most to its variance. This same technique was used to reduce the absolute values of each soil variable (i.e., not the CV) to four orthogonal factors (Tables S4 and S5) for structural equation modeling.
The 48 study plots were spread across three cantons in the region of Jipijapa (17 N 546,800 m, E 984,9274 m altitude 259 m), within the province of Manabi, Southwest Ecuador (for individual plot location details and a full description of the region see [60]). This area falls within the Choco-Manabi biodiversity hotspot, but large-scale agricultural conversion threatens the local biodiversity and the ecosystem services it provides [11]. There were 12 rice, 12 pasture, 12 coffee, six abandoned coffee, and six forest plots.
Nine standardized trap nests were positioned (in a 3 × 3 grid, 25 m between adjacent traps) in the center of each of the 48 plots, to provide nesting sites for naturally occurring communities of bees and wasps (hosts) and their natural enemies [11,60]. A PVC tube with a length of 22 cm and a diameter of 15 cm formed the outer case of the nest. Internodes of reeds Arundo donax L. (Poaceae) with varying diameter (2–20 mm) and a length of 20 cm were inserted into this tube and provided the nesting sites for bees and wasps. Exposure of standardized trap nests is similar to the exposure of other resources, e.g., phytometer plants, but because the guild of aboveground cavity-nesting species reproduces in these traps, the problem of species appearing as “tourists” in samples is eliminated. The traps were evaluated every month from June 2003 to October 2004. Occupied reeds were opened and the host larvae were reared to maturity for positive identification and to detect the presence of any parasitoids. Parasitism rate was defined as the proportion of host larvae attacked by parasitoids. Data from each of the nine traps per plot were pooled across all months for analyses.
Of all the potential host species found in our traps [60], the one with the greatest diversity of parasitoids attacking it (six species) was Pseudodynerus sp. (Hymenoptera: Eumenidae). This was the only host species found in every plot (n = 48), it was the second most abundant of all the host species (6,884 individuals, 487 of which were parasitized), and had the most even distribution among habitat types—allowing statistical analyses in all habitats. We therefore used this species as our focal host for this study. The natural enemies attacking Pseudodynerus were gregarious ectoparasitoids, solitary ectoparasitoids, endoparasitoids, or kleptoparasites [11], but here we refer to them collectively as “parasitoids.”
To quantify host heterogeneity (patchiness) in each plot, we used the CV in host (Pseudodynerus) abundance between the nine trap nests within one plot. Abundance was measured as the number of Pseudodynerus larvae in all occupied reeds over the entire sampling period.
The 24 study plots were spread across the agricultural landscape of the villages Wuasa, Watumaeta, Alitupu, and Kaduwaa at the margin of the Lore-Lindu-National Park, Central Sulawesi (Indonesia) (for site details see [15]). Sulawesi is a biodiversity hotspot, including many endemic species in and around the Lore-Lindu-National Park, but ongoing agricultural conversion at the forest margins threatens the endemic species and the local diversity.
We randomly selected four coffee (Coffea arabica L.) plants in each of the 24 coffee agroforests, and used one branch per plant for an open pollination treatment, and another one for a bagged pollination treatment. Bags of nylon mesh gauze (10 μm) were fixed on the bagged treatment coffee branches 1–6 d before blooming, to exclude pollen transferred by insects or wind. Tanglefoot was applied on the branch beneath the bagged flowers to exclude ants. We counted and tagged flowers on the observed bagged and open branches. After flowering, we removed the bags, and 5 wk later developing fruits were counted (see also [15]). As fruit set can vary significantly with plant quality and local soil and microclimatic conditions, we defined pollination benefit as the proportion of flowers that set fruit from the open pollination treatment, minus the proportion that set fruit in the bagged control treatment. This controlled for between-plant variability in fruit set due to genetic or environmental factors.
Because of time constraints during the short flowering period, flower-visiting bees and resulting fruit were sampled on only three of the four coffee shrubs in each site, following the methodology of Klein et al. [15]. We counted the flowering branches on each of these three shrubs, and the between-shrub CV in these values was used as an estimate of resource heterogeneity. We collected flower-visiting bees for 25 min on each of the three coffee plants, and further estimates of plot-scale bee diversity were obtained by sweep-netting for 5 min.
Unless stated otherwise, analyses were conducted in R v. 2.3.1 (R Development Core Team, http://www.r-project.org). Model residuals were tested for adherence to a normal distribution and homogeneity of variances, and response variables were transformed when necessary.
We tested the effects of plant diversity and soil heterogeneity on above and belowground standing biomass using multiple regressions. The initial (maximal) model contained standing biomass as the response variable, and plant diversity and all five soil heterogeneity factors from the PCA were included as continuous predictors. Interaction effects between plant diversity and each of the soil heterogeneity factors were also included in the model. Separate models were used for above and belowground biomass. Models were simplified by removing nonsignificant interaction terms then main effects, until model fit (assessed using Akaike Information Criterion [AIC]), no longer improved. When model fit did not differ significantly between two competing models (the difference in AIC score was <2), we selected the most parsimonious model (the model with fewest parameters). Full details of maximal and minimal adequate above- and belowground models are provided in Text S1A and S1B, respectively.
The proportion of Pseudodynerus larvae parasitized per plot was arcsine square root transformed prior to analyses to meet assumptions of the parametric tests. We tested the effects of habitat type, natural enemy diversity, and host heterogeneity on rates of parasitism using ANCOVA, with Type I sums of squares. No model simplification was needed, as we only included one measure of heterogeneity, compared with five factors in the plant model above. Habitat type entered the model first, then parasitoid species richness, then host heterogeneity (within-plot CV in abundance), followed by the main interaction of interest (host heterogeneity × parasitoid species richness interaction effect), and a three-way interaction between host heterogeneity, parasitoid richness, and habitat type (to determine if the mediating effect of host heterogeneity varied across habitats). To be conservative, we excluded all zero values from analyses, as incidents of zero diversity and/or zero parasitism could drive a positive diversity function relationship by default, and their inclusion only made the effects presented in the results and graphs even stronger.
We tested the effects of bee diversity and flower heterogeneity on pollination success (percent fruit set of open-bagged control flowers) using a multiple regression with Type I sums of squares. Bee richness entered the model first, followed by flower heterogeneity and the interaction of these two variables.
In addition to the above analyses, we applied NMDS ordination techniques to test for correlations between NMDS axes (community composition of plants/parasitoids/pollinators) and diversity in each of our systems. These analyses (Text S3) showed that species composition did not confound the effects of species diversity.
Although the ANCOVA and multiple regressions above tested our hypothesized effects, it was also possible that diversity responded to another variable that also has an effect on function. Similarly, resource abundance may have had complex indirect effects, possibly mediated through diversity, which was also correlated with species composition in the parasitoid and pollinator communities. Therefore, to control for these possible confounding variables, we used SEMs, performed in Amos v.16.0.1 (Amos Development Corporation, http://amosdevelopment.com). We included the above NMDS axes along with resource abundance, heterogeneity, and diversity variables in the SEMs. For each system we constructed an initial model with a variety of pathways allowing resource abundance and heterogeneity to affect diversity, composition, and function. We then simplified these models down to a final model using AIC scores (Text S4), and present the final simplified models in Figure 3.
|
10.1371/journal.pgen.1002244 | Separation of Recombination and SOS Response in Escherichia coli RecA Suggests LexA Interaction Sites | RecA plays a key role in homologous recombination, the induction of the DNA damage response through LexA cleavage and the activity of error-prone polymerase in Escherichia coli. RecA interacts with multiple partners to achieve this pleiotropic role, but the structural location and sequence determinants involved in these multiple interactions remain mostly unknown. Here, in a first application to prokaryotes, Evolutionary Trace (ET) analysis identifies clusters of evolutionarily important surface amino acids involved in RecA functions. Some of these clusters match the known ATP binding, DNA binding, and RecA-RecA homo-dimerization sites, but others are novel. Mutation analysis at these sites disrupted either recombination or LexA cleavage. This highlights distinct functional sites specific for recombination and DNA damage response induction. Finally, our analysis reveals a composite site for LexA binding and cleavage, which is formed only on the active RecA filament. These new sites can provide new drug targets to modulate one or more RecA functions, with the potential to address the problem of evolution of antibiotic resistance at its root.
| In eubacteria, genome integrity is in large part orchestrated by RecA, which directly participates in recombination, induction of DNA damage response through LexA repressor cleavage and error-prone DNA synthesis. Yet, most of the interaction sites necessary for these vital processes are largely unknown. By comparing divergences among RecA sequences and computing putative functional regions, we discovered four functional sites of RecA. Targeted point-mutations were then tested for both recombination and DNA damage induction and reveal distinct RecA functions at each one of these sites. In particular, one new set of mutants is deficient in promoting LexA cleavage and yet maintains the ability to induce the DNA damage response. These results reveal specific amino acid determinants of the RecA–LexA interaction and suggest that LexA binds RecAi and RecAi+6 at a composite site on the RecA filament, which could explain the role of the active filament during LexA cleavage.
| Genetic material is under constant environmental assault. The bacterial recombinase protein RecA is pivotal to DNA repair [1]–[4] and to orchestrate the bacterial DNA damage response (SOS response) against natural, or drug-induced, genotoxic conditions. It is part of an ancient and evolutionarily widespread protein family and, except for a few endosymbionts [5], homologs carry out related functions in archaea [6] and eukaryotes [7], and in some cases mutants are linked to human cancers [8], [9].
To perform its many roles, RecA interacts with multiple partners in E. coli [3]. It normally exists in an inactive conformation without bound DNA [10], [11]. Upon DNA damage, an essential first step is the RecA polymerization around a single stranded DNA (ssDNA) in an ATP-dependent fashion [12]–[14]. In this active filament form, it can direct homologous recombination [15], bind to DinI [16], [17] and RecX [18]–[20] to control filament growth [21], [22], and bind the RecFOR complex to repair ssDNA breaks [23]–[25]. RecA is also a co-protease that promotes cleavage of the transcriptional repressor LexA [26] to trigger the expression of over 40 SOS response genes [27]. It also promotes cleavage of UmuD [28]–[31] to become a constituent of the error-prone DNA polymerase V (pol V) [32], [33], in addition to direct interaction with pol V for its activity [33]. Alternately it also interacts with another Y family of DNA polymerase, DinB (pol IV), to directly modulate its mutagenic potential in the translesion DNA synthesis [34]–[36]. It also promotes cleavage of the phage repressor, λCI, triggering induction of the lytic cycle [37], [38]. Every one of these interactions is a potential target to design drugs or mutants that dissect the molecular basis of RecA-dependent genomic repair and stability.
There are many crystallographic structures of RecA, or homologs, but most do not include bound DNA, and so are thought to represent the inactive conformation [39]–[50]. More recently, the crystal structure of E. coli RecA bound to DNA in the active conformation was solved (hereafter PDB:3cmx) [51]. It showed the ATP binding site, the DNA binding site and RecA-RecA interfaces in a likely active form (Figure 1A). Still, the interaction sites for other partners (such as DinI, RecX, RecFOR, LexA, UmuD, UmuD2C, DinB and λCI, as mentioned above) remain unknown. Separately, several mutational studies sought to identify residue determinants of diverse RecA functions, but without yet producing a fully coherent view [52].
To investigate the biological roles of known structural sites and to discover other RecA functional sites, we turned to the Evolutionary Trace (ET). This phylogenomic method [53]–[55] ranks a protein's residues by relative evolutionary importance. A structural map of the top-ranked residues then reveals clusters that indicate active sites and binding sites on the protein surface and that efficiently guide site-directed mutations that block, separate, or rewire functions in eukaryotic proteins [56]–[68]. ET analysis revealed many clusters of top-ranked residues on the E. coli RecA surface, which were targeted for mutagenesis followed by functional analysis. This extended and confirmed the biological role of the interfaces revealed in the inactive and active filament structure [51] and, critically, revealed new sites in other regions where mutations separated recombinase activity from co-protease activity for LexA cleavage. Two structurally distant amino acids (G108 and G22) are linked to the RecA-LexA interaction, and their location on RecA subunits i and i+6 apart in the helical active filament, across the groove, suggests a constraint on a low-resolution, illustrative model of the LexA-RecA interaction.
In order to identify novel, biologically relevant functional sites in the E. coli RecA protein, ET analysis was performed on 201 RecA homologs of bacterial origin. Each residue sequence position was ranked by ET based on how well its variations among homologs correlated with phylogenetic divergences (Figures S1 and S2) [56], [69]. Residue positions ranked in the top 40th percentile rank (thereafter ET40) were mapped onto the monomer of the RecA crystal structure, in the active form [51] (Figure 1B, shaded red and maroon). ET40 residues formed statistically significant clusters, with a z-score of 1.9, and suggested a number of functional surfaces, including as expected known sites such as the RecA-ATP interface, RecA-DNA interface and the two RecA-RecA interfaces (Figure 1A and contoured with a thick line in Figure 1B).
One area of interest includes a cluster of ET40 residues that borders the RecA-RecA interface in the inactive structure (residues highlighted in cyan in Figure 1B) but within the RecA-RecA, RecA-DNA interfaces in the active structure (Figure 1A and 1B). It includes residues E123, E154, L126, G212, G165 and A168. The structural data and the ET rank of these residues suggest they may be functionally important for oligomerization, although no experimental evidence has indicated such a role. It is likewise for residues S172, R176 and Q173, which are within the RecA-RecA structural interface common to both active and inactive structures (residue positions shown in Figure 1B). All of these residues were therefore grouped together as the extended RecA-RecA/DNA interface patch and chosen for mutational and functional analysis, described below.
Besides this interface patch, other ET40 residues formed various other clusters elsewhere on the RecA structure. These were named, arbitrarily, ET site-1 (D224, R226 and K245), ET site-2 (G288, Q300 and N304), ET site-3 (G87, K88 and G108) and ET site-4 (G22, K23 and G24) (residue positions shown in Figure 1C). Since each of these sites suggest a new putative structural interface without any known function, site-specific mutagenesis was performed to probe their function. For all site-directed mutagenesis, amino acids were individually mutated to alanine unless the alanine substitution already existed in a member of the ET sequence dataset. For such exceptions, tyrosine, tryptophan or glycine residues were used depending on their absence from the substitution profile of the targeted position. All mutations were constructed on a low-copy plasmid-borne recA gene and transformed into a ΔrecA E. coli strain [70]. The mutant RecA strains were tested for their UV sensitivity to assess the global impact on RecA function. Representative mutant strains from each ET clusters were also tested for their sensitivity to mitomycin C to demonstrate that the survival phenotypes of these mutants were not specific to UV induced DNA damage (Figure S3). Then, to pinpoint the molecular basis of UV sensitivity, both a P1 recombination assay and a LexA western blot assay were performed to probe the recombinase activity and the induction of LexA cleavage of each RecA variant in vivo, respectively.
Finally, to validate our ET analysis on RecA, several poorly-ranked ET residues located on the RecA surface (in the worst quartile of importance) were analyzed through site-directed mutagenesis and functional analysis as described above. Such bottom-ranked residues near the known RecA interfaces included T89, N181, N186 and V238, and others that were away from any known RecA interfaces included K294 and N312 (Figure 1B, shown in blue letters). As expected, mutation of these residues displayed no UV sensitivity (Figure 2A), had relatively intact recombination efficiencies that ranged from 56 to 72% compared to wild-type RecA strain as judged by P1 recombination assay (Figure 2B) and were all fully capable of inducing LexA cleavage leading to upregulation of RecA protein (Figure 2C).
In order to test the functional role of the extended interface patch residues (Figure 1B), site-directed mutagenesis was performed. As expected from interference with RecA-RecA or RecA-DNA interactions, either of which would disrupt the basic ability of RecA to form active nucleoprotein filament, these RecA mutant strains within the interface patch were strongly sensitive to very low doses of UV damage (Figure 3A) similar to the empty vector in a ΔrecA background (Figure 2A) with the exception of the Q173A substitution. As a positive control, the E. coli strain with wild-type RecA overcame UV damage (Figure 2A).
To characterize the recombination efficiency, a P1 transduction assay was performed. All variants, except for Q173A, are disrupted for recombination similarly to the ΔrecA strain (Figure 3B). Likewise, these variants, significantly hindered RecA's ability to promote LexA cleavage upon DNA damage and subsequently failed to up-regulate RecA (Figure 3C). The observation that Q173A mutation showed no effect on RecA activity could be attributed to the lesser importance of this residue, which has the worst rank of the ET40 residues in this patch (30th percentile-rank).
Taken together, these mutations confirmed that top-ranked ET40 residues in this extended interface patch impair both the recombination and co-protease activities of RecA, and thus are crucial for UV survival. This is consistent with the structural data on the active RecA filament [51]. These residues are directly involved in RecA-RecA and RecA-DNA interaction, and their mutations are thus likely to interfere with the basic assembly or working of the nucleoprotein RecA-DNA filament.
Two of three ET site-1 RecA variants (R226A and D224A) are UV-sensitive (Figure 1C and Figure 4A). These mutations strongly disrupt recombinase activity of RecA (Figure 4B) and LexA self-cleavage (Figure 4C), similarly to the extended interface patch variants. Assuming that RecA folding is not affected, and given the structural contiguity to the RecA-RecA interface-1 (see Figure 1A and 1C) one reason may be some involvement in RecA-RecA interaction and filament formation. Another possibility is that ET site-1 could play a role in binding to proteins such as RecX, DinI and RecFOR that modulate RecA function.
Mutational analysis of ET site-2 residues (Figure 1C) showed separation of RecA function. First, two of the three mutant strains have abnormal UV sensitivity (Figure 5A). The N304D variant was the most sensitive, followed by Q300A. The G288Y variant displayed no UV sensitivity. Next, all three mutant strains had reduced recombination efficiency (Figure 5B), with the N304D variant being as deficient as the ΔrecA strain. Finally, LexA cleavage upon DNA damage was intact (Figure 5C), suggesting that the RecA folding and active filament formation required for SOS induction were unaffected. Thus, overall, all these ET site-2 mutations have more severe impact on the recombinase activity than on the SOS response. The N304D mutant, which is completely defective for recombinase activity, displays the clearest example of separation of function. Therefore, these data suggest that the ET site-2 is essential for the recombinase function of RecA. One explanation is that this site may bind to the dsDNA or to other partners essential for recombination events.
Mutagenesis of ET site-3 (G87, K88 and G108) and ET site-4 (G22, K23 and G24) (Figure 1C) also produced partial separation of function. In contrast to mutations affecting ET site-2 residues, these variants displayed no UV sensitivity (Figure 6A), except for the G22Y variant, which only becomes UV sensitive at higher doses. All variants displayed lower recombination efficiency compared to wild-type RecA but none as complete as the ΔrecA strain. The efficiency was down approximately to 9, 19 and 25% for G87Y, K88Y and G108Y in ET site-3 and to 5, 11 and 22% for G22Y, K23Y and G24Y, respectively, in ET site-4 (Figure 6B). Among these residues, the G108Y (in ET site-3), G22Y and K23Y (in ET site-4) showed strongly reduced LexA proteolysis upon DNA damage (Figure 6C, highlighted in red). Strikingly, these three mutant strains exhibited 3.5 to 4.6-fold upregulation of RecA levels, consistent with SOS induction, even in the absence of LexA cleavage after DNA damage. These variants show some similarity to a well-known recA mutant, recA430 (corresponding to G204S) [71]–[73], which is only slightly affected for recombination but deficient for LexA cleavage. In our assays, this variant showed an increase in UV-sensitivity (Figure 6A, lowermost panel), with relatively intact recombination activity, and no ability to induce LexA cleavage. However, this variant did not up-regulate RecA, unlike the G108Y, G22Y and K23Y mutant strains (Figure 6C). Thus RecA upregulation without LexA cleavage upon DNA damage is unique to our three mutant strains. Finally we asked whether this defect in co-protease activity was specific to LexA by testing another substrate of RecA, UmuD, which is also activated upon LexA cleavage. In this case, the active RecA filament mediates UmuD autoproteolysis yielding UmuD'. Since UmuD cleavage induction is a later SOS response than that of LexA, it was analyzed at later time points. Upon DNA damage, we observed an upregulation of UmuD levels in G108Y mutant strain and also a slight upregulation in G22Y and K23Y mutants respectively (data not shown). To analyze UmuD cleavage, we used a lexA (def) E. coli strain which is constitutive for UmuD expression. Formation of UmuD', the cleavage product of UmuD was visible in G108Y and G22Y mutant strains indicative of self-proteolysis of UmuD induced by RecA (Figure 6D). However in the G24Y mutant, that was shown to cause a slow LexA cleavage, there was a robust upregulation of UmuD like wild-type RecA (data not shown). In addition, the recA430 mutant strain, in our assay, could not cleave UmuD as well (Figure 6D). The upregulation of UmuD and its subsequent cleavage to UmuD' in G108Y, G22Y and to some extent in K23Y, strengthens the possibility that these variants alter most likely RecA-LexA interaction rather than affecting the overall co-protease activity of RecA.
The unexpected upregulation of RecA without LexA cleavage after DNA damage could suggest that LexA is sequestered by active RecA filaments, leading to SOS induction. Specifically, mutation of just one of the two residues, G108 or G22, could leave the ET-site with the other residue intact and able to bind LexA to RecA, effectively titrating it away from transcriptional repression irrespective of cleavage.
To test the possibility that LexA cleavage induction might require binding RecA at G108 and G22 residues at the same time, we made the double mutant G108Y/G22Y. We reasoned that with both ET-sites 3 and 4 mutated, LexA could not bind to RecA anymore, allowing it to repress the SOS response. In our assays, the double mutant (G108Y/G22Y) was weakly sensitive to UV (Figure 7A) comparable to the recA430 mutant (G204S) (Figure 6A). The recombination efficiency of the double mutant was intermediate between G108Y and G22Y individual mutants (20% as that of wild-type RecA) (Figure 7B). The mutant also did not induce LexA cleavage (Figure 7C), but could still cleave UmuD to UmuD' although, less efficiently (Figure 7D). Importantly, RecA upregulation was much reduced compared to the individual mutants. This supports our hypothesis that a joint disruption at ET-sites 3 and 4, via double mutations at residues G108 and G22, impairs LexA binding and prevents its sequestration to RecA with subsequent upregulation of SOS genes. The similar impact of individual mutations at residues G108 and G22 and the synergy of their coupled mutations suggest that they may play joint roles in both LexA binding and subsequent cleavage.
This work identifies new surface exposed domains of RecA critical for its recombinase and LexA cleavage functions. The discovery of these residues with the Evolutionary Trace (ET) shows that this computational strategy, based on phylogenetically correlated sequence variations, applies equally well here in prokaryotes as previously in eukaryotes, and that it can efficiently identify key functional residues that evaded detection by several past studies on this highly mutagenized protein [52]. Finally, this work validates the functional role of recent crystallographic evidence for RecA-RecA and RecA-DNA interfaces [51], and suggests a model for the RecA filament-LexA interaction.
To confirm past observations at RecA functional sites and also to validate the ET strategy, we targeted for site-directed mutagenesis the top-ranked Evolutionary Trace residues at the interface defined in the active RecA-ssDNA filament structure [51]. This structure has a ∼12 Å shift of the RecA-RecA interface upon ssDNA binding compared to the inactive structure [39], and it now includes additional residues important for filament formation (G165, S172, R176 and G212 and E123, A168). Consistent with both the inactive and active structures, top-ranked ET residues significantly overlapped the RecA-RecA and RecA-DNA dimer interfaces (Figure 1A and 1B) and their mutations prohibit both recombinase and LexA cleavage activities (Figure 3). This is in line with previous mutations of neighboring residues with similar defects in recombinase or co-protease activities [74]–[84], and it establishes a functional role for the residues in the extended RecA-RecA interface observed structurally in the active RecA filament.
A second set of top-ranked ET residues, ET site-1 (D224, R226, K245), was found in the cleft region of RecA and adds details on the determinants of overall RecA function. This cleft is located in between two adjacent RecA monomers and was previously proposed to bind repressors [39], [79], [80], [85]–[87] and dsDNA [88], [89], possibly through several of the positively charged side chains [86], [89]. The ET site-1 overlaps this region and mutations of D224 and R226 eradicated both recombinase activity and SOS functions of RecA. This suggests that, as above, this site also takes part in forming a functionally active filament, possibly as a functional extension of the neighboring extended interface patch. The inactive RecA filament structure (PDB:1u99) [39] shows ET site-1 next to the RecA homodimerization site. In addition, the active filament structure (PDB:3cmt) [51] shows both R226 and D224 residues binding to the previously disordered DNA binding loop 2 (L2). In fact, the binding partner of R226 in L2 is Glu207, which is an absolutely conserved residue among 64 RecA enzymes [90], [91] and does not tolerate any amino acid substitution without some loss of function, as seen from saturation mutagenesis [74]. Thus, the severe impact of R226 and D224 mutations on both recombination and SOS induction is consistent with ET site-1 contributing to the formation of the active filament and, indirectly, to DNA binding.
Another set of top-ranked ET residues, ET site-2 (N304, Q300 and G288), is located in the RecA C-terminal domain (CTD). The CTD region of RecA has been previously implicated in recombinase function [79], [92]–[94], acting as a secondary DNA (dsDNA) binding pocket on the outer surface of the filament. Mutations of all three ET site-2 residues impair recombinase activity but not LexA cleavage (Figure 5B and 5C). These residues are in the edge of the filament groove, and might provide binding stability to dsDNA for its efficient uptake into the filament. Of note, mutation N304D showed a striking separation of function with a complete destruction of recombinase activity similar to ΔrecA strain. Such marked defect in recombinase function was previously reported by a point mutation involving Gly301 to Asp in the CTD [95], [96], suggesting the intolerance of negatively charged amino acid side chains in the RecA CTD in dsDNA binding during the recombination process. Alternatively, these residues might modulate interaction between RecA and DinI [97], RecX [22] or RecFOR proteins.
The induction of the SOS response by RecA-mediated cleavage of LexA has been extensively studied both in vivo and in vitro, yet the sites involved in the interaction of these proteins remain unclear. Our ET analysis reveals two new sites with some potential to be determinants of the RecA-LexA interaction. Mutation of these sites preserves recombination but in majority, inhibits LexA cleavage. Paradoxically, levels of the LexA-repressed proteins, RecA (Figure 6C) and UmuD (data not shown) were up-regulated upon DNA damage; indicating that there was SOS induction, independent of LexA cleavage. These mutants could promote UmuD cleavage; indicating that this defect in co-protease function was highly specific to LexA (Figure 6D). To our knowledge, the activation of the SOS response by UV, independent of LexA cleavage, has not been previously observed. Electron micrograph [20], [98], [99] and mutational studies [71], [72], [79], [80], [82], [83], [85]–[87] point to the binding of LexA, cI and UmuD structural homologs deep within the RecA filament's helical groove. However, structural elements on the edge of the helical groove including the dynamic N-terminal helix/strand (1–30) and C-terminal domain (270–333), have also been found to contribute to cleavage of LexA, cI and UmuD [79], [99]. Consistent with these findings, the residues that we find to be highly specific to LexA hydrolysis lie between the N-terminal α-helix A and β-strand 0 (G22 and K23) or adjacent to the CTD (G108). We propose that LexA binds across the RecA filament's groove through direct contacts at both of the two distant ET-sites 3 and 4, and that these sites cooperate to enable LexA proteolysis (see Figure 8A). Then, as observed, the disruption of either one could permit binding but not cleavage of LexA, leading to SOS induction without efficient LexA degradation. Previous mutational studies [74], [79], [82], [83] implicating residues facing the helical groove in repressor cleavage functions (Figure 8A, shown in magenta) are consistent with this model. Moreover, this model predicts that the simultaneous disruption of both LexA binding sites 3 and 4 would prevent LexA binding and sequestration. Indeed, the G108Y/G22Y double-mutant does not up-regulate expression of RecA (Figure 7C) and UmuD (data not shown). An alternative possibility would be that each point-mutation changes the conformation of the active RecA filament to prevent LexA cleavage. However, such an allosteric effect would have to be subtle since both the recombination function of RecA and its co-protease activity towards UmuD are still present in both the point-mutant and the double-mutant (Figure 7B and 7D). Nevertheless, the less efficient co-protease activity of the double mutant towards UmuD also suggests that the binding sites for UmuD might be shared among these residues or their neighbors, so that the double mutation either directly disrupts the efficiency of UmuD binding and/or cleavage or indirectly affects the protein fold for UmuD binding. This is in agreement with previous electron micrograph and mutational studies suggesting the possibility of repressors sharing similar binding sites on the RecA filament [79], [87], [98].
Consistent with our model of LexA binding to a composite site, a geometric docking analysis of LexA dimer binding to the RecA filament identifies, among many other possible solutions, one in which the LexA dimer binds to ET site-3 and ET site-4 from RecA units at positions i and i+6, or one helical turn apart, across the filament groove (Figure 8B). This illustrates how, by wedging itself into the groove, the DNA binding domain of LexA may bind the core of the RecA filament and at the same time allow the catalytic C-terminal domain of LexA to span the helical filament's edge. The model could be further addressed by direct assays measuring LexA binding and proteolysis in these mutant proteins in vitro. In the future, these RecA mutants may become a useful tool for trapping the RecA-LexA interaction towards efforts to obtain a co-crystal structure. Overall our results suggest that a cooperative binding at RecA residues G108 and G22 is essential for triggering LexA proteolysis.
In conclusion, ET identified new functional sites and efficiently guided their mutational validation in RecA. These sites form important new targets for future biochemical studies of RecA function, and may prove useful for creating separation of function mutants that will help dissect the network of interactions responsible for DNA damage repair. The emergence of bacterial resistance to antibiotics is mediated in part by the SOS response and it has been proposed that blocking the SOS pathway may prevent the evolution of bacteria in contact with these antibiotics [100], [101]. The new RecA sites identified in this work may become useful for the design of new drugs preventing the evolution of bacteria to antibiotics.
The low copy plasmid pGE591 containing wild-type recA [81] was a kind gift from Dr. George Weinstock, Washington University in St. Louis, MO, and the E. coli strain SMR6765 [70] lacking functional RecA was provided by Dr. Susan Rosenberg, Baylor College of Medicine, TX. E. coli strain CH458 (lacZYA::gfp-cat) was used as donor strain for P1 phage lysate preparation. Rabbit anti-UmuDD' polyclonal antiserum [102] was generously provided by Dr. Roger Woodgate (Laboratory of Genomic Integrity, NIH, MD).
The genotypes and sources of the E. coli strains and plasmids used in this study are listed in Table S1. Strains made in this study were constructed by classical P1 transduction [103].
The Evolutionary Trace analysis [104] used a sequence alignment consisting of 201 RecA protein sequences, nearly all bacterial, that have LexA or LexA homolog (Table S2). The primary source of the alignment was the HSSP database and it was retrieved using the Evolutionary Trace Report Maker Server [105]. Each sequence was BLASTed against the NCBI non-redundant protein sequences (nr) database and the sequences with at most 20 gaps or additions relative to the RecA sequence of E. coli were aligned using MUSCLE [106]. This dedicated alignment spanned greater evolutionary distances than the one provided automatically by the ET viewer software [107] (ET servers and viewing tools are available for public use at http://mammoth.bcm.tmc.edu/). The ET phylogenetic tree and multiple sequence alignment of RecA sequences in text and image formats are also available at http://mammoth.bcm.tmc.edu/AdikesavanEtAl/Sup. The interfaces of RecA with ATP, DNA and other monomers were defined as the amino acids that are closer than 5 Å from the ligand in atom to atom distances, excluding hydrogens. The figures of RecA monomer and filament structures were generated by PyMOL (The PyMOL Molecular Graphics System, Version 1.3, Schrödinger, LLC) using the PDB structure: 3cmw. The RecA filament was extended by repeated duplications and space alignment of the terminal monomers. 34045 rigid-body protein-protein docking models of a LexA dimer bound to the RecA filament with good molecular shape complementarity were created with the program PatchDock [108].
The wild-type RecA plasmid (pGE591-recA-WT) was used as template for site-directed mutagenesis of RecA protein using QuikChange II XL site-directed mutagenesis kit (Stratagene) as per manufacturer's protocol. The plasmids containing mutations in the recA gene obtained by site-directed mutagenesis were transformed in E. coli SMR6765 (ΔrecA) strain. All the recA mutant plasmids were sequence verified. The mutant RecA proteins expressed from these strains were also checked for their stability by western analysis using anti-RecA antibody.
The semi-quantitative measurement of UV sensitivities of wild-type RecA and RecA mutants was done as described previously [79]. E. coli SMR6765 strains expressing either wild-type RecA or RecA mutants were grown overnight in Luria-Bertani (LB) medium containing selective antibiotic (kanamycin 25 µg/mL). The next day, subcultures were made and grown further till the OD600 reached 0.5. The bacterial cultures were streaked onto sterile LB/Kanamycin plates using sterile Q-tips. The plates were exposed to increasing doses (J/m2) of UV light using a UV Stratalinker, and incubated at 37°C for a further period of 16 hours protected from light. Different levels of UV survival between wild-type RecA and RecA mutant strains were analyzed. The assay was repeated at least three times independently and the representative results are shown.
Western blot analysis of in vivo LexA cleavage was carried out as described previously [79], [109] with minor modifications. E. coli SMR6765 strains carrying either wild-type RecA or RecA mutants were grown overnight and the next day, subcultures made and grown at 37°C till the OD600 reached 0.5. The DNA damaging agent, nalidixic acid (Sigma) was added to each culture at 100 µg/mL final concentration. The cultures were grown further at 37°C and 1 mL of culture from each strain was aliquoted at 0, 30 and 60 minutes. The culture aliquots were washed once in cold PBS and were stored at −80°C until further processing. Subsequently, the pellets were lysed using BugBuster Master Mix (Novagen) and the total lysate made as per the manufacturer's protocol. Total proteins in the lysates were estimated using the Micro BCA Protein Assay Kit (Thermo Scientific). The RecA protein levels were normalized to bacterial growth by using equal amount (50 µg) of total protein lysate collected at different time points for resolving in SDS-PAGE. The resolved bands were blotted to nitrocellulose membranes and probed with anti-LexA (1∶7000, ABR bioreagents) and anti-RecA (1∶15000, MBL International) antibodies. Goat anti-rabbit IgG-HRP (Chemicon International) was used as the secondary antibody at 1∶7000 dilutions. Chemiluminescence detection was done using Amersham ECL western blotting kit and autoradiographed as per manufacturer's protocol. All the experiments were repeated at least three times for each RecA mutants and the representative results are shown.
The recombination efficiency of the E. coli strains carrying wild-type RecA and RecA mutant proteins were assayed by P1 transduction as described [110]. The assay measures the efficiency of the wild-type RecA or its variants, to recombine the selectable genetic marker (gfp-cat gene) into their chromosome, using P1 phage mediated transduction. P1 lysate was prepared by growing the donor bacterial strain (CH458≡MG1655 lacZYA::gfp-cat) overnight in LB medium with chloramphenicol antibiotic. The overnight culture was diluted 1∶4 in fresh LB+ 5 mM CaCl2 and 0.2% glucose and allowed to stand for 30 min at room temperature. Then wild-type P1 phage lysate was added to the diluted overnight culture, incubated with shaking @ 37°C for 20 min followed by plating them on LB plates with 5 mM CaCl2 and 0.2% glucose. Next day after overnight incubation of the plates, the top layer of lysed cells were scrapped-off into sterile centrifuge tubes, and ∼300 µl of chloroform added to the lysate, vortexed and allowed to stand for 30 min at room temperature with intermittent vortexing followed by centrifugation @ 10000 rpm for 10 min to collect the supernatant P1 lysate. The P1 phage lysate was subsequently titred against E. coli strain SMR6765 containing wild-type RecA on pGE591 plasmid. The viable cell numbers for wild-type RecA and RecA mutant strains was also assayed, so that approximately 1 phage for every 100 viable cells was used in the P1 transduction assay. During the assay, the recipient bacterial strains (wild-type RecA and the RecA-mutant strains) were grown overnight and subcultured the next day till the OD600 reached 0.5. P1 lysate was added to the cultures in such a way that the ratio of phage to viable cell count was ∼1∶100, vortexed, and incubated with shaking @ 37°C for 18 min followed by centrifugation for 2 min at 7000 rpm to pellet the cells. The cells were resuspended in LB medium with 100 mM sodium citrate and plated on LB-citrate plates with chloramphenicol, incubated overnight at 37°C. Next day, the number of transductant colonies in each strain was counted. The transduction or recombination efficiency of the wild-type RecA and mutant RecA strains were calculated by the number of transductants relative to the phage titer. The assay was repeated at least 3 times for all the wild-type RecA or RecA mutant strains and the mean standard error values for recombination efficiency were used for graphical representation.
The cleavage of UmuD protein to UmuD' upon DNA damage were shown individually in E. coli strains with plasmid-borne wild-type RecA or empty vector or RecA mutants (G108Y, G22Y, K23Y and G24Y) by western blot [111]. The E. coli strains (OL53) used in this assay were lexA (def) to enable constitutive UmuD expression. UmuD cleavage to UmuD' was assayed similar to LexA cleavage analysis except that after DNA damage induction, the aliquots were collected at 0, 1, 2 and 4 hours (since UmuD induction is a late process in the SOS response). The culture aliquots were processed similarly as mentioned above for LexA cleavage analysis. 200 µg of total protein from lysates were resolved in SDS-PAGE and immunoblotting was done with anti-UmuDD' antisera (1∶2000). The analyses were repeated at least 3 times independently for each wild-type RecA or mutant strains and the representative data were shown.
|
10.1371/journal.pntd.0003282 | Matrix Metalloproteinase 9 Production by Monocytes is Enhanced by TNF and Participates in the Pathology of Human Cutaneous Leishmaniasis | Cutaneous leishmaniasis (CL) due to L.braziliensis infection is characterized by a strong inflammatory response with high levels of TNF and ulcer development. Less attention has been given to the role of mononuclear phagocytes to this process. Monocytes constitute a heterogeneous population subdivided into classical, intermediate and non-classical, and are known to migrate to inflammatory sites and secrete inflammatory mediators. TNF participates in the induction of matrix metalloproteinases (MMPs). MMP-9 is an enzyme that degrades basal membrane and its activity is controlled by the tissue inhibitor of metalloproteinase.
Mononuclear cells were obtained from ex-vivo labeling sub-populations of monocytes and MMP-9, and the frequency was determined by flow cytometry. Culture was performed during 72 hours, stimulating the cells with SLA, levels of MMP-9 and TIMP-1 in the supernatants were determined by ELISA.
We observed that cells from CL lesions secrete high amounts of MMP-9 when compared to healthy subjects. Although MMP-9 was produced by monocytes, non-classical ones were the main source of this enzyme. We also observed that TNF produced in high level during CL contributes to MMP-9 production.
These observations emphasize the role of monocytes, TNF and MMP-9 in the pathogenesis of L. braziliensis infection.
| To examine the participation of MMP-9 in the pathogenesis of L. braziliensis infection, we realized a cross-sectional study with CL patients in an early phase of the disease or with a classical ulcer, and healthy controls. We evaluated the frequency of MMP-9 in monocyte subsets and its mechanism of production. Our results showed that monocytes were the major cells producing MMP-9. The MMP-9 production by CL patients was presented in higher levels when compared with healthy subjects and early cutaneous leishmaniasis (ECL) patients, and the levels of MMP-9 inhibitor, TIMP-1, were lower in CL patients when compared to healthy subjects. The production of MMP-9 was enhanced by TNF, a cytokine associated with tissue damage in CL patients. Thus, therapeutic modulation of MMP-9 may be a useful approach for improving disease outcome in L. braziliensis patients.
| Human cutaneous leishmaniasis (CL) caused by Leishmania braziliensis infection is characterized by the presence of one or more ulcerated lesions with raised borders and few parasites [1]. Early after infection, most patients develop lymphadenopathy, followed by the appearance of a papule at the bite site, which subsequently becomes an ulcerated lesion. These lesions are composed of a robust inflammatory infiltrate including the presence of T and B lymphocytes, mononuclear phagocytes and plasma cells [2]. It is well known that both CD4+ and CD8+ T cells have important roles in the control of Leishmania parasites replication [3]. During L. braziliensis infection these cells are activated and TNF and IFN-γ are produced in high levels both by peripheral blood mononuclear cells (PBMC) and at lesion site of CL patients. However, this response can also lead to tissue damage and development of the ulcer [4].
In contrast to the role of T cells in the pathogenesis of L. braziliensis infection less attention has been given to the contribution of mononuclear phagocytes to the inflammatory process and tissue damage observed in CL. In spite of the presence of tissue resident mononuclear phagocytes, circulating monocytes migrate to the infection site. In the tissue, they can differentiate into macrophages and dendritic cells (DCs), which are the main cell types parasitized, by Leishmania. It was recently proposed that in humans circulating monocytes are a heterogeneous population based on the surface expression of CD14 and CD16 [5]. The monocyte subsets are subdivided into classical (CD14++CD16−), intermediate (CD14++CD16+) and non-classical (CD14+CD16++) [5], [6]. It has been shown that CD16+ monocytes are able to produce high levels of TNF and increased frequency of these monocytes is associated with the immunopathogenesis of inflammatory diseases, such as arthritis and sepsis [7], [8], [9].
TNF is an inflammatory cytokine produced in high levels in CL patients [4]. TNF participates in the inflammatory process through the induction of nitric oxide, necrosis, citotoxicity and expression of matrix metalloproteinases (MMPs) [10], [11], [12]. MMPs are zinc-dependent enzymes that degrade extracellular matrix proteins and are functionally classified according to the specificities of the substrates they degrade [13]. MMP-9 is involved in degradation of collagen type IV, the major component of basal membrane present at skin [14]. While MMPs are necessary for successful eradication of infection by stimulating the migration of effector cells to the inflammatory site, high production of these molecules may induce pathology [15]. An imbalanced production of MMPs and its natural regulator, tissue inhibitor of metalloproteinases (TIMPs) occurs in a variety of diseases where tissue damage occurs [16], [17], [18]. L. braziliensis-infected macrophages secrete MMP-9, and its production is increased in patients with mucosal leishmaniasis, the more severe form of L. braziliensis infection [19]. Thus, while the factors that induce the breakdown of the basal membrane leading to the development of skin ulcer in leishmaniasis are not yet elucidated, we hypothesize that the lesion formation is due to breakdown dysregulation of the basal membrane caused by imbalance in levels of MMP-9 and TIMP-1, together with other factors, such as cell recruitment and edema, [20].
In the present study we examined the participation of MMP-9 in the pathogenesis of L. braziliensis infection. CL patients in an early phase of the disease or with a classical ulcer were evaluated. Initially, we showed that MMP-9 genes were highly expressed in the CL skin lesions. We then evaluated the production of MMP-9 and its inhibitor, TIMP-1, and identified an imbalance between MMP-9 and TIMP-1 ratio in CL patients. Monocytes were the main source of the enzyme. Finally, we found that high levels of TNF produced during the disease contribute to the up-regulation of MMP-9 synthesis in CL.
Participants of the present study includes 19 early cutaneous leishmaniasis (ECL) and 85 CL patients from the L. braziliensis transmission area of Corte de Pedra, Bahia state, Brazil, and 29 healthy subjects (HS) living in areas not exposure to Leishmania. ECL patients were characterized by the presence of a lymphadenopathy or lymphadenopathy accompanied by a papule or an exoulcerative lesion. Diagnosis for CL was performed by a positive parasite culture or PCR as previously described [21]. All patients were evaluated before treatment.
For whole genome expression microarray, lesion biopsies preserved in RNAlater (Qiagen) were homogenized using a rotor-stator and RNA was isolated using the RNeasy Plus kit (Qiagen). Biotin-labeled complementary RNA (cRNA) was generated using the Illumina TotalPrep RNA amplification kit (Ambion). RNA and cRNA quality were assessed on a BioAnalyzer (Agilent). Illumina HumanHT-12 version 4 expression beadchips were hybridized with cRNA from 26 L. braziliensis lesion biopsies and 10 biopses collected from uninfected donors. Arrays were scanned using a BeadStation 500GX and raw image files were processed using GenomeStudio v1.8 software (Illumina). Data was variance stabilized, robust-spline normalized, and quality control analysis carried out using the Lumi package [22] in Bioconductor/R. Differential expression analysis of the data using linear models and empirical Bayes methods [23] was carried out using the Limma package [24]. Data was deposited on the Gene Expression Omnibus (GEO) database for public access (GSE#GSE43880). Heat map tools available on GenePattern [25] were used to graphically display differentially regulated genes in Figure 1A.
SLA was prepared with an isolate of L. braziliensis as previously described [26]. Briefly, promastigotes ressuspended in lysis solution (Tris, HCL, EDTA and leupeptin) were immersed in liquid nitrogen, and thawed at 37°C. After freezer-thaw procedure, they were sonicated and the disrupted parasites were centrifuged at 14,000G. The supernatant was filtered and assayed for protein concentration. SLA was used at a concentration of (5 µg/ml).
Flow cytometry was performed as previously described [27]. Briefly, PBMC were ressuspended in saline and adjusted to a concentration of 0.5×106 cells/ml. For ex-vivo, cell surface staining, we incubated cells with monoclonal antibodies anti-CD4 (APC), anti-CD8 (PE), anti-CD14 (APC) and anti-CD16 (PE) (BD bioscience), for 15 minutes, washed by centrifugation twice and fixed with 2% paraformaldehyde. For intracellular staining, cells were ressuspended in Perm/Wash (BD Cytofix/Cytoperm Plus – BD bioscience) for 15 minutes and intracellular labeling was performed using monoclonal antibody anti-MMP-9 (FITC) (BD bioscience) for 30 minutes. Cells were acquired on a FACS Canto II and analysis was done using Flowjo software (Tristar).
Peripheral blood mononuclear cells (PBMC) were isolated from heparinized venous blood by Ficoll-Hypaque (GE Healthcare Bio-Sciences AB, Sweden) gradient centrifugation. After washing three times in 0.9% NaCl, PBMC were adjusted to 3×106 cells/ml in 1 ml of RPMI-1640 (Gibco Laboratories, Grand Island, NY, USA) supplemented with 10% fetal bovine serum (FBS) (Gibco Laboratories, South America Invitrogen), 10 IU/ml penicillin and 100 µg/ml streptomycin. Cells were placed on 24 wells plates and incubated for 24 or 72 hours in the presence or absence of SLA (5 µg/ml) or recombinant TNF (5 ng/ml), or anti-TNF antibody (10 µg/ml) (R&D systems, Minneapolis, MN), as indicated in figures.
Biopsies from L. braziliensis patients and HS were cultured in complete RPMI media without stimuli. Tissue from CL patients and HS were cultured in RPMI for 12 hours. Supernatants from PBMC and biopsies were collected and stored at −70°C. The Levels of MMP-9, TIMP-1 (BD Biosciences, San Diego, CA, USA) and TNF (R&D Systems, Minneapolis, MN) were measured by ELISA according to the manufactures instructions. The results are expressed in pg/ml.
Mann-Whitney was used to compare HS and CL groups; Kruskal-Wallis test (nonparametric test) was used to compare the ECL, CL and HS groups; Wilcoxon matched pair test (paired and nonparametric t test) was used to analyze PBMC cultures in different conditions within the same group of individuals. Comparisons were considered statistically significant when p<0.05. All p values represented are two-sided.
This work was approved by the Ethics and Research Committee from Federal University of Bahia. All subjects provided witted informed consent; in case of illiterate subjects, a thumb print plus signature of an independent witness were used.
MMPs are enzymes that degrade extracellular matrix and in high levels MMPs may cause tissue damage [15]. Exaggerated inflammatory responses leads to tissue damage and ulcer development in CL [4]. To determine whether MMPs were expressed in CL lesions, we performed a whole genome expression profile from lesions of CL patients (n = 26) and compared that to normal skin (n = 10). The results showed that several MMPs, including MMP-9, had increased expression over normal skin (Figure 1A). We also assessed the expression of TIMP-1 in the skin and although TIMP-1 expression was increased in L. braziliensis infected skin over healthy skin, the ratio MMP-9/TIMP-1 expression was higher in CL skin when compared to normal tissue (Figure 1A and B). Because MMP-9 is known to be involved in basal membrane disruption, a process that precedes ulcer development in CL, we tested whether MMP-9 protein was produced in the lesion of CL patients and healthy subjects. To address that we cultured whole biopsies in the absence of stimulus for 12 hours and determined MMP-9 levels on culture supernatants by ELISA. Cells from biopsies from CL patients produced significantly more MMP-9 than those from healthy subjects (HS) (Figure 1C). These results show an imbalance between MMP-9 and TIMP-1 expression, suggesting that MMP-9 participate in lesion development in CL.
TIMPs are glycoproteins that inhibit MMPs and TIMP-1 specifically regulates the activity of MMP-9. Therefore, our next step was to determine the production MMP-9 and TIMP-1 by PBMC from HS and CL patients in the early and late phase of the disease. To do so, PBMC were stimulated in vitro with SLA or left untreated (media) for 72 hours and the production of MMP-9 and TIMP-1 were determined in the supernatants by ELISA. Our results show that CL patients produced high levels of MMP-9 and low levels of TIMP-1 when compared to HS (Figure 2A and 2B). Interestingly, PBMCs from CL patients produced high levels of MMP-9 even in the absence of stimulus, indicating that those cells are prone to produced MMP-9 when still in CL patient. A change in the ratio of MMPs and TIMPs may cause excessive degradation of the extracellular matrix and consequently tissue damage [28]. Therefore, we next sought to determine the ratio between MMP-9 production and TIMP-1. The high ratio MMP-9/TIMP-1 in cultures from CL individuals indicated an imbalance in the production of these enzymes in L. braziliensis infected individuals when compared to HS (Figure 2C).
It has been documented that leucocytes are the main source of MMP-9 [29], [30] and our results show that PBMCs from CL patients secrete high levels of MMP-9 in response to SLA. To further characterize the source of MMP-9 in CL patients, using flow cytometry we determined the ex-vivo production of MMP-9 by CD4+ and CD8+ T cells, and monocytes based on CD14 expression. Our results show that monocytes are the main source of this enzyme in HS and patients with CL (Figure 3A and B).
Recently, three monocyte subsets have been described based on the expression of CD14 and CD16, subdividing them into classical, intermediate and non-classical monocytes [5]. In our study the gate strategy to access the monocyte subsets was defined based on CD14 and CD16 expression. To evaluate the contribution of monocyte subsets to MMP-9 production we performed intracellular ex-vivo staining of PBMC from patients with ECL, CL and HS which were then analyzed by flow cytometry. The frequency of MMP-9 in each subsets of monocyte was performed using an isotype control (Figure 4A and B). Our data shows that monocytes from patients with ECL and CL express more MMP-9 than monocytes from HS and that non-classical monocytes (CD14+CD16++) are the major source of MMP-9 in patients with CL (Figure 4C and 4D).
TNF is an inflammatory cytokine that contributes to tissue damage by different mechanisms, including by induction of MMP-9 expression [31]. To determine if TNF production contributes to MMP-9 production in CL patients, we first measured TNF levels in supernatants of SLA-stimulated PBMC from HS and CL patients. As previously demonstrated [32], CL patients produced high levels of TNF in response to SLA (Figure 5A). In order to determine if the TNF produced had the ability to increase MMP-9 expression, we added recombinant TNF to PBMC cultures from HS and determined MMP-9 expression in the supernatants by ELISA. Our results show that there was a 5-fold increase in MMP-9 production in unstimulated PBMC cultures from healthy individuals upon exogenous addition of recombinant TNF (Figure 5B). Finally, we asked if TNF was playing a role in MMP-9 production in CL. In order to address this question, we added anti-TNF antibodies to PBMC cultures from HS and CL patients (Figure 5C) and we found that the blockage of TNF inhibits the production of MMP-9 in CL patients. Altogether these data show that TNF, in part, regulates MMP-9 production in CL patients.
CL is characterized by a well-defined ulcer with raised borders that appears a few weeks after the transmission of the parasite by sandflies [33]. One of the first signs of the disease is lymphadenopathy followed by the development of a papular or an exoulcerative lesion at the bite site, 2–3 weeks later. Finally, 1–2 weeks later the classical ulcer is observed [33], [34]. The evaluation of the immune response during the phase in which patients has not yet developed the cutaneous ulcer is important to determine factors contributing to disease development. Our group has described the immune response in CL patients and discovered the presence of many cytokines and chemokines that contributes to the maintenance of inflammatory response in these individuals [35]. However, much less has been done to understand the immune response at the early stages of the disease. This knowledge is particularly important, as it may unravel the factors that contribute to disease severity and/or ulcer development. Here, we investigated the contribution of leucocytes to MMP-9 production in CL patients and for the first time, we show how monocytes contribute to the production of MMP-9 in CL.
MMPs mediate several physiological processes, such as cell migration, extracellular matrix degradation and tissue remodeling. Our in situ gene expression data shows that MMP-1, MMP-3 and MMP-9 expression were particularly increased in CL over healthy skin. While MMP-1 and MMP-3 are more associated with matrix remodeling,uncontrolled secretion of MMP-9 has been associated with pathological processes [28], [36], [37]. There are a few examples of a role for MMPs in the pathogenesis of Leishmania infection: in a murine model of L. chagasi infection, production of MMP-9 by macrophages was associated with tissue damage [38]; high levels of mRNA to MMP-2 were documented in ulcers of CL patients and in macrophages from mucosal leishmaniasis patients; and upon infection with L. braziliensis human macrophages increased secretion of MMP-9 [19]. As MMP-9 degrades type IV collagen, the main component of the basement membrane in the skin, and their activation causes excessive tissue damage by facilitating the migration of inflammatory cells to the infection site [28]. In the present work we found that the MMP-9 gene was expressed during disease when compared to normal skin and we were also able to detect high levels of this enzyme being secreted by cells obtained from L. braziliensis lesions. With these findings we hypothesize that MMP-9 plays a role in ulcer development during CL. Our results led us to determine the levels of MMP-9 produced by PBMC. High levels of MMP-9 were observed in supernatants of PBMC from patients when compared with those from healthy subjects. Because exaggerated production of MMPs are associated with immunopathology [15], TIMPs, the natural inhibitors of these enzymes, play an important role in controlling MMPs production and activity. As expected, the levels of TIMP-1 were increased in healthy subjects when compared to CL patients, revealing an imbalance between MMP-9/TIMP-1 in CL individuals. MMP-9 activity was previously associated with development of mucosal leishmaniasis, the more inflammatory form of the disease with extensive tissue damage, and expression of MMPs were also correlated with therapeutic failure in CL [19], [39]. These data confirm the potential role for these enzymes in mediating immunopathology during leishmaniasis. Our documentation that patients in the pre-ulcerative phase of the disease produced lower levels of MMP-9 than patients with classical ulcer, raise the possibility that progression of the disease to the ulcerative phase may be associated with increasing in MMP-9 levels and, consequently, imbalance between MMP-9 and TIMP-1.
Evidence has been accumulated that activation of both CD4+ and CD8+ T cells are associated with tissue damage in CL [40], [41], [42]. On the other hand, less attention has been given to the role of mononuclear phagocytes in the pathogenesis of CL. Macrophages are the main reservoirs of the Leishmania, responsible for Leishmania parasite killing [43] but macrophages also contribute to the production of TNF-α and pro-inflammatory chemokines, which are associated with severe forms of the disease [44]. It was previously reported that different leukocytes contribute to deleterious MMPs secretion in infectious diseases [29], [30], [45], [46]. In a mouse model of toxoplasmosis, in addition to inflammatory monocytes and neutrophils, CD4+ and CD8+ T lymphocytes where also important source of MMP-8 and MMP-10 [47]. In malaria and tuberculosis MMP-9 was predominantly produced by activated monocytes [45], [46]. Here we show that in CL, although CD4+ and CD8+ T lymphocytes participate in the pathological immune response through the production of inflammatory mediators or cytolytic activity, monocytes were the main cells producing MMP-9. Neutrophils are another cell type known to produce high levels of MMP-9 in other inflammatory conditions [48], but since in CL the cellular infiltrate is predominantly mononuclear, neutrophils may not be a primary source of MMP9 in these patients [49].
Monocytes are subdivided in three subsets according to the expression of CD14 and CD16, and each subset has distinct roles during the infectious process. Classical monocytes (CD14++CD16−) have high phagocytic capacity [5], [6] and express more anti-microbicidal molecules, like reactive oxygen species, acting as an important line of defense against L. braziliensis (Novais and Nguyen et al., unpublished data). In contrast, intermediate monocytes (CD14++CD16+) express more MHC class II and act more effectively as antigen presenting cells than the other subsets of monocytes [5], [6]. Controversial studies have been published regarding the function of non-classical monocytes (CD14+CD16++) as some have reported that these cells produce more pro-inflammatory cytokines and are responsible for the recruitment of cells to the inflammatory sites, and others have documented a regulatory role for them [5], [6]. Although in this study all monocyte subsets produced MMP-9, the intensity of expression of this molecule was higher in non-classical monocytes from some individuals. It was previously shown that the expression of CD16 in monocytes is associated with high production of MMP-9 and increased frequency of CD16+ monocytes has been documented in CL patients [50] [51]. Thus, future studies will be performed to determine the ability of these cells to migrate into tissues to determine the contribution of different monocyte subsets to immunopathology at lesion site.
The production of MMP-9 is not spontaneous, being dependent on the interaction of cell to cell, cell to matrix and/or in response to cytokines [52]. As an important inflammatory cytokine TNF is involved in immune regulation and resistance to various microorganisms and exerts a variety of biological activities such as apoptosis, cytotoxicity and induction of MMPs [10], [11], [12]. We showed here that in CL TNF plays an important role in the regulation of MMP-9 since addition of exogenous TNF enhanced MMP-9 production in HS, whereas neutralization of this cytokine down-regulated MMP-9 synthesis in CL patients. Therefore we propose TNF as a major regulator of MMP-9 production during L. braziliensis infection. Our observations agree with what has been previously reported for other diseases [45], [46]. For example, in malaria, activation of MMP-9 was dependent upon the production of TNF and its activation was decrease by anti-TNF treatment [46].
This study brings evidence for the role of monocytes in the pathology associated with L. braziliensis infection and suggests that monocytes and MMP-9 may play key roles in tissue destruction. Specifically, the excessive production of TNF by monocytes observed in CL increases the production of MMP-9 leading to an imbalance between production of this enzyme and its inhibitor, TIMP-1. As a consequence, there is an excessive degradation of the basal membrane, migration of inflammatory cells to the site of infection and ulcer development. Thus, therapeutic modulation of MMP-9 may be a useful approach for improving disease outcome in L. braziliensis patients.
|
10.1371/journal.pbio.1001361 | Transport of Fibroblast Growth Factor 2 in the Pericellular Matrix Is Controlled by the Spatial Distribution of Its Binding Sites in Heparan Sulfate | The heparan sulfate (HS) chains of proteoglycans are a key regulatory component of the extracellular matrices of animal cells, including the pericellular matrix around the plasma membrane. In these matrices they regulate transport, gradient formation, and effector functions of over 400 proteins central to cell communication. HS from different matrices differs in its selectivity for its protein partners. However, there has been no direct test of how HS in the matrix regulates the transport of its partner proteins. We address this issue by single molecule imaging and tracking in fibroblast pericellular matrix of fibroblast growth factor 2 (FGF2), stoichiometrically labelled with small gold nanoparticles. Transmission electron microscopy and photothermal heterodyne imaging (PHI) show that the spatial distribution of the HS-binding sites for FGF2 in the pericellular matrix is heterogeneous over length scales ranging from 22 nm to several µm. Tracking of individual FGF2 by PHI in the pericellular matrix of living cells demonstrates that they undergo five distinct types of motion. They spend much of their time in confined motion (∼110 nm diameter), but they are not trapped and can escape by simple diffusion, which may be slow, fast, or directed. These substantial translocations (µm) cover distances far greater than the length of a single HS chain. Similar molecular motion persists in fixed cells, where the movement of membrane PGs is impeded. We conclude that FGF2 moves within the pericellular matrix by translocating from one HS-binding site to another. The binding sites on HS chains form non-random, heterogeneous networks. These promote FGF2 confinement or substantial translocation depending on their spatial organisation. We propose that this spatial organisation, coupled to the relative selectivity and the availability of HS-binding sites, determines the transport of FGF2 in matrices. Similar mechanisms are likely to underpin the movement of many other HS-binding effectors.
| The development, homeostasis, and repair of animal tissues requires communication between cells mediated by effector proteins, which are released from source cells and must move through the surrounding extracellular matrix to reach their receptors on target cells. A major component of the extracellular matrix is the polysaccharide heparan sulfate (HS); it binds the majority of these effectors and has the crucial function of regulating their transport. The mechanism underlying this function, however, is still unknown. To understand how HS regulates the transport of effectors, in this study we labelled molecules of the effector protein fibroblast growth factor 2 (FGF2) each with a gold nanoparticle, which we could visualise and quantify by electron microscopy and by a new approach called photothermal heterodyne imaging. By imaging the gold nanoparticles, we found that the binding sites for FGF2 on HS are distributed heterogeneously in the extracellular matrix that surrounds cells in culture. Single molecule tracking indicated that these binding sites are organised into local networks that confine the FGF2 and into paths that allow its translocation over long distances (up to several micrometers). Thus, the spatial distribution of the binding sites in HS and their physicochemical properties of binding are major factors controlling the transport of effectors in extracellular matrices.
| The notion of gradients of morphogens and of epithelial-mesenchymal signal relays is common currency in developmental biology [1]–[4]. Moreover, organism homeostasis often depends on similar transport of effector proteins, such as growth factors, cytokines, and chemokines from source to target cell, for example, in wound repair and in the regulation of immune responses [5]. Such transport occurs in the extracellular matrix that lies between cells, including the pericellular matrix adjacent to the plasma membrane, where the heparan sulfate (HS) chains of proteoglycans (PGs) are the dominant molecular species [6]. This dominance is due to their size (∼40 nm to 160 nm long), amount, and unlike the other extracellular glycans, their large array of protein partners (over 400), which they bind with varying degrees of selectivity [7],[8]. These protein partners include most protein effectors that mediate cell communication (e.g., morphogens, chemokines, cytokines, growth factors, matrix proteins, and their cognate cellular receptors).
HSPG possess a core protein (transmembrane, glycophosphatidyl inositol anchored or soluble), to which one or more HS chains are attached. A particular feature is the long, unbranched glycosaminoglycan chain, in which tracts of variably sulfated saccharides, responsible for the interaction with proteins, alternate with non-sulfated sequences of sugars [9]. A single chain of HS contains multiple, even overlapping, protein binding sites [10]. In addition, one particular sequence of sulfated sugars in a chain can bind different ligands with different affinities (e.g., [11]), and vice versa, a single ligand can bind to several sequences of sugars. The binding of ligands to HS chains is governed by relative selectivity rather than absolute specificity and there is substantial overlap at the level of the sugar sequences recognised by different proteins [12]. This conclusion is reinforced by the demonstration that some unrelated sulfated plant polysaccharides possess structures that allow effective interaction with HS-binding proteins [13].
Impairing the interaction of HS with its protein partners has been shown to alter gradient formation, as well as short- to long-range signalling for many morphogens and regulatory proteins [e.g., hedgehog, wingless (WNT), decapentaplegic (DPP, ortholog of vertebrate bone morphogenic protein), and fibroblast growth factors (FGF)] [1]–[3],[14]–[19]. The many experiments of this type demonstrate the crucial role of HS in the regulation of the transport of effectors. Despite the considerable overlap in the structures of the binding sites in HS recognised by its many protein partners, it is well established that HS can also be selective for these partners, which has been evidenced in matrices from different tissues (e.g., [20]–[23]). In addition, matrices are dynamic, so the selectivity of their HS for protein partners changes over time, which is particularly evident in development [24]. Thus, the expression of sequences of sulfated sugars can be spatially and temporally regulated in tissues, which tunes the interaction of protein partners with HS and regulates their effector and transport functions. However, how HSPG regulate the transport of its protein partners in matrices remains debated, because this has not been measured directly (reviewed in [19]).
To address this issue, we have used a new generation of gold nanoparticle probes (10 nm diameter) [25] to stoichiometrically label FGF2 morphogen, the archetypal HS-binding growth factor, and examine its distribution and dynamic fluctuations in the pericellular matrix of Rama 27 fibroblasts. To identify FGF2 associated with FGF receptor (FGFR), a heparin-derived dodecasaccharide, degree of polymerisation (DP) 12, was used to prevent interaction with endogenous HS. Ternary complexes of FGF2-NP:DP12:FGFR were found to be less mobile than FGF2 associated with HS. In the absence of exogenous DP12, we show that virtually all FGF2 bound to the pericellular matrix is engaged with HS, rather than the FGF receptor (FGFR). These HS-binding sites form non-random networks of heterogeneously distributed binding sites. The FGF2 moves from one HS-binding site to another in these networks, which determine whether it undergoes confined motion (∼110 nm) or substantial translocation (µm) in the pericellular matrix. The spatial organisation, the relative selectivity, and the availability of HS-binding sites thus lie at the heart of the mechanisms regulating the transport of FGF2 in matrices.
To examine, at single molecule resolution, the distribution and dynamic fluctuations of the FGF2 morphogen in the pericellular matrix of Rama 27 fibroblasts, we have used a new generation of 10 nm diameter gold nanoparticle probes [25]. The nanoparticles bear only one TrisNiNTA tag [26],[27], so they can specifically and stoichiometrically label the FGF2 (poly-histidine tagged FGF2, His-FGF2, see Materials and Methods). It has been demonstrated that, in the extrasynaptic membrane, protein diffusion parameters are similar when using probes as different as 500 nm diameter latex beads, 30 diameter nm quantum dots, and small organic dyes of ∼1 nm [28],[29]. Thus, within the pericellular matrix of Rama 27 cells, the 10 nm nanoparticles used here are not expected to interfere with the diffusion of the FGF2. Moreover, the N-terminus of FGF2 is an appropriate location for conjugation of a probe, because it is opposite the binding site for FGFR and the canonical heparin binding site and there are natural N-terminal extensions of FGF2 that do not affect its ability to bind heparin and activate FGFRs [30]–[32]. The Rama 27 cell line is representative of the mammary stroma from which it was derived; for example, it differentiates towards an adipocyte phenotype [33]. Its cytoplasm peripheral to the nucleus is very thin (∼2 µm) and flat, which allows it to be used for two-dimensional tracking of molecules in its pericellular matrix (thickness ∼1 HS chain). Moreover, purified HS from Rama 27 cells has been extensively characterised at the level of its FGF2 binding properties and the ability of this HS to act as a co-receptor and enable the growth-stimulatory activity of FGF2 [11].
Following purification, the functionality of FGF2-nanoparticle conjugates (one FGF2 for one nanoparticle, FGF2-NP) was assessed. At equimolar concentration, FGF2-NP was as potent as unlabelled His-FGF2 protein in stimulating DNA synthesis (Figure 1A). Similarly, FGF2-NP stimulated the sustained phosphorylation of fibroblast growth factor receptor substrate-2 (FRS2) and of mitogen-activated protein kinases (MAPK) p42/44MAPK, which are established signalling events downstream of the FGFR, to the same extent as unlabeled His-FGF2 (Figure 1B). A heparin-derived dodecasaccharide, DP 12, will prevent the binding of FGF2 to cellular HS by direct competition and replace endogenous HS in the formation of stable signalling complexes between the FGF2 and the FGFR [34]. A similar phosphorylation of FRS2 and of p42/44MAPK was observed in the presence or absence of the dodecasaccharide (Figure 1B). These results demonstrate that the FGF2-NP conjugate has the same growth-stimulatory and signalling activity as the free protein. As these effects are dose dependent [35], FGF2-NP conjugates and unlabelled FGF2 will be interacting with the HS co-receptor and FGFR similarly.
Since the FGF2-NP possessed the same activity as unlabelled FGF2, we were able to take advantage of the imaging versatility of the gold nanoparticle probe. Its electron density enables ready detection by TEM, while its strong plasmon absorbance allows optical imaging and tracking of individual NPs by PHI. In a first set of experiments, we examined whether the spatial distribution of binding sites for FGF2 in the HS of the pericellular matrix of fibroblasts was homogenous or heterogeneous. Previous coarser grained immunofluorescence and immunohistochemical data have shown that, although protein-binding structures in HS may be expressed differently between different matrices, within a particular matrix these have an apparently amorphous spatial distribution [20],[36],[37]. However, they have not had sufficient resolution to determine the distribution of such binding structures within a matrix.
In PHI, images are acquired by serial scans along the x-axis. The presence of lines (Figure 4A, rectangles) rather than spots (Figure 4A, circles) indicated that some of the FGF2-NP were moving along the direction of the scan in the pericellular matrix of living cells. It is important to note that FGF2 bound to HS in pericellular and extracellular matrices remains associated with these. It does not readily exchange into the bulk culture medium in the absence of competing exogenous soluble HS or heparin [32],[51],[52], though it may exchange into the medium within and nearby the matrix and then re-bind. Thus, these results demonstrate that FGF2-NP bound to HS of the pericellular matrix is mobile within the matrix.
Experiments performed in living cells were repeated in fixed cells, which will prevent the diffusion of the protein core of the HSPG, though the protein binding sites in the HS chains will be largely unaffected. This is because the overwhelming majority of glucosamine residues in protein binding domains are N-sulfated. Intriguingly, when FGF2-NP was added to fixed cells, the growth factor was still mobile (Figure 4B,C). It has been shown that some isolated membrane and glycosyl-phosphatidylinositol (GPI) anchored proteins might retain some mobility following fixation [53]. However, the cross-linking of the numerous endogenous protein partners of the HS chains and of the protein core of the HSPGs will severely restrict the freedom of the chains and protein cores, including GPI-anchored ones and, hence, their contribution to the observed motion. The mobility persisting in fixed cells cannot depend on cellular biochemistry. Comparison of sequential images taken in the same cell area at 70 min intervals shows that some of the immobile FGF2-NP have disappeared and that new FGF-NP have appeared. This suggests that there is a dynamic transition between immobile and mobile FGF2-NP (Figure 4B,C, dash circle).
PHI imaging indicated that some of the FGF2-NP was mobile in the pericellular matrix of both living and fixed cells. Such movement represents the transport of the FGF2 in the pericellular matrix. Therefore, we quantified the dynamic parameters of the movement of FGF2-NP by PHI single molecule tracking (see Materials and Methods). PHI tracking of gold nanoparticles uniquely allows very long trajectories to be captured, with a time frame of 42 ms and a pointing accuracy in the x, y dimensions of ∼10 nm (Materials and Methods). The motion of the FGF2-NP in the pericellular matrix is thus approximated to two dimensions. This is reasonable, given that the scale of the motion of FGF2-NP cannot exceed the depth of the pericellular matrix (no more than a single HS chain) by more than an order of magnitude and that the Rama 27 fibroblastic cells are flat.
FGF2-NP added to the cells will be virtually all associated with HS. In some experiments DP12 was included to compete for FGF2-NP binding to HS and so identify FGF2-NP associated with the FGFR, as a complex with DP12. Images were taken before and after the acquisition of tracks, which allowed the superimposition of tracks on a photothermal image (Figure 5A,C). It is apparent from inspection of exemplar trajectories and videos (Figure 5, Videos S1, S2, and S3) that an individual FGF2 molecule associated with HS undergoes various types of motion, ranging from confinements in a small area (e.g., expanded box, Figure 5E) to different types of travel phases, where the FGF2-NP undergoes substantial net displacement. The travel phases include motion that is nonetheless quite convoluted and interspersed with what appears to be confined motion (e.g., Figure 5D, grey track in dotted red circle, 434 s long), as well as straight-forward displacement that is more or less directional (e.g., Figure 5B, 18 s magenta track). When different tracks are superimposed (Figure 5B,D–E), this indicates that different FGF2-NP, which were tracked at different times in the same field, could travel the same path. Moreover, since there is a succession of different types of motion in individual tracks, it is clear that FGF2-NP were not restricted to any particular type of motion and were able to make transitions between these.
Discrimination between different diffusive behaviours was achieved by means of a plot of the distance travelled against displacement (Figure 6A) with a frame window of 12 points (0.5 s) (Materials and Methods “PHI Single Molecules Tracking Analysis” and Figure S1). Using this approach the data fell into five groups. As an illustration of this analysis, the exemplar tracks shown in Figure 6B and 6C are colour-coded according to the corresponding five diffusive behaviours in Figure 6A. All the physical parameters (diffusion coefficient, confinement diameter, mean square displacement over time, etc.) were calculated, as appropriate, for each group.
Group 1 corresponded to immobile/highly confined FGF2-NP. This was indistinguishable from the background noise of the tracker in a plot of distance travelled versus displacement (Figure 6). Group 2 corresponded to confined diffusion, where the FGF2-NP was clearly mobile in a plot of distance travelled versus displacement (Figure 6), yet confined to a small area. Group 3 was simple diffusive motion. Group 4 corresponded to slow directed diffusion, and Group 5 corresponded to long and fast directed diffusion. Only the last was restricted to living cells and for this reason was considered separately from Group 4, while Groups 2, 3, and 4 were statistically significantly different (Table S4).
Note that the mean square displacement (MSD) against time curves obtained for these five diffusive behaviours fit the physical description of protein diffusion, which has been characterised by others (Figure 7) [54]. This further supports our discrimination of the movement of FGF2-NP into these groups.
In living cells, individual FGF2 molecules spent most of their time (∼83%) in confined motion (Groups 1 and 2, Table S2A,B), which alternated with simple diffusive motion (Group 3, Table S2A,B, ∼13% of time). Occasionally (3% of time), the FGF2-NP underwent slow directed diffusion (Group 4, Table S2A,B) or more rarely fast directed diffusion (Group 5, Table S2A,B). It is important to note that the proportion of fast and directed diffusion may be underestimated, because the FGF2-NP undergoing such motion is near the speed limit of the tracker (∼0.2 µm2/s) (Figure 7A, Figure S2, Table S2A,B).
In the pericellular matrix of fixed cells, FGF2-NP spent more than 90% of their time in confined diffusion (Groups 1 and 2, Table S2C,D), with a commensurate decrease in simple diffusion (Group 3) and slow directed diffusion (Group 4) compared to living cells. Moreover, fast directed diffusion was absent (Group 5). Increasing the concentration of FGF2 from 22 pM to 220 pM had a clear effect on some of the parameters of the different types of motion, particularly on fixed cells (Table S2A–D, Figure 7B). However, it had no detectable effect on the proportion of time that an FGF2 spent in the five different types of motion (Table S2).
The signal intensity at each point in the trajectories was grouped into that corresponding to a single nanoparticle (below 0.147, see Materials and Methods) and that corresponding to two or more nanoparticles (Figure S3). In confined motion (Groups 1 and 2), FGF2 is more likely to be sufficiently close to one or more other FGF2 molecules to cause the photothermal signal to double or more than when the FGF2 undergoes diffusive motion (Figure S3). Thus, when FGF2 undergoes diffusive motion, it is less likely to be associated with other FGF2 molecules.
Competition by DP12 prevents FGF2-NP from binding to HS in the pericellular matrix, but allows the formation of a ternary signalling complex of FGF2-NP:DP12:FGFR. Thus, experiments with DP12 allow the motion of FGF2-NP associated with the signalling complex to be studied in isolation. In living cells, FGF2-NP associated with DP12 and FGFR spent 94% of their time undergoing confined motion (Groups 1 and 2, Figure S4) and just 5.5% of their time undergoing simple and slow directed diffusion. Despite the measurements being made on living cells, there was no Group 5 motion (long/fast directed diffusion). Thus, FGF2-NP associated with DP12 and FGFR were less mobile than FGF2-NP associated with HS (Figure S4 compared to Figure 7 and Table S2). This may reflect the progressive engagement of intracellular signalling platforms by the FGF2 ligand and DP12 co-receptor activated FGFR [55].
The change in MSD with time for the highly confined FGF2-NP motion (Group 1, Table S2) was not fitted by an exponential with an asymptote (Figure 7B), as the data were too close to the background noise of the tracker when using a time window of 12 points (0.5 s). However, we noted that the MSD increased with time compared to the control immobile nanoparticles fixed in polyvinyl alcohol, demonstrating that some of these FGF2 molecules, if not all, were indeed mobile. This mobility over time can be observed on exemplar trajectories (Figure 6B,C, black colour). Moreover, this mobility was somewhat higher in living compared to fixed cells. For confined subtrajectories of Group 2, the MSD over time was fitted using an exponential equation (Equation 4, Materials and Methods), with the asymptote of the curve corresponding to diameter of confinement (dconf) (Figure 7B,E) and the slope of the curve corresponding to the instantaneous diffusion coefficient (Dins) (Figure 7C,D). Moreover, since PHI of the nanoparticle probe is optically stable, our data covered sufficient time to estimate the asymptote directly from the graph. In fixed cells, the diameter of confinement was 94 nm (Figure 7B,E), whereas in living cells, it was 106 nm. These values diverged when the concentration of FGF2 was increased 10-fold to 220 pM, with the diameter of confinement in fixed cells being reduced to 61 nm, but increased to 122 nm in living cells.
What might the confined motion of FGF2 represent physically in the pericellular matrix? It may be due to the movement of a HS chain to which the FGF2-NP is bound (Figure 8, (c)). Such a view is consistent with the dimension of HS chains: the disaccharide unit is ∼1 nm and a chain is 40 to 160 disaccharides, so the chain is ∼40 to 160 nm long. In addition, the movement of the HSPG core protein may also contribute, since membrane proteins are known to undergo such confinements (Figure 8, (b)) [56]. Alternatively, HS chains and HSPG core proteins may actually be quite immobile. This is supported by the fact that there are many endogenous binding partners of HS chains and HSPG core proteins, which may severely restrict their movements (Figure 8, asterisks). In this instance the FGF2-NP would then be moving around a local network of binding sites on the chains (Figure 8, (a)).
Our data do allow some discrimination between these explanations of the confined motion of FGF2. Unlike the HSPG core proteins, the HS chains will be largely immune to fixatives, including amine reactive ones such as used here. However, the many endogenous proteins bound to HS chains are likely to be immobilised by fixation, so restricting their mobility. Indeed, fixation did reduce the diameter of confinement (Figure 7B,E). This reduction is likely to identify a component of the confined motion that may be due to the movement of the HS chain and/or the confined motion of the core protein. In fixed cells, raising the concentration of the FGF2 10-fold decreased the diameter of confinement (from 94 nm to 61 nm; Figure 7B,E), which is consistent with the increased occupancy of binding sites for FGF2 in a local network of HS chains having a crowding effect. This would decrease the capacity of the FGF2 to explore the entire network of binding sites. Thus, the results suggest that the confined motion of FGF2 represents the combined motion of the HSPG core protein carrying the chain, including the HS chain to which the FGF2 is bound, and of the translocation of FGF2 from binding site to binding site along the same or to a neighbouring HS chain (Figure 8). This translocation would involve the FGF2 in successive cycles of dissociation into the local medium and rebinding. It is likely that the translocation of the FGF2 from site to site is aided considerably by the fact that electrostatic binding dominates the kinetics of the interaction of FGF2 with HS, which ensures rapid rebinding following dissociation into the local medium [7],[10],[34]. The time FGF2 spends undergoing free diffusion in bulk medium is short compared to the measurement time (42 ms), since the tracker cannot measure such a fast event; a much higher time resolution would be needed to identify directly such processes. FGF2 possesses three binding sites for HS, one canonical, higher affinity site (Kd ∼10−8 M to 10−6 M) [11],[42], a secondary site of mM affinity, and a third of even lower affinity [31], which will also increase the probability of re-binding following dissociation. Moreover, these multiple sites may allow the FGF2 to bind to a site on a neighbouring chain, while still attached to its original site and so to move by sliding from one site on the polysaccharide to another.
In living cells, increasing the concentration of FGF2 did not decrease the diameter of confinement, but rather increased it. Thus, the greater freedom of HSPG core proteins and HS chains in the living cells allowed an adaptation to the increased concentration of FGF2. This might occur by the FGF2 competing for binding sites in the HS chains occupied by endogenous proteins, causing the chains to disengage from these and so increase their capacity for movement. In addition, the signalling activity of the FGF2-activated FGFR may affect the movement of FGF2 in the pericellular matrix through inside-out signalling and by changes in protein synthesis altering the extracellular heparin interactome of the pericellular matrix. It is also possible that the increased dissociation of endogenous proteins from HS caused by the increased concentration of FGF2 may have an impact on the intracellular signalling activity of these proteins, which could in turn impact the movement of FGF2 or the HSPG. None of these hypotheses are mutually exclusive.
The long tracking times that PHI allows demonstrate that confinements are interspersed by the various forms of non-confined motion (Groups 3–5, Figure 6, Videos S1, S2, S3, and Table S2). The displacement observed for individual FGF2-NP undergoing motion corresponding to Groups 3, 4, and 5 (Figures 6 and 7, Videos S1, S2, S3, and Table S2) is well beyond the scale of a single HS chain (Figure S2). For the fast and directed motion (Group 5), the diffusion coefficient value (Table S2A,B; ∼0.2 µm2/s) and the shape of the MSD over time (Figure 7) are consistent with the values measured for cytoskeleton-driven active transport [54],[56]. Thus, FGF2-NP motion corresponding to Group 5 may be due to the engagement of the HSPG core protein with cytoskeletal motor proteins. This is supported by the observation that it only occurs in living cells.
Simple and slow directed diffusive motion (Groups 3 and 4, Table S2) occurs in both living and fixed cells. In living cells, a 2-fold increase of the frequency of these non-confined motions (Table S2A,B) was observed compared to fixed cells (Table S2C,D). Therefore, the mobility of the protein core of the HSPG is likely to contribute to these types of motion. However, such motions were still observed in fixed cells. Thus, this suggests that an important mechanism underpinning simple and slow directed diffusive motion is the FGF2 moving from binding site to binding site. The amplitude of displacement of the FGF2 undergoing such diffusive motion corresponds to more than ∼10 HS chains (Figures 6, 7, and S1, Videos S1, S2, S3). Since the FGF2 does not dissociate from the pericellular matrix into the bulk cell culture medium, this indicates that its binding sites on successive HS chains are sufficiently close to enable it to undergo cycles of dissociation into the local medium of the matrix and rebinding to neighbouring sites in HS and/or to slide along and between chains. Therefore, these data suggest that HS-binding sites on multiple chains are spatially aligned so that FGF2 can undergo such major translocations. This is reinforced by the direct observation of some trajectories in fixed cells, such as in Figure 5E (asterisk), where two different FGF2 molecules (in green and purple) were at the same physical location in the pericellular matrix, but separated by several minutes and followed the same path. The observation of such super-imposable trajectories, which last for 101 s for one FGF2 molecule and more than 20 min for the second one, supports the notion that the HS chains form a well-defined path of binding sites for FGF2.
Though HS has a degree of selectivity for its numerous protein partners [20]–[24], it is clear that the motifs in the polysaccharide recognised by FGF2 are representative of the binding sites of a large number of other effector proteins [12]. Thus, the binding sites in HS probed by FGF2-NP in the present work represent structures in the pericellular matrix that will be similarly recognised by not just other FGFs but also many unrelated binding effectors. The data suggest that the HS chains possessing these binding sites in the pericellular matrix may be organised in two quite distinct ways: local networks, which support confinement, and paths, which support non-confined motion (Figure 8). Therefore, this long-range organisation of binding sites in the pericellular matrix is likely to impose similar types of motion on many other HS-binding effectors. The detailed physical properties of motion of each protein would depend on a number of factors. One is the actual binding parameters of the protein for HS, which will determine the properties of the cycles of dissociation and rebinding. This conclusion is supported by recent studies on the ensemble diffusion of FGF7, FGF10 [57], and FGF9 [2]. Another is the level of expression of the protein-binding structures in the HS of a particular matrix, though the HS interactome may be at least as important. This is due to the interactome determining the number of free binding sites in HS. These factors are not independent. For example, introducing a HS-binding effector into a matrix, as done in the present experiments, may alter the balance of interactions of HS chains with the polysaccharide's endogenous interactome and hence the spatial distribution of the effector's binding sites.
All extracellular matrices contain the same general recipe of molecules: HS and HS-binding proteins. Although many effectors mediating cell-cell communication bind similar sites in HS [12], their selectivity and affinity may differ [20]–[24]. Therefore, the structured networks of HS-binding sites presented to effector proteins by a matrix may be sufficiently different in the fine detail of the protein's binding properties to allow the tuning of the movement of different effectors. This would contribute to the shape of effector gradients and the rate of their delivery to target cells and ultimately to their signalling receptors on the cell membrane.
By using a novel gold nanoparticle probe to label FGF2 stoichiometrically, we have been able to determine the spatial distribution of FGF2 from the nano- to the microscale and to measure the dynamics of individual FGF2 at unprecedented time scales. Here we show that the binding sites in the sugar chains of HSPGs are directly involved in the transport of FGF2 within the pericellular matrix. An important mechanism whereby they achieve this is by their HS chains forming local networks ( = confinements) and paths ( = non-confined motion) of binding sites for FGF2. We propose that extracellular matrices are highly structured rather than amorphous. Networks and paths of HS-binding sites consequent of such structure would represent a fundamental mechanism that enables HS-binding effectors to move through matrices and, therefore, drive cell communication in development and disease.
Phosphate-buffered saline (PBS) is 8.1 mM Na2HPO4, 1.2 mM KH2PO4, 140 mM NaCl, and 2.7 mM KCl. Acquisition buffer is 10 mM Hepes pH 7.4, 140 mM NaCl, 5 mM KCl, 2 mM CaCl2, 2 mM MgCl2, and 11 mM glucose, supplemented with 250 µg/mL bovine serum albumin (BSA). Binding buffer is a 9∶1 mixture of PBS∶acquisition buffer, supplemented with BSA, 10 mg/mL. KOAc buffer is 25 mM Hepes, pH 7.4 (KOH), 115 mM potassium acetate, and 2.5 mM MgCl2.
Ten nm diameter Mix-capped gold nanoparticles (HS-PEG∶CVVVT-ol, ratio 30∶70) bearing only one TrisNiNTA function per nanoparticle (TrisNiNTA-NP, n = 1) were prepared and coupled to in-house-produced FGF2 ligand, as described in [25]. Briefly, purified poly-histidine-tagged FGF2 (His-FGF2) at 6.5 µM final concentration was mixed with purified TrisNiNTA-NP, n = 1, at 160 nM final concentration in a 10 µL final volume of PBS supplemented with 0.005% Tween (v/v) (PBST). The reaction was left 2 h at room temperature and PBST then added to a final volume of 200 µL. Centrifugation was performed for 90 min at 17,000 g 4°C, and the supernatant, corresponding to free soluble FGF2 (unlabelled), was removed. The pellet was resuspended in 200 µL of PBST and centrifuged again; a total of five cycles of centrifugation were performed. At the end, the pellet, which corresponds to the purified FGF2-NP conjugate (stoichiometry 1∶1), was resuspended in PBS at a final concentration of 11 nM. Pure recombinant FGF2 protein concentration was calculated using its value of Σ280 nm (1.6×104). FGF2-NP conjugate concentration was calculated using the epsilon value of 10 nm gold nanoparticles, Σ520 nm (9.5×108) [25].
Rama 27 fibroblasts were cultured in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% (v/v) fetal calf serum, 50 ng/mL insulin, and 50 ng/mL hydrocortisone [33].
Proliferation assays were performed as described previously [35]. Briefly, cells were rendered quiescent by 30 h incubation in step down medium (SDM; Dulbecco's modified Eagle's medium supplemented with 250 µg/mL BSA) before the addition of growth factors for 18 h. [methyl- 3H] thymidine (ICN, Basingstoke, UK) was then added directly to the culture medium for 1 h, and radioactivity in DNA, precipitated with 5% (w/v) trichloroacetic acid, was measured by liquid scintillation counting.
SDS PAGE and Western blotting were performed as described in [35] with minor variations. Briefly, after 18 h incubation in SDM, FGF2 or NP-FGF2 (55 pM) were added for 10 or 40 min at 37°C. Plates were placed on ice, washed with PBS, and Laemmli buffer added to extract the proteins. Anti-phospho-p44/42 MAPK (Thr183/202/Tyr185/204) (E10) and anti-phospho FRS2-α (Tyr196) were from Cell Signalling Technology (Hitchin, UK). Anti-actin was from Sigma-Aldrich Co. Secondary peroxidase-labelled anti-IgG antibodies (anti-rabbit and anti-mouse) were from Pierce UK. Visualization was performed using enhanced chemiluminescence (SuperSignal West Dura Substrate, Pierce).
Thirteen mm diameter glass coverslips were washed in ethanol, rinsed with milliQ water, and then used as is. Rama 27 fibroblasts, seeded on a coverslip, were rinsed 3 times with PBS and incubated in 500 µL SDM for 2 h. Three washes with 500 µL PBS were performed and cells were incubated with 100 µL of binding buffer with control TrisNiNTA-NP or FGF2-NP, in absence or presence of DP12 (50 µg/mL) and/or excess of unlabelled FGF2 protein (50 µM). Coverslips were then washed 3 times with PBS and plasma membrane sheets on EM grids were prepared as described in [38]. Cells on a coverslip were pressed onto two coated grids. The coverslip was turned over and 200 µL of KOAc buffer added quickly around the grids to separate them from the coverslip and to generate plasma membrane sheets on the grids (inner leaflet face up). Samples were then fixed with a solution of 4% (w/v) paraformaldehyde and 0.1% glutaraldehyde (v/v) in KOAc for 15 min. The fixative was quenched with three washes with 25 mM glycine in PBS for 10 min in total. Five washes of 2 min were then performed with de-ionized water, and the grids were incubated with a solution of 1.8% (w/v) methyl cellulose, 0.3% (w/v) uranyl acetate for 10 min on ice, and then individually picked up with 5 mm copper wire loops and left to dry for at least 10 min before storage or viewing. Plasma membrane sheets were digitally imaged using an FEI Tecnai G2 120 kV transmission electron microscope and data analysed as described in [38].
PHI, alternatively called LISNA (Laser Induced Scattering around NanoAbsorber), allows detection and tracking of single noble metal nanoparticles down to 2 nm diameter. PHI is a confocal technique with a focal depth of ∼1.6 µm. The optical set-up of the microscope was as described previously [44],[58], with a heating beam intensity of 4 mW/cm2 and an integration time of 1 ms for image acquisition and 7 ms for tracking. Before each experiment, the microscope was calibrated by measuring the mean signal and performing tracks on isolated NP embedded in a thin film of polyvinyl alcohol. Signal to noise ratio was ∼400. To track NP in the pericellular matrix, we used a triangulation procedure knowing the point spread function of the microscope [45]. A 2-D Gaussian fit based on the signal measurement of three points around the NP gives the NP position and the signal intensity. The sampling time Δt is 42 ms. This methodology allows tracking of one NP at a time with a pointing accuracy of ∼10 nm. In our experiments, the calculated diffusion coefficient for NPs embedded in polyvinyl alcohol (a measure of the noise and pointing accuracy of the tracker) was 4.10−6±0.9.10−6 µm2/s (mean ± sem) with a mean square displacement (MSD) after 60 s of 0.17×10−3±1.7×10−6 µm2 (mean ± sem). This gives a deviation length of 13 nm±1.3 nm after 60 s (mean ± sem), which is well below the calculated parameter for mobile FGF2-NP in the pericellular matrix of living and of fixed cells (Table S2, Figure 7B). In PHI, the signal is proportional to the volume of nanoparticle, and n nanoparticles of similar diameter close to each other (≤10–15 nm) will provoke an n-fold increase in the PHI signal intensity [43],[45]. Therefore, the number of nanoparticles in close vicinity of the tracked one can be estimated at each acquisition point of a trajectory. Signal intensity for a single FGF2-NP was calculated at 0.11±0.037 volt (mean ± STD, n = 600,000).
Thirteen nm diameter glass coverslips were washed in ethanol, rinsed with milliQ water, and then used as is. Rama 27 fibroblasts, seeded on a coverslip, were rinsed 3 times with PBS and incubated in 500 µL SDM for 2 h. Three washes with 500 µL PBS were performed and cells were incubated with 100 µL of binding buffer with control TrisNiNTA-NP (22 or 222 pM) or FGF2-NP (22 pM±200 pM unlabelled FGF2) in the absence or presence of DP12 50 µg/mL for 30 min at 37°C. Additional controls were performed by adding FGF2-NP at 22 pM in the presence of a large excess of unlabelled FGF2 (50 µM) or in the presence of unlabelled FGF2 (50 µM) and DP12 (50 µg/mL). Three washes with 500 µL of PBS were performed and cells were placed in 500 µL of acquisition buffer for immediate microscope acquisition.
For fixed cells, following the incubation in SDM, cells were washed 3 times with 500 mL of PBS, rinsed once with 500 µL of fresh paraformaldhehyde solution 4% (w/v) in PBS and then incubated 45 min at room temperature in 500 µL paraformaldhehyde 4% (w/v). PBS washes (5×1 mL) were then performed and 500 mL of binding buffer added. Fixed cells were kept at 4°C in the fridge overnight prior to the addition of the appropriate nanoparticle sample. In some experiments, fixed cells were treated with heparinases I, II, and III (10 mU/mL each in a 100 mM sodium acetate and 0.1 mM calcium acetate buffer, pH 7.0; produced in-house, a kind gift of Prof. Jerry Turnbull, University of Liverpool). Heparinase treatment was achieved by incubating fixed cells overnight at 37°C with 200 µL of the three enzymes at 10 mU/mL prior to washing and labelling with FGF2-NP. Chondroitinase treatment was achieved by adding 20 µL chondroitinase ABC (Sigma) at 333 mU/mL in 100 mM Tris acetate, pH 8.0, to cells in 2 mL step-down medium and incubating overnight prior to washing, fixation, and labelling with FGF2-NP.
PHI images (30×30 µm) were converted to 8 bit greyscale images, thresholded, and colours were inverted. Areas of images 4×4 µm (devoid of mitochondrial signal) were selected, duplicated, and the percentage of labelled pixels (black) versus unlabelled pixels was calculated for each 4×4 µm image using ImageJ software.
All analyses were performed using MATLAB R2009a. Subsequent graphs and statistics analysis were performed using OriginPro 8.5 software.
Each trajectory was characterised by the number of points it contained, N, and four vectors of length N representing sampling time, x and y position, and signal strength, denoted:Each trajectory was split into segments of length s frames, and the net displacement and total distance travelled for each segment was calculated. This created a set of displacement-distance pairs (Pj, Qj) for each trajectory:The displacement-distance pairs of multiple trajectories were plotted on a scatter plot.
First, the selection of obvious confined, slow diffusive, and fast/directed diffusive events, etc., on dozens of trajectory images were performed to obtain the corresponding set of displacement-distance pair (Pj, Qj) values on the scatter plot. This step was repeated using different segments of length s. This allowed the identification of the best choice of the length s and best partitioning of the scatter plot to discriminate the different behaviours within the trajectories. At the limit length s→1, one calculates displacement over time for movement between two points and confinement cannot be measured. As the length s progressively increases, different types of movement inevitably become averaged. With windows of 6, 12, and 18 points we found that the analyses were found to be similar (Figure S1). A window of 12 points, s = 12, corresponding to a time interval of 0.504 s, was used for all subsequent analyses, because it was likely to be more robust than the windows of 6 and 18 points. This led to the identification of five groups, as shown in Figure 6A. “Group 1” (black) was defined according to the parameters obtained for tacking NPs embedded in thin film of polyvinyl alcohol. Note that when plotting the Group 1 population according to the MSD over time, a difference of mobility over time between polyvinyl alcohol embedded nanoparticles and FGF2-NP is seen (Figure 7B). Thus, Group 1 corresponds to the population of molecules that are non-mobile and/or too confined to be accurately discriminated from the noise of the tracker within the time frame of 0.504 s used (displacement lower than 0.036 µm2 with a distance travelled under 0.16 µm within the time frame used). “Group 2” (grey) corresponds to mobile but confined events (displacement lower than 0.11 µm2 within the time frame used, but excluding the events of Group 1); “Group 3” (green) to simple/slow diffusion (displacement between 0.11 to 0.33 µm2 within the time frame); “Group 4” (magenta) to directed diffusion (displacement between 0.33 to 0.68 µm2 within the time frame); and “Group 5” (blue) to unidirectional diffusion events that were only observed in living cells (displacement over 0.68 µm2 within the time frame). These parameters were then used for batch processing all PHI data and for dividing each trajectory into sub-trajectories (Figure 6B,C). Sub-trajectories were constructed by joining together trajectory pieces adjacent in the original trajectory data and belonging to the same behavioural group k in the displacement-distance (Pj, Qj) scatter plot. For a specified group k, each sub-trajectory was characterised by its length, M, and four vectors of length M representing sampling time, x and y position, and signal strength S, denoted:These were further analysed to compute the mean squared displacement (MSD) over time (t) within each sub-trajectory according to the following expressions:(1)The over-bar represents averaging over all sub-trajectories in group k of duration at least t. The diffusion coefficients (D) were calculated according to the following equations [54]:(2)(3)The confinement domain size was obtained by fitting the MSD over time plot of the trajectories in confined motion to the following exponential equation:(4)v in Equation (4) is the velocity and dconf is the measured diameter of confinement and corresponds to the asymptote of the curve. Dins is the instantaneous diffusion coefficient (before the confinement arises) and corresponds to the slope of the curve at the origin.
Statistical analyses were performed using OriginPro 8.5 software. The p values were obtained using Kolmogorov-Smirnov non-parametric test and confirmed using Mann-Whitney non-parametric test. The t values were obtained using Student's t test (parametric).
|
10.1371/journal.ppat.1007115 | A cell-based infection assay identifies efflux pump modulators that reduce bacterial intracellular load | Bacterial efflux pumps transport small molecules from the cytoplasm or periplasm outside the cell. Efflux pump activity is typically increased in multi-drug resistant (MDR) pathogens; chemicals that inhibit efflux pumps may have potential for antibiotic development. Using an in-cell screen, we identified three efflux pump modulators (EPMs) from a drug diversity library. The screening platform uses macrophages infected with the human Gram-negative pathogen Salmonella enterica (Salmonella) to identify small molecules that prevent bacterial replication or survival within the host environment. A secondary screen for hit compounds that increase the accumulation of an efflux pump substrate, Hoechst 33342, identified three small molecules with activity comparable to the known efflux pump inhibitor PAβN (Phe-Arg β-naphthylamide). The three putative EPMs demonstrated significant antibacterial activity against Salmonella within primary and cell culture macrophages and within a human epithelial cell line. Unlike traditional antibiotics, the three compounds did not inhibit bacterial growth in standard microbiological media. The three compounds prevented energy-dependent efflux pump activity in Salmonella and bound the AcrB subunit of the AcrAB-TolC efflux system with KDs in the micromolar range. Moreover, the EPMs display antibacterial synergy with antimicrobial peptides, a class of host innate immune defense molecules present in body fluids and cells. The EPMs also had synergistic activity with antibiotics exported by AcrAB-TolC in broth and in macrophages and inhibited efflux pump activity in MDR Gram-negative ESKAPE clinical isolates. Thus, an in-cell screening approach identified EPMs that synergize with innate immunity to kill bacteria and have potential for development as adjuvants to antibiotics.
| Bacteria evolved molecular machines called efflux pumps to export toxic chemicals, including antibiotics encountered in the environment. Multi-drug resistant (MDR) bacteria use efflux pumps to rapidly transport clinical antibiotics out of the cell and thereby increase the dosage at which they tolerate antibiotics. One way to combat MDR pathogens may be to reduce the activity of efflux pumps and thereby increase pathogen sensitivity to existing antibiotics. We describe an infection-based screen that identified chemicals that inhibit bacterial efflux pump activity and show that these compounds bind to and block the activity of bacterial efflux pumps.
| Human pathogens have become increasingly resistant to clinical antibiotics. Gram-negative bacterial pathogens are particularly problematic because their outer membranes are impermeable to many chemicals, and because many compounds that do enter the periplasm or cross the cellular membrane are immediately exported by efflux pumps. Multi-drug resistant (MDR) bacteria typically have increased gene copy number and/or production of efflux pumps, features demonstrated to contribute to the failure of clinical antibiotic treatment [1]. For these reasons, compounds that reduce efflux pump activity (efflux pump modulators, EPMs) are under investigation for their potential use in re-sensitizing MDR pathogens to existing antibiotics [2].
Three synthetic small molecules with EPM activity against Gram-negative bacterial pathogens have been well characterized. Phe-Arg β-naphthylamide (PAβN) was identified in a screen for compounds that increase the sensitivity of Pseudomonas aeruginosa to levofloxacin, an antibiotic and efflux pump substrate [3,4]. PAβN binds AcrB, the main component of the efflux system AcrAB-TolC, a member of the RND (resistance-nodulation-cell division) family of pumps. However, this compound was not developed as an antibiotic because medicinal chemistry could not separate EPM activity from unfavorable pharmacokinetics and toxicology, possibly reflecting off-target effects [5,6]. A second series of EPMs was identified in the same screen as PAβN. These pyridopyrimidines were subjected to medicinal chemistry, and the lead compound D13-9001 has efficacy against Pseudomonas aeruginosa during infection of rats [7,8]. Finally, a screen for chemicals that increase the sensitivity of E. coli to ciprofloxacin identified the pyranopyridine MBX2319 as an EPM that targets AcrB and has activity against multiple Enterobacteriaceae [9–12]. We identified three compounds that have activity as EPMs using a different approach—an in-cell screen for small molecules that prevent the replication of the Gram-negative pathogen Salmonella enterica (Salmonella) in mammalian cells.
SAFIRE (Screen for Anti-infectives using Fluorescence microscopy of IntracellulaR Enterobacteriaceae) is a high-content, medium-throughput screening platform that identifies compounds active against Gram-negative bacteria within the context of host cells (Fig 1A). The platform uses fluorescence microscopy and automated image analysis to monitor Salmonella within RAW 264.7 cells, a macrophage-like cell line in which the virulent Salmonella laboratory strain SL1344 replicates 10-15-fold [13–15].
We used SAFIRE to screen 14,400 compounds from a drug-like diversity library, the Maybridge HitFinderTM v11 [16–19]. Macrophages in 384-well plates were infected with Salmonella expressing GFP under the control of a promoter, sifB, that is induced within macrophages (Table 1) [20]. After 45 minutes, infected macrophages were treated with the antibiotic gentamicin to prevent the replication of extracellular bacteria [21]. At two hours post-infection, test compounds [25 μM] were added. The compounds remained for the duration of the experiment. At 17.5 hours post-infection, when optimal Salmonella replication was observed, cells were stained with a marker of macrophage vitality, MitoTracker Red CMXRos, to help identify compounds toxic to eukaryotic cells. Thirty minutes later, cells were fixed and incubated with DAPI to stain DNA and imaged on an automated microscope. A MATLAB-based algorithm was used to quantify bacterial infection, specifically, the percentage of infected cells. Macrophage boundaries were first established using MitoTracker and DAPI signals (Fig 1B). The percentage of infected cells was determined by setting a threshold for the GFP signal based on infected and uninfected controls; cells containing at least two GFP-positive pixels were labeled infected. The library was screened in duplicate, and well-to-well variability was addressed using B-score normalization [22–24]. Assay positives were called based on an activity threshold greater than one standard deviation below the mean B-score, and a p-value of less than 0.05 calculated using a modified t-test assuming an inverse gamma distribution of variances (Fig 1D) [23,25]. The micrographs of the 461 assay positives were manually reviewed to eliminate host-toxic and/or autofluorescent chemicals. The remaining 309 hits were retested using SAFIRE in the 96-well format and ranked based on reduction of bacterial load as determined by SAFIRE and traditional lysis and plating for colony forming units (CFU) (S1 Table). Sixty-four (85%) of the top 75 compounds, including chloramphenicol, reduced CFU by at least 25% (S1 Fig). The top 60 compounds (excluding chloramphenicol) were repurchased and validated using SAFIRE in a 96-well format; 58 repurchased compounds were active.
SAFIRE has the potential to identify EPMs because Salmonella requires at least two efflux pumps, AcrAB and MacAB, to replicate and/or survive within macrophages and mice [30–34]. The fluorescent dye Hoechst 33342 is an efflux pump substrate, and increased Hoechst accumulation relative to controls identifies potential modulators of efflux pumps [35]. We incubated bacteria with each of the 58 repurchased, validated hits and Hoechst 33342. As expected, heat-killed bacteria exhibited high fluorescence immediately after exposure to Hoechst because an electrochemical gradient is required to export pump substrates. Live, wild-type Salmonella demonstrated low fluorescence, and a strain lacking the AcrAB efflux pump had a modest level of fluorescence. PAβN treatment resulted in higher levels of fluorescence, as expected [35]. Under the same conditions, treatment with three of the 58 compounds (EPM30, EPM35 and EPM43) resulted in fluorescence comparable to that of PAβN (Fig 2A). Further examination revealed that the three compounds had half maximum effective concentrations (EC50s) four-fold lower than that of PAβN in the Hoechst assay (Fig 2B). These observations show that the hit compounds increase bacterial accumulation of an efflux pump substrate.
We next tested the ability of the three hit compounds to inhibit efflux of nitrocefin, a chromogenic beta-lactam and known AcrAB substrate [9,36]. E. coli strain RAM121 encodes a porin with a large diameter, which allows rapid influx of nitrocefin, and the periplasmic AmpC beta-lactamase, which hydrolyzes nitrocefin and results in a color change. Carbonyl cyanide m-chlorophenylhydrazone (CCCP) is a protonophore that inhibits efflux pumps, causing increased nitrocefin hydrolysis (Fig 2C and 2D). Treatment with EPM30 or EPM35 yielded a similar result, whereas treatment with EPM43 only modestly increased hydrolysis. Thus, the three hit compounds (Fig 2E) may inhibit bacterial efflux pumps.
We performed a more thorough characterization of the putative EPMs regarding anti-Salmonella activity in multiple mammalian cell types. Micrographs from RAW264.7 macrophages treated with 25 μM of each compound demonstrated a significant reduction in the percentage of GFP-positive cells compared to treatment with vehicle alone (Fig 3A). The SAFIRE inhibitory concentration-50 (IC50) for the three compounds in macrophages ranged from 3 to 7 μM (Fig 3B). To establish whether reduced GFP signal correlates with bacterial killing, we quantified bacterial survival by enumerating CFU from infected cells. The IC50s for the 3 compounds by CFU ranged from 2 to 5 μM (Fig 3C). HeLa cells harboring a Salmonella-GFP expressing strain and treated with 25 μM of each compound also demonstrated a reduction in the percentage of GFP-positive cells compared to treatment with vehicle alone (Fig 3D and 3E). In primary bone marrow-derived mouse macrophages (BMDMs) all three hit compounds reduced the number of recoverable Salmonella by approximately 20-fold (Fig 3F). Finally, while the macrophage and HeLa cell culture assays require gentamicin to prevent the replication of extracellular Salmonella, we established in broth assays that the three compounds do not synergize with the antibacterial activity of gentamicin (S2 Fig). Thus, the putative EPMs inhibit bacterial replication and/or survival in at least two cell types, macrophages and epithelial cells, which are relevant to whole animal infection.
We next examined whether the bacterial load of MDR Salmonella in macrophages is reduced upon treatment of infected cells with the hit compounds. A clinical MDR Salmonella isolate (S10801) was recovered from hit compound-treated macrophages at levels 1000-fold lower than from DMSO-treated macrophages (Fig 4). These results indicate that the three compounds inhibit not only SL1344 but also an MDR clinical isolate during infection of cells.
Clinical MDR isolates frequently express high levels of efflux pumps [1]. The two Salmonella efflux pumps needed for bacterial survival in cells and mice, AcrAB and MacAB, both use the TolC channel to export cargo across the outer membrane [30–34,37]. We first confirmed that Salmonella strains lacking acrAB, macAB, or tolC replicate poorly in macrophages compared to the wild-type parent strain (Fig 4). Treatment of macrophages with any of the three compounds reduced loads of wild-type bacterial below those observed in DMSO-treated cells infected with the acrAB, macAB, or tolC mutant strains. Compound treatment of the mutant strains in macrophages further reduced the levels of mutant bacteria. Given that Salmonella encodes other efflux pumps that may contribute to survival in macrophages in the absence of macAB, acrAB, or tolC [34,38,39], these data could suggest that the hit compounds target other efflux pumps.
Having established that the hit compounds are antimicrobial in mammalian cells, we returned to the analysis of their activity. While the Hoechst accumulation assay is a good first approximation of anti-efflux pump activity, quantification of export in real time based on energy (glucose) dependence is a more specific assay for pump inhibition.
Ethidium bromide is an efflux pump substrate that fluoresces upon intercalating into DNA. We pre-loaded cells with ethidium bromide and then treated with glucose to energize the efflux pumps and stimulate export [40]. Incubation with PAβN or any of the three putative EPMs reduced ethidium bromide export upon glucose addition in a dose-dependent manner (Fig 5A; S3A Fig). A similar assay using the efflux pump substrate Nile red further demonstrated that EPM30 and EPM35 reduced pump export (Fig 5B; S3B Fig; S4 Fig). Nile red becomes strongly fluorescent upon partitioning into the cytoplasmic membrane and possibly the inner leaflet of the outer membrane but is rapidly exported [41–44]. Washout of the EPMs still reduced Nile red export, suggesting the activity of these compounds is not readily reversible (Fig 5C; S3C Fig). These observations indicate that the three hit compounds may inhibit energy-dependent efflux pump activity.
Since efflux pumps rely upon the proton motive force or ATP to provide the energy for the transport of substrates, chemicals that disrupt the inner membrane may indirectly inhibit efflux. To establish whether the three EPMs alter bacterial inner membrane potential, we observed their effect on the incorporation of the voltage-sensitive dye tetramethylrhodamine methyl ester (TMRM). After 30 minutes of exposure to the ionophore CCCP, TMRM levels in cells were approximately 50-fold lower than upon treatment with DMSO, but treatment with any of the three EPMs did not alter TMRM signal based on analyses by flow cytometry (Fig 6A and 6B). These observations suggest that membrane potential remains intact in the presence of the EPMs. To establish whether a longer incubation with the EPMs may compromise membrane integrity, we monitored the effect of the EPMs on bacterial swimming, an energy intensive activity, over 15 hours [45] (Fig 6C; S5 Fig). Salmonella were injected into the center of soft-agar plates and 10 μl of compound was pipetted onto paper disks on the periphery [45]. Control compounds included CCCP and PAβN, which disrupts membranes over long (> 30 minutes) exposures [46–48]. Since the swimming assay also requires bacterial growth, we tested whether filters containing bacteriostatic antibiotics that are not known to disrupt membranes prevented swimming. Neither the three EPMs nor the conventional antibiotics inhibited swimming relative to their MIC, as compared to CCCP and PAβN, further suggesting that the EPMs do not interfere with bacterial energy production across the inner membrane.
A second class of chemicals that appears to interfere with bacterial efflux does so by permeabilizing the outer membrane, which allows substrates to diffuse into the periplasm [36]. We therefore tested whether exposure of bacteria with an intact outer membrane to the EPMs would enable the beta-lactam nitrocefin to enter the periplasm, as monitored by nitrocefin hydrolysis in bacteria with high expression levels of bla beta-lactamase [49]. In this assay, monitoring of absorbance showed that CCCP had no effect on nitrocefin hydrolysis, whereas polymyxin B, a pore-forming antimicrobial peptide, increased hydrolysis by 10-fold or more. The efflux pump inhibitor PAβN slightly increased hydrolysis but the EPMs did not (Fig 6D and 6E). Thus, the EPMs did not appear to increase nitrocefin access to the periplasm, suggesting they do not increase bacterial outer membrane permeability.
Since all three EPMs reduced Nile red or ethidium bromide export in Salmonella, well-studied substrates of the AcrAB-TolC efflux system [40,50], we established whether any of the compounds bind E. coli AcrB using isothermal titration calorimetry (ITC) [51–53] (Fig 7; S6 Fig; Table 2). The equilibrium dissociation constants (KDs) are within the micromolar range, demonstrating that AcrB is capable of binding each of the EPMs. It appears that AcrB interacts with these three EPMs using different binding mechanisms. The free energies of binding for EPM35 and EPM43 are dominated by the large negative enthalpies (ΔHs) of -7.7 and -6.0 kcal/mol, respectively. However, the binding for EPM30 is predominantly determined by its entropic contribution (-TΔS), which is -6.8 kcal/mol at 25 oC. These data suggest that the AcrB efflux pump tends to bind the EPMs with a 1:1 protein monomer-to-ligand molar ratio.
To establish whether the hit compounds have minimum inhibitory concentrations (MICs) in broth that are similar to their IC50s in host cells (Fig 3B and 3C), we examined bacterial growth in their presence using standard rich laboratory media, Mueller Hinton Broth (MHB). The MICs of EPM30, EPM35, and EPM43 respectively were 100 μM [36 μg/mL], 400 μM [186 μg/mL], and >400 μM [113 μg/mL], for two Salmonella strains: the strain used in the SAFIRE screen (SL1344), and a clinical MDR Salmonella isolate (S10801). These broth MIC values (> 100 μM) are considerably higher than the IC50s observed in host cells (< 10 μM). Thus, the three putative EPMs may not function like traditional antibiotics and yet are potent in the context of the host cell.
We next addressed why the EPMs kill bacteria in mammalian cells at concentrations 10-fold lower than they inhibit efflux in broth. One possibility is that the presence of antimicrobial peptides (AMPs) within host cells plays a role. Mammalian cells constitutively express AMPs and increase AMP expression in response to infection [54,55]. We found that in broth the combination of each EPM with either the bacterial-derived polymyxin B or the human cathelicidin AMP LL37, but not individual treatments, significantly inhibited Salmonella growth; EPM35 and EPM43 display stronger synergy than EPM30 in this regard (Fig 8). These data suggest at least three possibilities: AMPs may potentiate EPM activity, EPMs may potentiate AMP activity, or both.
To distinguish between these possibilities, we first determined that bacterial exposure to polymyxin B concentrations high enough to allow nitrocefin access to the periplasm (5 μg/mL, Fig 6D and 6E) did not enhance the ability of the EPMs to increase Hoechst accumulation (S7A Fig). Similarly, co-treatment of polymyxin B with EPMs did not synergistically increase Nile red retention compared to polymyxin B or EPMs alone (S7B Fig). These observations suggest that the membrane-damaging activity of polymyxin B did not potentiate EPM blockage of efflux pumps. We next observed that low concentrations of polymyxin B [1 μg/mL], which do not by themselves allow nitrocefin access to the periplasm (Fig 6D and 6E), increased the rate of nitrocefin hydrolysis in a bla-expressing Salmonella strain when EPMs were present [25 μM] (Fig 9). It thus appears that EPMs potentiate AMPs with regard to both nitrocefin entry into the periplasm and bacterial growth inhibition. Therefore, EPMs may decrease the effective concentration of AMPs and, for this reason, have indirect antibacterial activity in the context of the host.
We next established whether co-incubation with EPMs reduced the MIC of established AcrB substrates in the wild-type SL1344 strain and the S10801 clinical isolate. Exposure to EPM35 or EPM43 decreased by four-fold the MIC of chloramphenicol for one or both strains (Table 3). EPM35 was also synergistic with tetracycline and erythromycin. Similar effects were observed in the SL1344-derived macAB mutant strain but not in the acrAB or tolC mutant strains. These data suggest that the AcrAB-TolC efflux pump may be a relevant target for EPM35 and EPM43 with regard to antibiotic potentiation, and more importantly, that these EPMs appear to reduce the effective dose in broth of some clinical antibiotics.
Since intracellular conditions are distinct from microbiological media, we tested whether the EPMs also synergize with AcrAB-exported antibiotics for bacterial killing in macrophages. RAW264.7 macrophages were infected with wild type Salmonella followed two hours later by treatment with antibiotic over a dose range with or without an EPM [6.25 μM]. SAFIRE analysis revealed that both erythromycin and ciprofloxacin were potentiated by the EPMs, as indicated by a difference between the co-treatment data and calculated additivity curves (Fig 10). Thus, the EPMs may reduce the effective dose of erythromycin and ciprofloxacin in macrophages.
Six pathogens that cause the bulk of MDR nosocomial infections have been dubbed the ESKAPE pathogens: Enterococcus faecium, Staphylococcus aureus, Klebsiella pneumoniae, Acinetobacter baumannii, Pseudomonas aeruginosa, and Enterobacter species [56]. EPM35 and EPM43 significantly reduced Nile red export in MDR clinical isolates of K. pneumoniae and Enterobacter cloacae in addition to E. coli (Fig 11), suggesting these two compounds have biological relevance in MDR strains beyond Salmonella.
In-cell screens aimed at identifying chemicals that prevent pathogen intracellular replication have been described. For example screens of FDA-approved drug libraries have uncovered modulators of Listeria monocytogenes or Salmonella enterica infection [57,58]. Similar screens were performed with macrophages infected with Mycobacterium tuberculosis [59] or the yeast Cryptococcus neoformans [60]. There are also many reports of high-content screens aimed at identifying genetic disruptors of host-pathogen interactions [61–63]. SAFIRE adopts what we thought would be the most useful aspects of these earlier studies, including a GFP-expressing Salmonella to track the microbe, the addition of compounds after infection, automated fluorescence microscopy-based image analysis with MATLAB, and estimation of compound toxicity based on manual visualization of cell morphology and MitoTracker staining. In addition, because the SAFIRE assay includes serum, it avoids compounds that are poor candidates for drug development because of high affinity for serum proteins [64]. These features enabled us to develop SAFIRE as a medium-throughput assay useful for identifying small molecules that interfere with the host-pathogen relationship.
Pathogens require efflux pump activity to survive in host tissues, suggesting modulators of bacterial efflux may be identified with in-cell screens for pathogen survival [30,34]. Salmonella encodes nine efflux pumps [65]. The two demonstrated to be required for bacterial survival in cells and in mice are AcrAB-TolC and MacAB [30–33,39]. The first hint that several of the hit compounds may modulate bacterial efflux was the observation that treatment of wild-type Salmonella with PAβN or with any of the three EPMs allowed Hoechst to accumulate to higher levels than in vehicle-treated bacteria. We speculate that treatment with PAβN or the EPMs was more effective at raising Hoechst levels than was deletion of the acrAB::kan locus because the compounds may target other efflux pumps. We also note that EPMs are not expected to function as clinical antibiotics: EPMs have high MICs in standard broth-based assays [66]. However, EPMs are of interest because of their potential to enhance the activity of existing antibiotics and/or host antimicrobials. These properties further underscore the biology of efflux pumps and highlight the importance of looking beyond MIC assays to identify chemicals with antimicrobial activity under conditions that approximate infection.
All three of our hit compounds bind the efflux pump subunit AcrB, a subunit of the most thoroughly studied RND efflux pump. AcrB integrates into cellular membranes and captures substrates from the outer leaflet of the cytoplasmic membrane or the periplasm [67–69]. The compounds identified bind AcrB more tightly than several known AcrB substrates, such as ethidium (KD of 8.7 +/- 1.9 μM), proflavin (KD of 14.5 +/- 1.1 μM), and ciprofloxacin (KD of 74.1 +/- 2.6 μM) [70]. The chemical structures of the three compounds have some resemblance to known efflux pump inhibitors (Fig 2E). EPM30 is a small compound with an aminothiazole core, and several aminothiazole compounds have been identified that inhibit efflux [7,71]. EPM35 is a trifluoro-pyrimidine linked to a piperidine. A very similar compound was suggested to bind the AcrB substrate-binding pocket in an in silico screen [72]. EPM43 is a small quinazoline, a planar moiety which is a common drug pharmacophore. Other quinazolines have been identified as inhibitors of bacterial and fungal efflux pumps [73,74]. EPM43 itself has been identified as an inhibitor of fungal dihydrofolate reductase (DHFR), but is not known to inhibit bacterial or human DHFR [75,76]. Where on the AcrB protein the EPMs bind remains unknown. EPM35 and EPM43 potentiate multiple AcrB substrates, suggesting they bind in the hydrophobic trap [77,78]. Alternatively, the EPMs may bind outside of the substrate pocket and, for instance, disrupt AcrB folding, localization or interactions with AcrA. It is notable that the three EPMs do not behave identically in broth assays that monitor export of AcrB substrates, potentiation of antibiotics, or activity against other Gram-negative pathogens, emphasizing that they may not interact identically with AcrB and/or any other molecules they may target.
Why the three EPMs are more potent as antibacterials in mammalian cells than they are as efflux pump inhibitors in broth is not completely clear. A simple model supported by existing data is that the EPMs increase bacterial sensitivity to host AMPs by binding efflux pump subunits, thereby reducing AMP export [79] and decreasing the effective concentration of AMPs. During infection of a whole animal, endogenous AMPs, which are ubiquitous in body fluids, may synergize with EPMs, even in severely immunocompromised patients for whom innate immunity typically remains intact [80,81]. While Salmonella RND efflux pumps have not been demonstrated to export AMPs, it is nevertheless encouraging that two of the EPMs inhibit efflux in other major MDR bacterial pathogens, suggesting they may have utility beyond Salmonella. Another possible explanation for higher potency during infection than in broth is that the EPMs could accumulate in the SCV, thereby increasing the concentration of compound experienced by the bacterium within host cells. An EPM could also target the host cell and, for instance, increase production of antimicrobial mediators. Alternatively, or in addition, EPMs may interfere with other bacterial processes and/or bind targets that are not present or accessible under the broth conditions tested. To facilitate our understanding of how the EPMs function, it may be useful to identify more potent, less toxic chemical derivatives. Desirable derivatives would have efficacy in SAFIRE, potency in efflux assays, and, most importantly, resensitize MDR pathogens to clinical antibiotics by reducing the antibiotic dosage needed to treat an infection.
The work presented as a whole suggests that SAFIRE enables discovery of chemicals that interfere with the host-pathogen relationship and may have potential as lead compounds for therapeutic development. Advantages of SAFIRE-identified chemicals include that they are unlikely to be inactivated by serum and are fairly non-toxic to mammalian cells. As proof-of-principle, SAFIRE identified not only a traditional antibiotic, chloramphenicol, but also three small molecules with activity against Gram-negative bacterial pathogens. Moreover, because SAFIRE identifies compounds that decrease bacterial load within mammalian cells, the platform may single out chemicals with antibacterial activity that is facilitated by endogenous host antimicrobial peptides, which are broadly distributed across extracellular and intracellular niches. In summary, SAFIRE followed by secondary screening has the potential to identify new and previously overlooked compounds that may be useful as lead compounds for biological and/or antibacterial discovery.
Animal work was carried out in accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. All protocols were approved by the University of Colorado Boulder Institutional Committee for Animal Care and Use (protocol number 2445). Euthanasia was carried out by carbon dioxide asphyxiation followed by cervical dislocation.
The wild-type S. enterica serovar Typhimurium strain, SL1344, was initially isolated from the blood of an infected calf [26]. For screening and validation in macrophages, SL1344 sifB::GFP [20] was grown in Luria-Bertani Broth (LB) with 30 μg/ml streptomycin and 30 μg/ml kanamycin to saturation overnight, diluted to an OD of 0.001 and frozen in 100 μL aliquots in 20% glycerol at -80°C. Prior to infection, aliquots were grown in 5 mL cultures of LB with 30 μg/ml streptomycin and 30 μg/ml kanamycin for 18 hours at 37°C with aeration. Bacterial strains were routinely grown in LB with antibiotics (Table 1): 30 μg/ml streptomycin, 30 μg/ml kanamycin, 50 μg/ml ampicillin, 10 μg/ml tetracycline, and/or 1.15 μg/ml meropenem. The acrAB::kan and macAB::kan strains were constructed as described [82]. S. enterica subsp. enterica, serovar Typhimurium strain S10801, NR-22067 is a multidrug resistant isolate from a calf with sepsis [83]. This strain and others as indicated (Table 1) were obtained through BEI resources, NIAID, NIH.
Murine macrophage-like RAW 264.7 cells and HeLa human epithelial cells were obtained from the American Type Tissue Collection. BMDMs were isolated as previously described [21]. Briefly, marrow was flushed from the femurs of 1- to 4-month-old 129SvEvTac mice (Taconic Laboratories) bred in-house. Mononuclear cells were separated using Histopaque-1083 (Sigma), washed, and directly seeded into assay plates at 1 x 105 cells/ml in complete medium supplemented with 35% conditioned media from 3T3 cells expressing MCSF. Media were refreshed three days later. After 1 week, media were replaced with 100 μL fresh media and cells were infected as described below. All three types of cells were grown in DMEM high glucose (Sigma) supplemented with 10% fetal bovine serum, 2 mM L-glutamine, 1 mM sodium pyruvate, 10 mM HEPES, and 50 μM β-mercaptoethanol. Cells were maintained in a 5% CO2 humidified atmosphere at 37°C. For screening, frozen aliquots of RAW 264.7 were thawed and allowed to expand for three days prior to seeding; other experiments were performed with cultures between passages four and 20.
SAFIRE—RAW 264.7 macrophages (7 x 103 in 40 μL or 5 x 104 in 100 μL) were seeded, respectively, in 384- or 96-well black-walled glass-bottomed plates (Brooks Automation). Twenty-four hours post-seeding, bacteria in 20 or 50 μL PBS were added to a final concentration of 1 x 107 CFU/mL, conditions yielding infection of approximately 70% of macrophages at 18 hours post-infection with minimal macrophage toxicity. The sifB::GFP bacterial reporter strain was used to minimize green signal from extracellular bacteria. Forty-five minutes after bacterial addition, 20 or 50 μL gentamicin was added to a final concentration of 40 μg/mL, which did not affect intracellular infection but inhibited replication of extracellular bacteria. At two hours post-infection, 200 or 500 nL compound was added using a pin tool (CyBio) to yield a final concentration of 25 μM. Each assay plate included rifampicin and DMSO controls. In some experiments, media were removed and replaced with fresh media containing 40 μg/mL gentamicin and the indicated concentrations of drugs. At 17.5 hours post-infection, PBS containing MitoTracker Red CMXRos (Life Technologies) was added to a final concentration of 300 nM or 100 nM, for 384- or 96-well, respectively. Thirty minutes later, 16% paraformaldehyde was added to a final concentration of 1–2% and incubated at room temperature for 15 minutes. Wells were washed twice with PBS and stained for 20 minutes with 1 μM DAPI; wells were washed twice and stored in 90% glycerol in PBS until imaging. The Z’-factor of the screening platform was 0.59 and 0.48 in 96-well and 384-well plates, respectively, within published ranges for complicated cell-based screens [22,60–62].
HeLa cells—Infections of HeLa cells with Salmonella were performed as above except that 1 x 104 cells were seeded, and cells were infected with Salmonella constitutively expressing GFP from the rpsM locus because sifB::GFP is poorly expressed in HeLa cells. In addition, plates were spun for five minutes at 500 x g after addition of bacteria to enhance infection.
CFU—Infections were performed as described above, except cells were seeded in 96-well tissue culture coated plates (Greiner). At 18 hours post-infection, wells were washed three times in PBS, lysed with 30 μL 0.1% Triton X-100, diluted and plated to determine CFU.
High magnification images were acquired on an Olympus IX81 inverted widefield microscope (40X) or CV1000 spinning disk confocal microscope (20X). For screening, three-color images were acquired at 10X on a Cellomics ArrayScan VTI (Thermo) and exported to DIB files. At least two fields were imaged per well for all experiments, and compounds were screened in duplicate. To quantify intracellular bacterial load we developed an automated MATLAB script (“SAFIRE_ArrayScan”, “SAFIRE_OlympusIX81” and “SAFIRE_CV1000” on MATLAB File Exchange, https://www.mathworks.com/matlabcentral/fileexchange/). Briefly, the algorithm identifies macrophage borders via watershed segmentation using DAPI and MitoTracker. To identify bacteria, the user supplies an empirically determined GFP threshold that maximizes signal to noise based on uninfected and untreated controls. Within each macrophage, the number of pixels above the GFP threshold is counted. If more than two pixels are above the GFP threshold, the macrophage is labeled infected. The script calculates the percentage of macrophages infected in the image. Raw data for at least two images from the same well are averaged to yield one value for each well. Raw screening data were subjected to B-score normalization because we identified significant row and column effects by a method previously described [22]. To determine significance, we employed the modified one-sample t-test [23] by fitting the variances of replicates to an inverse gamma distribution [25]. The micrographs of the 461 preliminary positives were examined by the human eye to eliminate compounds that clearly destroyed the macrophages (host toxic). Assay positives were defined as having a p-value less than 0.05 and a B-score one standard deviation below the mean.
Overnight Salmonella cultures were washed three times in PBS and diluted to an OD600 of 0.01 in MHB in 96-well flat-bottom plates. Compound was added using a pin tool (CyBio) or manually, yielding a final concentration of no more than 1% DMSO. Plates were grown at 37°C shaking and OD600 was monitored using a BioTek Eon or Synergy H1 incubator shaker microplate absorbance reader. MICs were defined as the concentration at which no growth was visually observed. The Fractional Inhibitory Concentration Index (FICI) was calculated for wells showing no visible growth (FICI = [agent A] / [MIC agent A] + [agent B] / [MIC agent B]) [84]. For experiments with polymyxin B, bacteria were grown in M9 minimal media supplemented with 100 mM Tris pH 7.4, 0.35% glycerol, 0.002% histidine, 10 mM MgCl2, and 0.1% casamino acids and 5 μg/mL polymyxin B. For experiments with LL37, bacteria were grown in M9 minimal media supplemented with 0.4% dextrose, 0.004% histidine, 1 mM MgSO4, and 5 μg/ml LL37.
Hoechst accumulation assays were performed essentially as described [35]. Briefly, overnight Salmonella cultures were washed three times in PBS and diluted to an OD600 of 0.1 in PBS with 2.5 μM Hoechst 33342 in the presence of the indicated concentrations of compounds. Fluorescence was monitored on a Biotek Synergy 2 with a 360/40 nm excitation filter and 460/40 nm emission filter. The maximum Hoechst fluorescence over 60 minutes of incubation was normalized to the signal from the equivalent number of heat-killed bacteria, after subtraction of autofluorescence signal determined from compound incubated in the absence of bacteria. EC50s were determined using a 4-parameter nonlinear fit constrained using DMSO-treated Wild-Type as the minimum and Heat Killed as the maximum (GraphPad Prism). Concentrations of PAβN used in all assays were determined by titration with the Salmonella wild-type strain in the corresponding assay.
Nile red assays were adapted from an established protocol [50]. Briefly, overnight Salmonella LB cultures were washed in PBS with 1 mM MgCl2 and resuspended at an OD600 of 2.0. Cells were incubated in 10 μM Nile red for three hours at 37°C in glass tubes on a roller drum and then at room temperature standing for one hour. Cells were pelleted at 2,050 x g, resuspended in PBS with 1 mM MgCl2, and 200 μl was added to 96-well black walled plates (Greiner) with compound at the indicated concentrations. In washout experiments, after 35 minutes of incubation with compound, cells were centrifuged at 16,000 x g, resuspended in PBS with 1mM MgCl2 without compound and aliquoted into 96-well black walled plates (Greiner). During loading into plates (~ 20 minutes), bacteria effluxed some Nile red even in the absence of glucose (S4 Fig). Samples were read using a Varioskan Flash Multimode Reader at 540 nm (excitation) and 625 nm (emission) or a Biotek Synergy H1 at 560 nm and 655 nm. To activate efflux, glucose was added to a final concentration of 2 mM.
Ethidium bromide (EtBr) assays were performed similarly to Nile red assays with the following changes. Cells were incubated with 10 μM of carbonyl cyanide m-chlorophenylhydrazone (CCCP) for 15 minutes, then incubated with 10 μM of EtBr for one hour at 37°C with aeration, pelleted at 2,050 x g, aliquoted and combined with compound at the indicated concentrations in 96-well black walled plates (Greiner). Plates were monitored with a Biotek Synergy H1 at 510 nm and 600 nm.
Bacteria were subcultured to mid-log phase, washed, and combined with 100 μM nitrocefin and the indicated concentrations of drugs in 96-well plates in 200 μL [41]. Washes and incubations were performed in 20 mM KPO4, pH 7.0, 1 mM MgCl2. Absorbance (486 nm) was measured on a BioTek Eon or Synergy H1 spectrophotometer every 60 seconds for one hour. To observe efflux inhibition, E. coli RAM121 [28] were added to plates at a final OD600 of 10. This strain produces an OmpC variant with a larger pore size to allow increased influx of nitrocefin and other bulky molecules, and nitrocefin is hydrolyzed by the endogenous AmpC beta-lactamase. To measure outer membrane permeability, wild-type Salmonella harboring beta-lactamase (bla)-expressing pACYC177-mTagBFP2 [29] were added to plates at a final OD600 of 0.1.
Wild-type Salmonella harboring pACYC177-mTagBFP2 [29] were subcultured to mid-log phase, diluted to 1 x 106 CFU/ml in PBS, aliquoted into flow cytometry tubes, and treated with the indicated concentrations of compounds. TMRM was immediately added to a final concentration of 100 nM. After incubation for 30 minutes at 37°C, samples were analyzed on a CyAn ADP (Beckman Coulter) in channels FL6 and FL2. Data were analyzed using FlowJo; bacteria were gated based on side scatter and BFP signal in the FL6 channel, and the FL2 median fluorescence intensity (MFI) was calculated.
Saturated overnight cultures were diluted to an OD600 of 0.01 in LB, and 1 μL was injected into the center of low (0.25%) agar LB plates. Ten microliters of the indicated compounds up to concentrations of 100 mM were added to sterilized Whatman paper disks (diameter 0.7 cm) placed equidistant from the plate center; solubility issues arose at concentrations above 100 mM. Plates were incubated face up at 37°C overnight; no change in halo size was observed between 14–24 hours incubation. Plates were imaged using a Gel Logic 200 imaging system, and halo radius (distance from center of disk to outermost edge of halo) was measured using ImageJ.
ITC was used to examine the binding of EPMs to the purified AcrB transporter. Measurements were performed on a Microcal iTC200 (Northampton, MA) at 25°C. Before titration, the protein was thoroughly dialyzed against buffer containing 20 mM Na-HEPES (pH 7.5), 0.05% n-dodecyl-μ-maltoside (DDM) and 5% dimethyl sulfoxide (DMSO). The protein concentration was determined using the Bradford assay. The protein sample was then adjusted to a final monomeric concentration of 10 μM. Ligand solution consisting of 100 μM EPM30, EPM35 or EPM43 in 20 mM Na-HEPES (pH 7.5), 0.05% DDM and 5% DMSO was prepared as the titrant. The protein and ligand samples were degassed before they were loaded into the cell and syringe. Binding experiments were carried out with the protein solution (0.2 ml) in the cell and the ligand as the injectant. Two microliter injections of the ligand solution were used for data collection.
Injections occurred at intervals of 60 s, and the duration time of each injection was 4 s. Heat transfer (μcal/s) was measured as a function of elapsed time (s). The mean enthalpies measured from injection of the ligand in the buffer were subtracted from raw titration data before data analysis with ORIGIN software (MicroCal). Titration curves were fitted by a nonlinear least-squares method to a function for the binding of a ligand to a macromolecule. Nonlinear regression fitting to the binding isotherm provided us with the equilibrium binding constant (KA = 1/KD) and enthalpy of binding (ΔH). Based on the values of KA, the change in free energy (ΔG) and entropy (ΔS) were calculated with the equation: ΔG = —RT lnKA = ΔH—TΔS, where T is 273 K and R is 1.9872 cal/K per mol. Calorimetry trials were also carried out in the absence of AcrB using the same experimental conditions. No change in heat was observed in the injections throughout the experiment.
Statistical tests were applied as indicated in figure legends using GraphPad/Prism and/or JMP software.
|
10.1371/journal.pntd.0002250 | Replacing a Native Wolbachia with a Novel Strain Results in an Increase in Endosymbiont Load and Resistance to Dengue Virus in a Mosquito Vector | Wolbachia is a maternally transmitted endosymbiotic bacterium that is estimated to infect up to 65% of insect species. The ability of Wolbachia to both induce pathogen interference and spread into mosquito vector populations makes it possible to develop Wolbachia as a biological control agent for vector-borne disease control. Although Wolbachia induces resistance to dengue virus (DENV), filarial worms, and Plasmodium in mosquitoes, species like Aedes polynesiensis and Aedes albopictus, which carry native Wolbachia infections, are able to transmit dengue and filariasis. In a previous study, the native wPolA in Ae. polynesiensis was replaced with wAlbB from Ae. albopictus, and resulted in the generation of the transinfected “MTB” strain with low susceptibility for filarial worms. In this study, we compare the dynamics of DENV serotype 2 (DENV-2) within the wild type “APM” strain and the MTB strain of Ae. polynesiensis by measuring viral infection in the mosquito whole body, midgut, head, and saliva at different time points post infection. The results show that wAlbB can induce a strong resistance to DENV-2 in the MTB mosquito. Evidence also supports that this resistance is related to a dramatic increase in Wolbachia density in the MTB's somatic tissues, including the midgut and salivary gland. Our results suggests that replacement of a native Wolbachia with a novel infection could serve as a strategy for developing a Wolbachia-based approach to target naturally infected insects for vector-borne disease control.
| Aedes polynesiensis is a vector for both dengue and filariasis in the South Pacific. Efforts are ongoing to utilize Wolbachia as a biological control agent targeting this vector through either population suppression via releases of incompatible males or population replacement for spreading disease resistance into a population. Replacing the native Wolbachia with a novel infection from Ae. albopictus has generated the “MTB” strain of Ae. polynesiensis. This MTB mosquito is reproductively-incompatible with the wild type of Ae. polynesiensis and has a low susceptibility for filarial worms. In this work, we show that the MTB mosquito is resistant to dengue virus with a reduced viral infection in the mosquito whole body, midgut, head, and saliva. Our results further show its refractoriness to dengue virus is associated with a dramatic increase in Wolbachia density in those mosquito tissues where dengue virus needs to reside, replicate, and travel in order to be transmitted to humans. These results suggest that the MTB strain has the potential to be used in Wolbachia-based strategies to control both dengue and filariasis in the South Pacific.
| Wolbachia is a maternally transmitted endosymbiotic bacterium that infects an estimated 65% of insect species [1], [2]. The ability of Wolbachia to both induce viral interference and spread into mosquito vector populations makes it a potential candidate as a biological control agent for dengue control [3], [4], [5], [6], [7]. The primary dengue vector, Aedes aegypti, does not naturally carry Wolbachia. When artificially introduced into Ae. aegypti, three types of Wolbachia (wAlbB, wMelPop-CLA, and wMel) show significant inhibition of dengue virus (DENV) replication and dissemination, resulting in either complete or partial block of viral transmission [4], [6], [7]. Furthermore, this Wolbachia-mediated pathogen interference has a broad spectrum and can inhibit a variety of pathogens in Ae. aegypti, including Chikungunya, Plasmodium, and filarial worms [7], [8].
In addition to Ae. aegypti, dengue is transmitted by secondary mosquito vectors including Ae. albopictus and Ae. polynesiensis [9], [10], [11]. Both species of mosquitoes can transmit other pathogens to humans as well. While Ae. albopictus is a competent vector for at least 22 arboviruses [10], Ae. polynesiensis is the primary vector of Wuchereria bancrofti in the South Pacific [12]. In contrast to Ae. aegypti, both of them naturally carry Wolbachia infections. Ae. albopictus is infected with wAlbA and wAlbB [13], and Ae. polynesiensis with wPolA [14]. These Wolbachia can induce cytoplasmic incompatibility (CI) when infected males mate with females that are uninfected or carry different type of Wolbachia [14], [15]. Although native Wolbachia infections were reported to confer host resistance to pathogen infection in both the Drosophila and Culex mosquitoes [16], [17], neither wAlbA and wAlbB, nor wPolA appear to induce resistance to DENV in Ae. albopictus or Ae. polynesiensis, respectively [6], [9], [18].
Wolbachia-induced viral interference depends on the density of Wolbachia. In the original Ae. albopictus Aa23 cell line, wAlbB reaches a density high enough to completely clear dengue infection [18]. A strong negative linear correlation was observed between the genome copy numbers of wAlbB and DENV [18]. In the mosquito vector, this density-dependent viral inhibition occurs in a tissue-specific manner. A very low density of native Wolbachia in mosquito somatic tissues (e.g., midguts and salivary glands) makes its resistance to DENV undetectable [18], or detected with a weak effect [19]. When the native Wolbachia was replaced with a wMel infection, the transinfected Ae. albopictus becomes strongly resistant to DENV. This resistance is associated with a 7-fold increase in wMel density compared to the native infection [20].
Following ingestion by mosquitoes in a DENV-infected blood meal, virus first enters the midgut epithelial cells and infects the midgut tissue, in which it replicates to produce more viral particles. DENV then escapes from the midgut and disseminates through the hemolymph to other tissues, including the salivary glands, where it replicates and resides until it is injected into a human host [21]. During this journey virus needs to pass a number of physiological barriers, such as the midgut infection/escape barrier and salivary gland infection/escape barrier, and is subjected to a series of attacks by the mosquito immune system [21], [22], [23], [24]. The dynamics and tropism of DENV within a mosquito are dependent on the mosquito strain, the virus genotype, and environmental factors [25]. Recent studies show that endogenous microbial flora, including Wolbachia, also influence the mosquito's tissue- and cell-specific susceptibility to DENV by boosting basal immunity [22], [26].
Due to its importance in transmitting both Lymphatic filariasis and dengue in the South Pacific, Ae. polynesiensis was targeted for control through either population suppression via releases of incompatible males or population replacement to harness CI as a gene-drive mechanism for spreading disease resistance into a population [12], [14]. A strain of Ae. polynesiensis was previously infected with Wolbachia from Ae. riversi through interspecific hybridization and introgression. This strain was used to test a population suppression strategy in semi-field conditions with encouraging results [12]. With successful experiences in Wolbachia transinfection through embryo microinjection, additional studies were conducted to transfer Wolbachia from Ae. albopictus into the aposymbiotic strain of Ae. polynesiensis (APMT), resulting in the “MTB” strain stably infected by wAlbB only [14]. There is a strong bi-directional CI when the MTB strain crosses with the wild type “APM” strain. Furthermore, the number of successfully developing infective stage filarial worms was reduced in the MTB strain [14], which was consistent with previous findings that interaction of wAlbB with mosquito hosts can trigger a process of pathogen interference [26].
The mosquito MTB strain has a decreased ability to regulate oxidative stress [14]. We previously showed that Wolbachia can induce reactive oxygen species (ROS)-dependent activation of the Toll pathway to control DENV in Ae. aegypti [26]. Due to the role of Ae. polynesiensis in dengue transmission, it is important to know how wAlbB will influence MTB's susceptibility for DENV. In this study, we measured the vector competence of the MTB strain for DENV serotype 2 (DENV-2) as compared to the naturally infected APM strain of Ae. polynesiensis. We also compared the Wolbachia density between the MTB and APM strain. These results indicate that wAlbB induces a strong resistance to DENV in the MTB strain, which is associated with an increased Wolbachia density in mosquito somatic tissues.
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 of the Michigan State University (Application #: 03/11-040-00).
All mosquito strains used in these experiments, including the wild-type APM strain and the transfected line MTB of Ae. polynesiensis, were maintained on sugar solution at 27°C and 85% humidity with a 12-hr light/dark cycle according to standard rearing procedures. After transinfection of wAlbB into AMPT, MTB females were out-crossed with AMPT males for six consecutive generations [14]. APM and MTB were reared in a regular condition for at least two generations before used for experiments with the goal to recover and re-colonize gut bacteria [7]. Female mosquitoes, 3–5 d after eclosion, were fed on the blood of anesthetized white mice to initiate egg development. The Ae. albopictus cell line C6/36 was grown in minimal essential medium (MEM) with 10% heat-inactivated FBS, 1% L-glutamine, and 1% non-essential amino acids at 32°C and 5% CO2.
The New Guinea C strain of DENV-2 was propagated in C6/36 cells according to standard conditions [27]: In brief, 0.5-ml aliquots of virus stock were used to infect 75-cm2 flasks of C6/36 cells, at 80% confluence, with a multiplicity of infection (MOI) of 3.5 virus particles/cell. Infected cells were incubated for 7 days. Then, cells were harvested with a cell scraper and lysed by repeated freezing and thawing in dry ice and a 37°C water bath. Virus isolated from these cells was combined with the supernatant [28], resulting in virus suspension with a titer of 2×107 PFU/ml. For infection through intrathoracic injection, 69 nl of the above virus suspension was used for injection into each female. For infection through oral feeding, the resulting virus suspension was mixed 1∶1 with commercial human blood. A flask with uninfected C6/36 cells was maintained under similar conditions and used to create the noninfectious blood meal that served as our control. The blood meal was maintained at 37°C for 30 min prior to use for feeding 7-day-old mosquitoes [28].
Mosquitoes were dissected to collect the midguts in RNALater at 4, 7 and 10 days post infection (dpi) and thorax at 14 dpi, with three individual mosquitoes in a single replicate. At least five replicate biological assays were performed. Total RNA was extracted using the RNeasy kit (QIAGEN). To measure the virus titers in mosquito bodies, at 14 days after a blood meal, mosquitoes were briefly washed in 70% ethanol, then rinsed in sterile distilled water. The midgut and thorax were dissected in sterile PBS and transferred separately to microcentrifuge tubes containing 150 µl of MEM, and then homogenized with a Kontes pellet pestle motor in a sterile environment. To measure the density of Wolbachia in mosquito tissues, the midguts, salivary glands, fat bodies, and ovaries of 7-day-old non-blood-fed females were dissected and transferred separately to microcentrifuge tubes containing 50 µl of STE buffer for extraction of total genomic DNA.
To measure the number of viral genome copies, total virus RNA was extracted using the RNeasy kit (QIAGEN) and reverse-transcribed using Superscript III (Invitrogen, Carlsbad, California, USA) with random hexamers. qRT-PCR was conducted using primers targeting the dengue NS5 gene and the host RPS6 [29]. The dengue genome copy number was normalized using the RPS6 results. Two recombinant plasmids containing the targeted fragments were diluted from 101 to 108 copies/reaction and used to generate separate standard curves for NS5 and RPS6. Real-time quantitation was performed using the QuantiTect SYBR Green PCR Kit (Qiagen) and ABI Detection System ABI Prism 7000 (Applied Biosystems, Foster City, California, USA). Three independent biological replicates were assayed, and all PCR reactions were performed in triplicate. To determine the number of copies of the Wolbachia genome, real-time PCR was carried out as previously described [22], [30].
Virus titers in the tissue homogenates were measured as previously reported [28]: The virus-containing homogenates were serially diluted and inoculated into C6/36 cells in 24-well plates. After incubation for 5 days at 32°C and 5% CO2, the plates were assayed for plaque formation by peroxidase immunostaining, using mouse hyperimmune ascitic fluid (MHIAF, specific for DENV-2) and a goat anti-mouse HRP conjugate as the primary and secondary antibodies, respectively.
At 7 dpi, the viral antigen in the midguts of mosquitoes was detected by using an indirect IFA. Mosquito midguts were dissected in PBS and fixed in 4% paraformaldehyde for 5 h, and then incubated for 1 h at room temperature in a PBS-BSA-Triton solution (1× PBS, 1% Bovine Serum Albumin, and 0.1% Triton X-100), a mouse anti-dengue complex monoclonal antibody (obtained from the Centers for Disease Control, Atlanta, GA), and a fluorescein-conjugated affinity-purified secondary antibody (Millipore) were used in all midguts assays. Specimens were examined with a Zeiss (Germany) fluorescence microscope.
Mosquitoes that had been infected with DENV-2 as described above were maintained for 14 days for forced salivation assays. The assays were conducted as previously reported [31], [32]: In brief, mosquitoes were deprived of food for 24 h prior to forced salivation. The legs and wings of each mosquito were cut away, and the proboscis was inserted into 25 µl of feeding solution (50% FBS/164 mM NaCl/100 mM NaHCO3/0.2 mM ATP/≈50 µg sucrose/phenol red, pH 7.0) [32] in a 0.2-ml PCR tube. After 90 min, the mosquitoes were removed, and the feeding solution from each mosquito was sterilized by Millex-GV filter for plaque assays.
The potential use of the MTB strain to control Ae. polynesiensis in a dengue endemic area provides a rationale to test if the MTB strain is resistant to DENV as compared to the wild type of Ae. polynesiensis. To test this, we first compared DENV-2 genome copies in the whole body between the MTB strain and the wild type APM strain at 14 days post infection (dpi). Mosquitoes were infected through either oral feeding with DENV-2 infected blood or intrathoracic injection with cell medium containing DENV-2. As a result, we observed that the median number of viral genomes in the whole bodies of MTB mosquitoes was 1.4×104 times lower than that of APM mosquitoes when mosquitoes were infected through oral feeding (Mann-Whitney U test, P<0.05) (Fig. 1). Similarly, mosquito infection through intrathoracic injection led to a significantly lower viral infection in whole bodies of MTB mosquitoes as compared to APM mosquitoes (Mann-Whitney U test, P<0.001). There was a 28.3-fold reduction in the median number of viral genome copies in MTB mosquitoes comparing to APM mosquitoes (Fig. 1).
In order to determine how viral inhibition occurs in a tissue-specific manner in MTB mosquitoes, we examined whether or not viral replication in the midguts of MTB mosquitoes was suppressed. Both MTB and APM mosquitoes were fed with the DENV-2 infected blood and viral genome copy numbers in the midguts were measured at three different time points: 4, 7 and 10 dpi. The DENV-2 copy numbers were significantly lower at all of the three time points in the midguts of MTB mosquitoes than in APM mosquitoes (Mann-Whitney U test, P<0.01) (Fig. 2A). The viral infection in mosquito midguts at 7 dpi was further assayed by IFA to visualize the intensity and distribution of DENV-2 using an antibody against the viral envelop protein. Consistently, 76.2% (16/21) of APM midguts showed strong positive signals while only 10% (2/20) of MTB midguts had weak positive signals (Fisher's Exact Test, P<0.001) (Fig. 2B; Table 1). This indicates that viral replication is strongly inhibited in the midguts of MTB mosquitoes.
Following replication in the midgut, DENV-2 migrates to the mosquito head where it reaches a peak infection 14 dpi [25]. To test whether viral dissemination was affected in MTB mosquitoes, we assayed the viral infection at this time point by RT-PCR and compared the head infection rate between MTB and APM mosquitoes. As shown in Table 1, 100% (30/30) of APM heads were positive whereas viruses were only detected in 26.7% (8/30) of MTB heads (Fisher's Exact Test, P<0.001). This indicates the possibility that viral dissemination to mosquito heads is significantly reduced in MTB mosquitoes as compared with APM mosquitoes.
To determine whether viral transmission potential is reduced in MTB mosquitoes, we measured infectious viral particles in the saliva released during feeding from the proboscises of mosquitoes at 14 dpi by plaque assay. Mosquitoes were infected through either oral feeding with DENV-2 infected blood or by intrathoracic injection with cell medium containing DENV-2. When mosquitoes were infected through a bloodmeal, 21.4% (6/28) of the APM saliva samples were positive for the envelope protein of DENV-2, whereas DENV-2 was detected in 3.6% (1/28) of the MTB saliva samples (Chi-Square test, P<0.05) (Fig. 3). In the intrathoracic injection experiment, 100% (8/8) of saliva samples were positive for both mosquito strains, although the viral infection level was significantly lower in MTB than APM mosquitoes (Mann-Whitney U test, P<0.05) (Fig. 3). These results indicate MTB mosquitoes have a lower viral transmission potential than APM mosquitoes.
We previously found that Wolbachia can induce resistance to dengue infection in a Wolbachia-density dependent and a tissue-specific manner [18]. To test whether the above viral inhibition is caused by an increased Wolbachia density in the somatic tissues of MTB mosquitoes, we compared the Wolbachia density between MTB and APM mosquitoes using qPCR. We found that the Wolbachia density is significantly higher in the salivary glands (5,738-fold), fat bodies (68-fold), and midguts (269-fold) of MTB mosquitoes as compared to APM mosquitoes, while no differences were observed in the ovaries (1-fold) of APM and MTB mosquitoes (Fig. 4). These results suggest that the high density of Wolbachia in somatic tissues of MTB mosquitoes may contribute directly to their low susceptibility for DENV.
The Wolbachia strain wAlbB is able to induce resistance to DENV in both mosquito vectors and cell lines [6], [18]. After the native wPolA was replaced with wAlbB in Ae. polynesiensis, we observed that wAlbB inhibits dengue viral replication in mosquito midguts and dissemination to mosquito heads. This artificial infection also reduces viral transmission potential through mosquito bites and suppresses the viral infection level in the whole bodies of mosquitoes. Our results further show that the density of wAlbB is 269-fold and 5,738-fold higher in the midguts and salivary glands, respectively, of MTB mosquitoes than the native wPolA infection in APM mosquitoes. This supports the previous finding that the strength of Wolbachia-mediated viral inhibition depends on the Wolbachia density [18].
Although it naturally carries a Wolbachia infection, Ae. polynesiensis is a compatible vector for both DENV and Wuchereria bancrofti in South Pacific [9], [12]. This is similar to Ae. albopictus and Culex quinquefasciatus in that the presence of the native Wolbachia infection in mosquitoes does not inhibit their abilities as vectors to transmit human pathogens. Evidence supports that Wolbachia-mediated viral inhibition occurs in a tissue-specific manner with the effect dependent on the Wolbachia density in the local tissue [18]. In a host cell with a very low level of Wolbachia infection, coexistence of Wolbachia and DENV can be observed in the cytoplasm [18]. Before a mosquito can transmit DENV to humans, viruses ingested by a mosquito in a blood-meal must sequentially migrate through and replicate in the midgut and salivary glands. We previously found that the Wolbachia density in both the midgut and salivary gland of Ae. albopictus was too low to induce a resistance to DENV [6], [18]. A similar situation may be true for Ae. polynesiensis. Consistent with this predication, the Wolbachia density in the midgut and salivary gland of the wild type APM strain is hundreds of times lower than that of the resistant MTB strain.
It is interesting to note that the above difference in the Wolbachia density between APM and MTB mosquitoes occurs only in the somatic tissues, and not the reproductive tissues (ovaries). This distribution pattern could be due to the strain of Wolbachia or the history of Wolbachia-host association. Previous studies showed a Wolbachia strain-specific distribution in insect hosts [33]. The native Wolbachia in APM mosquitoes may not be able to develop an infection at as high a level as wAlbB in somatic tissues. Alternatively, the native infection may have initially been at high levels, similar to wAlbB in the somatic tissues of MTB mosquitoes, but was gradually reduced to the current level in APM mosquitoes as Wolbachia and host co-adapted during evolution. Previous studies show a long-term attenuation of wMelPop in a non-native host, Drosophila simulans [34]. If wAlbB behaves similarly, it may compromise the effectiveness of this bacterium in pest control.
Our results indicate that Wolbachia in midguts can have a significant impact in the development of mosquito resistance to DENV. DENV has to pass the midgut infection barrier with a peroral infection. When infected by an intrathoracic injection, viruses bypass the midgut infection barrier and directly disseminate through the hemolymph. We observed a 1.4×104 –fold reduction in the viral infection in the whole bodies of MTB mosquitoes as compared to those of APM mosquitoes with a peroral infection by DENV-2, while just a 28.3-fold reduction was observed for the same comparison with the viral intrathoracic injection. In addition, only 3.6% of MTB saliva was dengue-positive with the viral peroral infection, as compared to 100% of MTB saliva positive with the viral intrathoracic injection. The above results suggest that Wolbachia-induced viral interference in midguts may contribute to the majority of the virus-blocking effect in MTB mosquitoes.
The fact that Wolbachia-mediated resistance has a broad spectrum against a variety of pathogens in Ae. aegypti indicates that general killing mechanisms should be triggered by Wolbachia [7], [8], [16], [26]. Similarly, wAlbB is also found to suppress filarial worm loads in MTB mosquitoes as conducted by a parallel study in a recent publication [14]. Thus, wAlbB is the second example showing that Wolbachia can interact with mosquito hosts to inhibit both DENV and worm infection. We previously found that Wolbachia can induce ROS production, which activates the Toll pathway to control DENV in Ae. aegypti [26]. When the levels of H2O2 were compared before blood meal, there was a significantly higher level of ROS in MTB mosquitoes than in APM mosquitoes [14]. This suggests that a high density of wAlbB induces more oxidative stress in MTB mosquitoes than the native wPolA does in APM mosquitoes, consistent with our prediction that a high ROS production can contribute to the antiviral resistance [26].
Until now, wAlbB has been observed inducing resistance to DENV in the Ae. albopictus Aa23 cell line [18], Ae. aegypti [6], and Ae. polynesiensis mosquitoes. Among them, wAlbB is a native infection in Aa23 cells, but an artificial infection in Ae. aegypti and Ae. polynesiensis. It is important to note here that the native wAlbB infection can completely clear DENV in Aa23 cells [18]. This suggests that wAlbB has the ability to block DENV in a host cell when the cellular physiological environment is optimized for wAlbB to grow, such that it reaches a high infection level. Until now, both wMel and wMelPop were reported to induce complete blockage of DENV transmission in Ae. aegypti [4], [7], while wAlbB induces a strong resistance to DENV with some viruses leaking from its inhibition in Ae. aegypti [6]. Considering that experimental methods, viral strains and mosquito strains are different in these studies, a direct comparison of viral interference and fitness between wAlbB and wMel/wMelPop in the same host genetic background is needed to ascertain which Wolbachia strain is better suited for use in dengue control. From the disease control standpoint, an ideal strain of Wolbachia should be able to block pathogens in mosquitoes effectively enough to interrupt the disease transmission, and also be benign enough to the mosquito hosts that it can be persistently maintained in populations. It would be interesting to know if a perfect viral blockage is essential for disease control, and what extent of the host fitness cost associated with Wolbachia is acceptable.
Previous studies show Wolbachia-host interactions are determined by the Wolbachia strain, the host genotype, and the environment [35], [36], [37], [38]. It is still unclear why some native Wolbachia are unable to induce pathogen interference in their original hosts, although it appears related to their density. Although the artificial infections all reported inducing a resistance to pathogens [4], [6], [7], [8], this does not mean that any strain of Wolbachia, when introducing into a new mosquito host, will induce pathogen interference. We also cannot exclude the possibility that certain Wolbachia-host associations may even facilitate pathogen infection in an insect vector [39]. Those effects are determined by how a mosquito physiological system is perturbed or modified by its interaction with each specific strain of Wolbachia. The outcome becomes even more of a challenge to predict when considering that Wolbachia-host interactions may evolve over time. Despite these gaps in our knowledge, our results show replacement of native Wolbachia with a novel infection is a potentially practical way to develop pathogen resistant mosquitoes for modification of those vector populations.
Both the ability of wAlbB to induce resistance to filarial worms and DENV and the bi-directional nature of CI between wAlbB and wPolA in Ae. polynesiensis make it possible to develop MTB line as a tool to block disease transmission through population replacement. This result can be achieved by large-scale releases of MTB females. Alternatively, a strategy can be designed to start with population suppression by inundative releases of MTB males, followed by release of MTB females to initiate population replacement. Technically, the latter will be more acceptable to the public because males do not feed on human blood or transmit disease. When the population is then suppressed to a low density, the number of females that need to be released will be reduced. This would make population replacement happen parallel to a decrease in annoyance by mosquito biting. However, its deployment would require the development of a system with a high efficacy for mass rearing and sex separation, which is an ongoing effort in the field.
|
10.1371/journal.pcbi.1003119 | Target Essentiality and Centrality Characterize Drug Side Effects | To investigate factors contributing to drug side effects, we systematically examine relationships between 4,199 side effects associated with 996 drugs and their 647 human protein targets. We find that it is the number of essential targets, not the number of total targets, that determines the side effects of corresponding drugs. Furthermore, within the context of a three-dimensional interaction network with atomic-resolution interaction interfaces, we find that drugs causing more side effects are also characterized by high degree and betweenness of their targets and highly shared interaction interfaces on these targets. Our findings suggest that both essentiality and centrality of a drug target are key factors contributing to side effects and should be taken into consideration in rational drug design.
| The ultimate goal of medical research is to develop effective treatments for disease with minimal side effects. Currently, about 20% of drug candidates failed at clinical trial phases II and III due to safety issues. Therefore, understanding the determining factors of drug side effects is of paramount importance to human health and the pharmaceutical industry. Here, we present the first systematic study to uncover key factors leading to drug side effects within the framework of the human protein interactome network. Our results show that it is the number of essential targets, not the number of total targets, of a drug that determines the occurrence of its side effects. Furthermore, we find that the centrality, both degree and betweenness, of the drug targets is also an important determining factor of drug side effects. Our findings will shed light on new factors to be incorporated into the drug development pipeline.
| Regardless of their effectiveness, most drugs come with side effects of different types that affect patients' life quality and may even bring up additional health problems. It is estimated that around two million patients suffer from serious drug side effects each year and that the fourth leading cause of death in the United States is severe side effects of medication [1], [2]. Of the total number of drug candidates failed during clinical trial phases II and III, 20% of these failures are because of safety issues [3]. Hence, evaluating potential side effects of drugs is important in rational drug design and development, as well as successful marketing. Binding of drugs to their on- and off-targets modifies the functions of these targets and therefore is believed to account for their efficacies as well as side effects [4]. Traditionally, properties of a drug such as binding fingerprint and chemical structure are evaluated to anticipate side effects [5], [6]. Moreover, in vitro assays or phenotypic tests in model organisms may not be able to capture the same spectrum of side effects in human [7], [8].
Recently, an increasingly accepted view is that integrating biological networks would provide unique insights into understanding disease mechanisms and identifying novel drug targets [9], . Network-based methods have been explored and successfully applied in finding disease-associated genes and inferring underlying molecular mechanisms [11], [12]. Similarly, phenotypic responses to drugs can be better rationalized by considering their overall effects in the context of molecular networks. Previous studies have shown that drugs with shared targets or those that are close in the interactome network often share similar side effects [13], [14]. Also, similar side effect profiles have been used to predict drug-target interactions for potential drug repositioning [13]. Hase et al. examined network degree distribution of different categories of genes and suggested that connectivity is potentially important in inferring drug side effects [15]. However, no actual adverse effect data were used in their study. The relationships between drug target properties, especially in the context of biological networks, and its potential toxicity to human remains unexplored. Here, we systematically investigate major contributing factors of drug side effects, taking into consideration their direct targets and the local network structures of these targets.
We obtained a list of 996 drugs and the associated 4,199 side effects from SIDER 2 [16] and analyzed 645 FDA-approved drugs that have at least one known human protein target based on the DrugBank database [17]. Evaluation of severity of adverse effects varies among individuals and is often affected by an individual's underlying health conditions. In general, drugs that cause more side effects tend to have higher likelihood leading to severe outcomes, including death (Figure 1). Although tremendous efforts have been made on studying drug side effects in the pharmaceutical industry, the number of side effects for FDA-approved drugs significantly increases for those that were approved recently (Figure S1), indicating the necessity in further studying the contributing factors underlying drug adverse effects. By grouping drugs into the categories of “nutraceutical”, “approved”, and “withdrawn” drugs, we find that, unsurprisingly, the nutraceutical drugs have the least number of side effects (P-value = 0.00023, when compared to the approved therapeutical drugs; Figure 2A), while the withdrawn drugs cause significantly more side effects compared to the approved ones (P-value = 0.04; Figure 2A). However, there is no significant difference between the average numbers of targets of the three drug groups (Figure 2B). This indicates that the occurrence of side effects may not simply be explained by the number of targets a drug binds to. To investigate this further, we performed a generalized linear regression with negative binomial distribution for side effects over the number of targets. At first, we observed that the number of side effects significantly correlates with the number of targets (β = 0.045; P-value = 0.0033; Figure 2C). However, further dissection of properties of drug targets reveals that the positive correlation is due to the presence of essential targets, those drug targets encoded by essential genes. We find that the positive correlation between the number of side effects and that of essential targets is much more significant (β = 0.17; P-value = 1.8×10−5; Figure 2D). On the contrary, by analyzing drugs with no known essential targets, we find that the positive correlation between the number of side effects and targets no longer holds (β = 0.004; P-value = 0.93; Figure 2D; see Figure S2 for the illustration separating the effects of essential and non-essential targets). This discovery suggests that it is the number of essential targets, rather than the number of total targets, that governs the occurrence of drug side effects.
Moreover, the human interactome network has been demonstrated to be highly valuable in understanding pathogenic mechanisms of many disease genes [9], since most proteins interact with other proteins to carry out their functions [18]. Therefore, it is also important to assess drug side effects by considering network properties of their targets within the human protein interactome. Here, we examined whether the degree (number of proteins that directly interact with the targets) and betweenness (number of shortest paths going through the targets) [19] of drug targets in the network contribute to side effects. These are two of the most important network parameters, measuring the centrality of the target proteins within the network. We constructed a high-quality human protein-protein interactome network that consists of 30,713 interactions between 8,357 proteins and then mapped all the drug targets onto the interactome (Materials and Methods; the sub-network containing the drug targets is shown in Figure 3A). This high-quality human protein-protein interactome network can provide insights into potential toxicity of drugs based on the network properties of their targets.
To systematically investigate the relationship between a drug's side effects and its target degree within the interactome network, we focused on drugs with only one non-essential target to separate potential confounding effects of the number of total and essential targets. The results show that the number of side effects correlates significantly with the degree of drug targets (β = 0.31; P-value = 0.041; Figure 3B). Furthermore, we analyzed the occurrence of side effects with respect to the number of targets that are bottlenecks [19] (network nodes with betweenness among top 20%) and found significant positive correlation between them (β = 0.21; P-value = 0.0057; Figure 3C). This positive correlation is consistent when we set the betweenness cutoff at top 5%, 10%, and 40% for identifying bottleneck proteins (Figure S3). This observation indicates that the centrality of drug targets in biological networks also plays a key role in producing various side effects. We further partitioned the drugs into cancer and non-cancer drugs and repeated the calculations for essentiality and centrality that we presented above. We found the same conclusions for both cancer (Figure S4) and non-cancer drugs (Figure S5).
Our recent study has shown that reconstructing the human protein interactome into a three-dimensional (3D) structurally resolved network can provide insights into molecular mechanisms of disease genes and their mutations [12]. To understand distinct perturbations of the interactome network by various drugs, we then examined the properties of their targets within the framework of our 3D-interaction network. The structural details in this 3D-interaction network allow us to distinguish the effects of drug targets with distinct binding interfaces (i.e., multi-interface targets, which bind their different interaction partners at different interfaces) and those with a common interface (i.e., single-interface targets, which bind their different partners at the same interfaces) [20]. We hypothesize that more adverse effects are expected for a single-interface target due to a higher likelihood of altering all of its interactions by a drug disrupting its only interaction interface. By analyzing side effects of a drug with the proportion of shared interaction interfaces of each drug target with its interaction partners, we observe that the number of side effects increases significantly with the proportion of shared interaction interfaces on a target (β = 1.5; P-value = 0.00014; Figure 3D). This observation confirms our hypothesis that single-interface targets are likely to cause more side effects than multi-interface ones. We show that this finding is not due to potential biases contributed by hubs or bottlenecks since these nodes tend to have smaller proportions of shared interaction interfaces (Figure S6).
We further identified genes associated with human genetic disease and mapped them onto our human protein interactome network [12]. We calculated the average shortest distances between drug targets and disease-associated genes to represent potential molecular steps needed for a drug to affect the corresponding disease module/pathway. We find that although there is an enrichment of shorter distance between drug targets and their “indicated disease” genes, the distribution largely overlaps with that of distance between targets and unrelated disease genes (Figure 4A). Furthermore, the drugs that fail to specifically interfere with the disease-associated module/pathway result in many more side effects (Figure 4B). This result further demonstrates the importance of incorporating network properties of drug targets and corresponding disease genes in rational drug design and development.
In summary, for the first time, we show that the number of essential targets, not the number of total targets, is a determinant of drug side effects. Furthermore, high incidence of drug side effects is also characterized by high degree and betweenness of their targets in the interactome network, as well as highly shared interaction interfaces on these targets. Our findings reveal that both essentiality and centrality of a drug target are important factors to be considered in the drug development pipeline in order to improve the efficiency of this lengthy and costly process. Incorporation of these factors will be useful in the selection of drug candidates at the early stages of the drug development pipeline. When choosing from several drug candidates with similar chemical properties, the one binding to proteins that are not essential and not central in the network would have a higher chance of passing clinical trials later. Moreover, in the efforts of computationally predicting drug side effects [21], the inclusion of target essentiality and centrality as additional features would also improve the prediction performance. Furthermore, our results can serve as guidance for minimizing side effects in clinical applications, especially when prescribing multi-drug cocktails, which have been proven to be much more effective than single drug approaches [22]. With the increasing coverage of the protein-protein interaction network in human and the accessibility of interactions of high confidence levels [23], more interesting analyses can be performed to further dissect the properties of drug targets and the associated side effects. This study of adverse effects of drugs within the framework of the protein-protein interactome network demonstrates that network-based pharmacology is of great importance in the field of drug development and application.
We downloaded 4,199 side effects associated with 996 drugs from the SIDER database release 2 [16]. For the drugs in SIDER 2, we mapped them based on the generic drug names or PubChem IDs [24] to the DrugBank database [17] downloaded on November 6, 2011, and extracted all of their direct binding human protein targets (647 in total) with available uniprot IDs. We did not differentiate on- and off-targets in all of our analyses with the rationality that they could all potentially produce side effects when bound by the corresponding drugs. Furthermore, we downloaded the database containing the approval dates for each drug from the Drugs@FDA database (http://www.accessdata.fda.gov/scripts/cder/drugsatfda/) and the Orange Book (http://www.accessdata.fda.gov/scripts/cder/ob/eclink.cfm). The earliest approval date was used when a drug had a history of multiple approval events. We then cross-checked the list with the ones reported by Rask-Andersen et al. [25] and removed the drugs with conflicting dates. A list of essential genes was obtained by taking the union of the human orthologs of mouse genes that result in embryonic or postnatal lethality when disrupted [26] and the genes reported as essential from a large-scale RNAi screen in human mammary cells [27]. A drug target that belongs to the essential gene list is abbreviated as an “essential target”.
To find key factors contributing to the incidence of side effects, we performed a series of generalized linear regressions based on negative binomial distribution for side effects with the following probability density function:with mean μ and shape parameter θ. The expected value and variance for the number of side effects are:
This model is used because we observed over-dispersion with Poisson distribution, which is normally modeled for count data. The generalized linear regressions were built using the log-link function:where X is the independent variable (such as the number of targets), β is the unknown parameter, and is the linear predictor. To minimize the effects of extreme observations, we used median numbers of side effects as response variables for regression analysis. For each regression, we obtained a P-value for the effect of a tested factor based on the hypothesis testing: H0: β = 0 (there is no effect of the tested factor) vs. HA: β≠0 (the incidence of side effects is contributed by the factor). Due to the lack of data points, a few observations at the margin were binned together. We first fitted regression for the number of side effects over that of total targets and that of essential targets. To distinguish the effect of total targets and essential targets on the incidence of side effects, we repeated the regression analysis on the drugs that do not have any essential targets.
We compiled a list of human protein-protein interactions combining high-throughput high-quality yeast two-hybrid interaction datasets [28]–[31] with six major protein-protein interaction databases [32]–[37]. Since literature-curated interactions could contain low-quality interactions [38], [39], we filtered the dataset by applying the criteria that each interaction has to be either from a high-throughput high-quality experiment or supported by at least two independent publications. The interactome network contains 30,713 binary and co-complex interactions between 8,357 proteins. To evaluate network properties of drug targets, we mapped them to the high-quality protein-protein interactome network and calculated their network properties.
To reconstruct the three-dimensional (3D) structurally resolved network, we further filtered the interactions with binary evidence codes, since the concept of interaction interface does not apply when two proteins do not bind each other directly [12]. We then constructed the 3D-interaction network based on known co-crystal structures in the Protein Data Bank (PDB) [40] using a homology modeling approach as described earlier [12]. This approach has been demonstrated to be very effective and accurate in inferring protein-protein interaction interfaces [12]. The resulting structurally resolved protein interactome is composed of 6,594 interactions between 3,630 proteins.
We compiled a list of diseases for each drug based on the “indication” field from the DrugBank database. For each drug, we then obtained the disease-associated genes for these diseases from the disease-gene association map we compiled earlier based on OMIM and HGMD databases [12], [41], [42]. We then calculated the average shortest distance on the binary interactome network for 1) pairs of target proteins and the genes associated with the “indicated” diseases and 2) pairs of target proteins and all other disease-associated genes (Figure 4).
For each drug target protein T that can be mapped to the structurally resolved network with at least two interaction partners, we measured the proportion of shared interaction interfaces by calculating the Jaccard similarity coefficient [43]:where is the number of interacting domains on drug target protein T involved in both T-A and T-B interactions, and is the number of interacting domains involved in either T-A or T-B interaction. The mean of the Jaccard similarity coefficient was taken when a target protein has more than two interaction partners. To minimize potential confounding effects of essentiality, we analyzed the drugs with only one non-essential target to evaluate the effects of shared interaction interfaces of a drug target on the number of side effects.
While the vast majority of drugs have average distances between their targets and corresponding disease genes comparable to network mean distance (mean distance = 4.4), there are some drugs enriched with much smaller distances (distance<3; Figure 4A). We categorized the drugs into two classes using an average distance of 3 as cutoff to compare the median number of side effects. We carried out the bootstrapping approach to evaluate the difference of median number of side effects due to the observation of extremely unequal sample sizes (12 drugs with distance less than 3 and 319 drugs with distance equal to or bigger than 3) and variances between the two classes. For each drug class, we randomly sampled 10 observations with replacement and generated the median of these observations. The procedure was repeated 1000 times to obtain distributions of median number of side effects for each of the two drug classes. Then the Wilcoxon rank-sum test was used to evaluate the differences of median drug side effects between the two drug classes (Figure 4B). By randomizing the protein-protein interactions, the disease gene sets, and the drug target sets, we demonstrated that the observation is not due to potential biases in the data (Figure S7).
|
10.1371/journal.pgen.1002014 | Polycomb Repressive Complex 2 Controls the Embryo-to-Seedling Phase Transition | Polycomb repressive complex 2 (PRC2) is a key regulator of epigenetic states catalyzing histone H3 lysine 27 trimethylation (H3K27me3), a repressive chromatin mark. PRC2 composition is conserved from humans to plants, but the function of PRC2 during the early stage of plant life is unclear beyond the fact that it is required for the development of endosperm, a nutritive tissue that supports embryo growth. Circumventing the requirement of PRC2 in endosperm allowed us to generate viable homozygous null mutants for FERTILIZATION INDEPENDENT ENDOSPERM (FIE), which is the single Arabidopsis homolog of Extra Sex Combs, an indispensable component of Drosophila and mammalian PRC2. Here we show that H3K27me3 deposition is abolished genome-wide in fie mutants demonstrating the essential function of PRC2 in placing this mark in plants as in animals. In contrast to animals, we find that PRC2 function is not required for initial body plan formation in Arabidopsis. Rather, our results show that fie mutant seeds exhibit enhanced dormancy and germination defects, indicating a deficiency in terminating the embryonic phase. After germination, fie mutant seedlings switch to generative development that is not sustained, giving rise to neoplastic, callus-like structures. Further genome-wide studies showed that only a fraction of PRC2 targets are transcriptionally activated in fie seedlings and that this activation is accompanied in only a few cases with deposition of H3K4me3, a mark associated with gene activity and considered to act antagonistically to H3K27me3. Up-regulated PRC2 target genes were found to act at different hierarchical levels from transcriptional master regulators to a wide range of downstream targets. Collectively, our findings demonstrate that PRC2-mediated regulation represents a robust system controlling developmental phase transitions, not only from vegetative phase to flowering but also especially from embryonic phase to the seedling stage.
| Epigenetic regulation of gene expression through modifications of histone tails is fundamental for growth and development of multicellular organisms. The trimethylation of lysine 27 of histone 3 (H3K27me3) is the landmark of Polycomb Repressive Complex2 (PRC2) function and is associated with gene repression. Here we present the development of a genetic system to generate homozygous null mutants of Arabidopsis PRC2. A first major finding is that H3K27me3 is globally lost in these mutants. Surprisingly, we found that initial body plant organization and embryo development is largely independent of PRC2 action, which is in sharp contrast to embryonic lethality of PRC2 mutants in animals. However, we show here that PRC2 is required to switch from embryonic to seedling phase, and mutant seeds showed enhanced dormancy and germination defects. Indeed, many genes controlling seed maturation and dormancy are marked by H3K27me3 and are upregulated upon loss of PRC2. The invention of seed dormancy of land plants is regarded as one of the major reasons for the evolutionary success of flowering plants, and the here-discovered key role of PRC2 during the developmental phase transition from embryo to seedling growth reveals the adaptation of conserved molecular mechanisms to carry out new functions.
| One common principle of flowering plants and probably one of the main reasons for their evolutionary success is the alternation of a dormant seed stage with a growing plant that will eventually reproduce and again generate seeds. Seeds harbor not only the plant embryo, i.e. the next plant generation, but typically contain a nourishing tissue, called the endosperm that supports embryo growth and often provides the nutrients for the germinating seedling. Moreover, the embryo and the endosperm are protected by a hard shall, the seed coat, that also facilitates the distribution of seeds. Remarkably, seeds often will stay dormant after ripening and require for germination a defined order of environmental conditions reflecting the progression of the seasons in moderate climates, i.e. they will germinate only after exposure to warmth after a period of cold temperatures. Many factors have been identified to influence this transition from a dormant embryonic phase to a germinating seedling (for review see [1]). However, a unifying molecular framework has not been established so far.
For the other major phase transition in plants, e.g. from vegetative growth to flowering, it has been found that Polycomb repressive complex 2 (PRC2) regulation is crucial [2]–[4]. PRC2 activity was also found to be required for repression of flower formation in young seedlings indicating a function in maintaining and/or establishing vegetative growth [5], [6]. Moreover, severely compromising PRC2 function revealed its function in maintaining overall cell and tissue organization, e.g. the distinction between root and shoot fates [5], [6].
PRC2 catalyzes the deposition of trimethylation of Lysine 27 on histone H3 (H3K27me3), a repressive chromatin mark [7], [8]. The core PRC2 complex is conserved between animals and plants and contains at least four components, which were first identified in Drosophila: the HMTase Enhancer of Zeste (E(Z)), the WD40 domain protein Extra sex combs (ESC), the Zn-finger protein Suppressor of zeste-12 (SU(Z)12) and the nucleosome-remodeling factor 55 (NURF-55) [9]–[12]. Arabidopsis contains three presumptive H3K27me3 HMTases, CURLY LEAF (CLF), SWINGER (SWN) and MEDEA (MEA) that have been found to at least partially compensate for each other. Similarly, Drosophila Su(Z)12 function is represented by three partially redundantly acting genes, EMBRYONIC FLOWER 2 (EMF2), FERTILIZATION INDEPENDENT SEED DEVELOPMENT 2 (FIS2), and VERNALIZATION 2 (VRN2). The homolog of Drosophila ESC, FIE, is the only PRC2 component that is represented by a single member in Arabidopsis.
In the past few years, much progress has been made in the understanding of the modus operandi of PRC2. However, a major obstacle in studying the function of chromatin regulators is their essential role in early development as for instance mutants in ESC in Drosophila and its murine ortholog EED are embryonic lethal [13]–[15]. Similarly, PRC2 function is crucial already for endosperm formation in flowering plants by controlling the parent-of-origin dependent activity of a number of genes in the endosperm (imprinting). PRC2 function is maternal gametophytically required and loss of the maternal PRC2 function releases targets genes from their repression leading to endosperm overproliferation and ultimately to seed abortion [16]–[19]. This requirement for endosperm formation has also precluded so far an analysis of PRC2 action during later stages of seed development and it also remained an open question whether PRC2 function is required for initial body plan formation in flowering plants during which an embryo with shoot, root, and one (Monocotyledons) or two (Dicotyledons) cotyledons is formed. In contrast to animals, the two stem cell populations established in embryogenesis, i.e. the root and shoot meristem, will produce the body of the adult plant and it has been shown previously that PRC2 is involved in postembryonic shoot meristem function [20].
We and others have previously identified a mutant in the cell cycle regulator CDKA;1 in which the second mitosis during pollen development is missing or substantially delayed [21]–[23]. However, mutant pollen can successfully fertilize the egg cell giving rise to an embryo while triggering the onset of endosperm development without a paternal contribution. This type of fertilization was found to bypass the maternal requirement of PRC2-dependent repression during endosperm development resulting in a mutual rescue of the paternal effect of cdka;1 mutant pollen and the maternal effect caused by mutations in MEA, FIS2 or FIE [24].
Here we have used cdka;1 mutant fertilization to generate homozygous fie mutant plants allowing us to functionally address the requirement of PRC2 action during embryogenesis and subsequent plant growth and development. Our results show that PRC2 is required neither for the generation nor maintenance of embryonic organization in striking contrast to animal PRC2 function. However, PRC2 in plants is vital for the reprogramming of developmental fates mediating the switch from embryonic states to growing seedlings. Furthermore, our genome-wide ChIP- and transcriptional profiling experiments gave insights into the circuitry of PRC2 action indicating that developmental phase transitions are robustly controlled by PRC2 through simultaneously targeting genes at different hierarchical levels and triggering positive feed back loops. This network design allows the transduction of environmental cues into stable and self-maintaining developmental fates likely underlying the enormously adaptable yet enduring growth of plants.
Since the female gametophytic defect of mutants in FIS class genes can be bypassed by fertilization with cdka;1 mutant pollen [24], we asked whether this would allow the generation of homozygous fie mutant plants in crosses of heterozygous fie mutant mother plants with pollen of cdka;1-fie double heterozygous plants. Indeed, in the progeny of this cross and amid the descendents of a self-pollinated double heterozygous cdka;1-fie mutant a morphologically distinguishable class of plants was identified that was never found among the progeny of heterozygous fie or cdka;1 mutants. Subsequent genotyping confirmed that these plants were homozygous mutant for fie (Figure 1). Reciprocal crosses corroborated that the appearance of fie resulted solely from fertilization with paternal cdka;1 whereas maternal cdka;1 did not contribute to the generation of viable fie mutants (Table 1).
The fie mutant used as reference allele in this study is a T-DNA insertion line in a central exon and represents a transcriptional null mutant (Figure S1). In the same way generated homozygous seedlings for three additional fie alleles resulted in the same mutant phenotype (Figure S1 and data not shown). Thus, circumventing the requirement of FIS action in the endosperm is sufficient to generate homozygous null mutants for the PRC2 core gene FIE.
Loss of ESC function in flies or mammals causes embryo lethality and is essential for the patterning of the body plan [15], [25]. In contrast, macro- and microscopical analyses revealed that fie mutant seedlings initially showed a wild-type-like body plan with a root and a shoot, two cotyledons, and newly forming rosette leaves that were at this stage morphologically indistinguishable from wild-type sister plants (Figure 1A, 1D). However, fie mutants grew more slowly than the wild type and around 10 days after stratification (10 DAS) already initiated flower buds (Figure 1E shows a flower bud at 15 DAS) whereas the wild type started to flower only after more than 30 DAS. During the next 10 days, homozygous fie mutants developed an increasing number of ectopic cells (Figure 1K, 1L) and organs (Figure 1H, 1I), showed signs of organ transformations (Figure 1G) and generated somatic embryos (Figure 1J). The loss of spatial and temporal organization continued and homozygous fie seedlings transformed into callus-like structures that could be maintained for several months displaying an increasing number of small cells (Figure 1F). This neoplastic behavior was confirmed by flow cytometrical analyses showing that shortly after germination fie cells started to endoreplicate as a sign of differentiation, a cellular behavior typically found in maturing wild-type plants (Figure S2A, S2B, S2D) [26], whereas at three months after germination the peaks corresponding to 8C and 16C were very much reduced and the remaining cell population gave rise to a DNA profile with cells being mostly in a G1 and a G2 phase, suggesting a massively dividing cell population (Figure S2C, S2D).
Thus, PRC2 in plants does not appear to be required for initial body plan organization, indicating a major difference with animal PRC2 function. After germination homozygous fie mutants displayed a progressive loss of cell differentiation states resembling the previously characterized clf-swn double mutant or a special fie mutant allele that results from the incomplete rescue of a fie mutant with a FIE-expressing transgene [5], [6].
As H3K27me3 is essential in animal embryogenesis we asked if this mark is in fact missing in the viable fie mutants. Therefore we first analyzed by immuno-cytology the distribution of H3K27me3 in the nuclei of wild-type control plants and homozygous fie mutants (Figure S3). In two-week old wild-type plants, a clear nuclear signal that is widely dispersed along the entire chromatin was found (Figure S3A–S3C) consistent with previous studies [27], [28]. The spotted antibody signal is excluded from compacted heterochromatic regions, as visualized by DAPI staining. In contrast, no signal was observed in nuclei of two-week old fie plants (Figure S3D–S3F).
To obtain a high-resolution molecular map of the genome-wide distribution of H3K27me3 in wild type versus fie seedlings, chromatin immunoprecipitation (ChIP) was performed, followed by hybridization to a whole genome tiling array. A total of 5634 genes were identified as putative PRC2 targets in wild type seedlings, in good agreement (68% overlap) with a previous analysis (Table S1, Figure S4) [29].
In fie seedlings, the H3K27me3 signal was absent or extremely reduced throughout the genome (Figure 2). However, out of 5634 H3K27me3 positive genes in wild type, 1384 (24.6%) still passed the detection threshold in fie seedlings (Figure S4). Furthermore, 2014 genes appeared to be marked de novo by H3K27me3 in fie. Yet, in addition to being much weaker, the H3K27me3 signal in fie showed an atypical distribution pattern over genes and the marked genes were on average larger and slightly closer to transposable elements (Figure S5, Table S2). Notably, the most prominent signal in the mutant was found over heterochromatic regions, i.e. repeat-regions and transposable elements although H3K27me3 is typically excluded from these locations (Figure 2A, Figure S5B) [29]. Such an apparent re-localization of H3K27me3 signal to heterochromatic regions has also been seen on immuno-localization level in other mutants in PRC2 components [27].
To test the H3K27me3 signal found in wild type and fie tiling arrays, we performed locus-specific qPCR on our ChIP-derived DNA-material. We analyzed seven gene loci and corroborated a H3K27me3 signal in wild type and its absence in fie (Figure S6A, S6D). Moreover, we could detect in qPCR experiments only a slight increase in H3K27me3 over heterochromatic regions in fie in contrast to the array signal (Figure S6A, S6E, S6F). In any case, the signal over heterochromatic regions was much below the level of H3K27me3-positive genic regions in wild type. These findings suggest that the antibody recognizes additional epitopes besides the H3K27me3 mark, preferentially in the absence of the proper antigen. A weak signal may get artificially enhanced in ChIP-on-chip experiments due to the global amplification procedure that is not applied in gene-specific ChIP-qPCR experiments.
To investigate the specificity of the antibody, we performed peptide competition assays. Nuclear protein extracts isolated from wild type showed a strong signal in Western blots while no band corresponding to the H3K27me3 mark could be detected in homozygous fie mutants under standard conditions (Figure 3A). However, when over-exposed or under less stringent conditions a faint signal also became visible in fie (Figure 3A, 3B). Using increasing peptide concentrations of up to 10 µg of H3K27me2 and H3K27me1 peptides, a gradual decrease in signal strength was observed in the case of H3K27me2 and H3K27me1 in the fie mutant, with the mono-methylated peptide being the most effective (Figure 3C). As the signal was strongly reduced by the peptides the cross-reactivity of the antibody might account to a large extent for the remaining H3K27me3 signal in our Western and ChIP-experiments. Moreover the H3K27me3-peptide could not deplete the signal further in fie as would be expected when the mutant is already largely devoid of any H3K27me3 (Figure 3B). In contrast, the trimethylated peptide effectively reduced the signal in wild type to a level comparable to the detection level in fie whereas the H3K27me2- and H3K27me1-peptides did not show any obvious effect in wild type samples up to concentrations of 10 µg (Figure 3B, 3D). Thus, we conclude that the remaining signal in fie is not H3K27me3 but to some extent H3K27me2 and more pronounced H3K27me1 demonstrating a slight cross-reactivity of the antibody. Given that H3K27me1 is found mainly over heterochromatic regions in Arabidopsis wild-type plants [30], we conclude that a cross-reactivity of the H3K27me3 antibody with H3K27me1 accounts for the gain in signal over repeat-regions and transposable elements in the fie mutant.
To unravel the transcriptional consequences upon the loss of PRC2 function we compared genome-wide expression levels of homozygous fie mutants with wild type at two different time points. At 7 DAS, no major transformations were observed, yet the plants could be unambiguously and reproducibly identified as homozygous fie mutant plants due to their aberrant root growth and subsequent genotyping (Figure 1). At the second time point at 20 DAS, substantial morphological transformations were clearly visible.
Within our reference set (Table S3), a total of 1115 genes were significantly up-regulated at 7 DAS and 1735 genes at 20 DAS in fie versus wild-type plants (Bonferroni P-value ≤0.05, see Material and Methods). Conversely, we also found genes to be significantly weaker expressed in fie versus wild type: 1308 and 1843 genes at 7 DAS and 20 DAS, respectively (Bonferroni P-value ≤0.05; Figure S7'). Next, we compared the expression data with our PRC2 target gene set. Only a fraction of all identified PRC2 target genes became up-regulated in fie mutants, indicating that PRC2 is not the only repressive system and/or besides the revelation of the repression activators are required for gene expression (Figure S7). Still, our data are consistent with the concept that H3K27me3 mark is associated with inhibition of gene expression since the overlap of the group of up-regulated genes at 7 DAS and 20 DAS with the group of H3K27me3 marked genes is larger than expected by random (7 DAS and 20 DAS: representation factor (rf) = 7.1, p<1.0e−99 *; 7 DAS and H3K27me3: rf = 1.6, p<1.8e−21; 20 DAS and H3K27me3: rf = 1.1, p<0.009; Figure S7). Conversely, for down-regulated genes at 7 DAS we see the opposite effect, i.e. the overlap of both gene sets is smaller than expected at random (7 DAS and 20 DAS: rf = 7.5, p<1.0e−99; 7 DAS and H3K27me3: rf = 0.6, p<1.3e−13; 20 DAS and H3K27me3: rf = 0.9, p<0.122; Figure S7).
To evaluate the PRC2 targets that are up-regulated, potentially in direct response to the loss of H3K27me3, we examined which gene ontology (GO) categories are overrepresented among the up-regulated genes that lost H3K27me3 in fie mutants using the BINGO analysis software [31]. Most overrepresented GO categories in the classification system biological function relate to reproduction with two distinct subcategories: Flower- and seed development (Figure 4, Figure S8). Besides reproduction, a few additional small categories were overrepresented such as abscisic acid (ABA) signaling and lipid-transport and –sequestering. However, a closer analysis of the corresponding genes revealed that they are also linked with reproduction, in particular seed development (see below).
H3K27me3 appears to be a key repressive mechanism for the expression of many genes controlling different aspects of flower development and consistent with this, homozygous fie mutants are very early flowering, i.e. as early as 10 DAS and produce ectopical flowers, e.g. on roots. A similar early flowering phenotype has been found in mutants with compromised PRC2 activity [5], [6], [32], and has been related to the early deregulation of LEAFY (LFY), AGAMOUS (AG) and PISTILLATA (PI), which starts as early as the embryonic stage. Our analysis identifies several additional genes controlling flower development as PRC2 targets that are significantly up-regulated in fie mutants, including genes involved in the establishment of a floral meristem (e.g. FLC, AGL24, LFY, FUL and CAL), genes involved in promoting a determinate floral meristem (e.g. ULT1, PAN, LFY) and genes involved in organ identity specification (e.g. AP3, SEP3, LFY, PI) (Figure S9). The results of our microarray experiments could be validated by qRT PCRs confirming the significant up-regulation of 5 genes (AP3, CRC, FLC, PI, SEP3). In addition, 2 genes that were only slightly (but not significantly) up-regulated in our microarray experiment were also found to have significantly elevated expression levels in the qRT PCR on fie mutant material (AG, AP1), whereas the flowering regulator FWA shows neither upregulation in the array nor in qRT-PCR experiments (Table S4).
The second principal category of PRC2 target genes that became up-regulated in fie mutants are genes functioning in late seed development (Figure 4, Figure S8, Figure S10). Among the PRC2 targets that are up-regulated in fie we find genes acting at different hierarchical levels in late seed development, from master regulators (e.g. AGL15, LEC2, ABI3, FUS3) and more specific modulators (e.g. WRI, FLC) over genes promoting ABA and/or inhibiting GA signaling (e.g. ABI4, DOG1, CHO1, SOM, SPL8) down to target genes such as storage compounds (e.g. CRU3, CRA1, LEAs, oleosins) (Figure 5).
The up-regulation of many important seed regulatory genes raised the hypothesis that fie seedlings, albeit macroscopically resembling wild type seedlings, display seed phase characteristics. To test this, we first analyzed the lipid content using the dye Fat Red that stains for triacylglycerol-lipids in red color. Whereas wild-type seedlings displayed a sharply decreasing lipid content from 5 to 8 DAS, fie mutants showed an intense red color indicating a high lipid content that is typical for late seed maturation stages in wild type (Figure 1M, 1N).
To test whether the failure to repress late seed genes during the seed maturation process interferes with germination, we performed seed germination assays of clf-swn and fie mutants in comparison with wild-type plants. Whereas all wild-type seeds germinated within 2 DAS, both clf-swn and fie mutants show delayed germination (Figure 6A). Eventually, over 90% of clf-swn mutants germinated around 4 DAS. In contrast, approximately 40 percent of the homozygous fie mutants stayed dormant for the course of the entire experiment (20 days), as revealed by dissecting dormant seeds and genotyping of the embryos.
Dissected dormant embryos were then allowed to develop on agar plates. As a reference wild-type embryos were isolated from seeds 24h after imbibition. Initially, dormant fie embryos are indistinguishable from wild type embryos (Figure 6C, 6F) and around 1/4 of these fie mutants started to develop in a similar manner as wild type, showing root- and root hair formation, unfolding and greening of the cotyledons and the accumulation of anthocyanin (Figure 6C–6H). However the remaining 3/4 of fie embryos stayed dormant for a period of several days. Some of these finally could break dormancy and started to develop although proliferation was extremely delayed (Figure 6L–6N). Notably, heterozygous cdka;1 mutants behaved like wild type seeds consistent with the previous finding that cdka;1 mutants are sporophytically recessive. Similarly, double heterozygous cdka;1-fie mutants also did not show any germination defects demonstrating that the observed dormancy phenotypes are due to the loss of PRC2 function.
Germination is associated with a low ABA to gibberellic acid (GA) ratio [1]. Intriguingly, the development of some of the dissected, initially dormant fie seedlings resembled the development of wild–type plants that germinated on high concentration of ABA, lacking proper greening of aerial tissue, root formation and expansion of true leaves (Figure 7I–7J, 7L–7M). However, applying high dosage of GA did not lead to higher germination rates of fie mutants (Figure 7B), suggesting that the primary defect in the class of non-germinating fie seeds is dormancy release and not the germination itself.
Among the up-regulated genes in fie controlling seed and flower development certain gene families appeared to be overrepresented, e.g. transcription factors, consistent with previous studies showing that those are in particular marked by H3K27me3 (Table 2) [7], [29], [33]. To get a more detailed picture, we tested whether all transcription factor families are equally subject to regulation by PRC2 (Figure S11A, S11B). Approximately 2/3 of all transcription factor families have members that are marked by H3K27me3 at 20 DAS. Notably, one of the largest transcription factor families within our reference set in which none of the member was found to carry H3K27me3 was the group of AUXIN RESPONSIVE FACTORS (ARFs) that mediate auxin signaling (Table S3). However, at a more general level, we found that other genes involved in the auxin signal transduction network are targets of PRC2 regulation, for instance several IAAs and PIN auxin transport facilitators (Table S1).
Among transcription factor families that are marked with H3K27me3, the fraction of PRC2 targets varies substantially. A particular high proportion of PRC2 targets (≥60%) were found in MIKC subclass of MADS transcription factors, in the WOX-class, the HD-Zip-IV Homeobox class and in the C2C2-Dof and C2C2-YABBY zinc finger classes, for the latter even all 6 members were found to be PRC2 targets. The transcription factor subfamily with the most (in absolute numbers as well as in percentages) PRC2 targets that also showed transcriptional up-regulation in fie is the MIKC class, among which we find central regulators of seed and flower development (Figure S11, Table S1, Table S5) [34].
In addition to transcription factors, a few other gene families were overrepresented among the PRC2 targets that are up-regulated in fie; the most prominent are oleosins and LATE EMBRYONIC ABUNDANT PROTEIN genes (LEAs). Oleosins are structural components of oil bodies and were found to be expressed preferentially in seeds or the tapetum layer of developing anthers [35] (Figure 5, Table S5). 11 of the 17 oleosin genes in our reference set are H3K27me3-marked and 8 are in addition up-regulated in fie (Table 2), which matches the observation of storage lipid accumulation in fie (Figure 1M, 1N). This is a strong overrepresentation since from all genes in our reference set, we find not more than 21% marked by H3K27me3 and only 2% being up-regulated in fie as well.
Another gene family that is highly overrepresented in the class of up-regulated PRC2-targets are LEAs, most of which are expressed in seeds. Of 54 LEAs covered by our reference set, we find 27 (50%) H3K27me3-marked and 16 (30%) being in addition up-regulated in fie. Interestingly, we found 7 (13%) of the LEAs down-regulated in fie and with a single exception these are not H3K27me3 targets and show a non-seed specific expression (Table S5, Figure S10) [36], [37].
In Drosophila the function of the Polycomb complex Group (PcG) is counteracted by the action of the trithorax Group (trxG) [7]. In Arabidopsis, the role of TRX has been assigned to ATX1, ATXR7/SET DOMAIN GROUP25 (SDG25), PICKLE (PKL)/PICKLE-RELATED 2 (PKR2) and ULTRAPETALA 1 (ULT1) [38]–[42]. Our data showed that ULT1, ULT2 and PKR2 are PRC2 targets and at least ULT1 and ULT2 were substantially up-regulated at 7 and 20 DAS (Table S3). ULT1 has been shown to act as an anti-repressor (i.e. limiting H3K27me3 deposition) and as an activator of the flower regulator AG by mediating Lysine 4 H3 trimethylation [42]. To test whether the loss of H3K27me3 is accompanied with a gain in H3K4me3, as suggested by our finding of a possible negative feed-back of PRC2 on ULT1/ULT2 activation, we analyzed the genome-wide distribution of H3K4me3 in wild type and fie.
Consistent with previous studies [43], we found that in wild-type plants a large number of genes (approximately 1/3 of the genome) are marked with H3K4me3 at 20 DAS (Figure 2, Figure 7). However, the number of genes that are marked by both H3K27me3 and H3K4me3 is significantly smaller than expected for an independent distribution, as was observed previously [43] (Figure 7). This indicates repulsion of these two marks in accordance with the model that H3K27me3 and H3K4me3 signifies repressed and activated genes, respectively. None-the-less, a small set of 501 genes was identified as marked by both histone modifications.
In our ChIP-chip experiments we found only a slight increase in number of genes that carry the H3K4me3 mark in fie in comparison to wild-type plants (13945 vs. 13211 genes) and we do not see an elevated level of H3K4me3 in Western blot analyses (Figure S6C). Thus, genes that loose H3K27me3 do not in general gain H3K4me3 in fie (Figure 7, Table S3). On the other hand, gene up-regulation in fie is positively correlated with loss of H3K27me3 and gain in H3K4me3. The global proportion of genes with elevated expression in fie at 7 DAS is 4.5% while the percentage of up-regulated genes reaches 28% amongst those that loose H3K27me3 and concomitantly gain H3K4me3 (Table 3). In addition, for certain gene families, such as the MIKC group of MADS transcription factors, the WOX group of Homeobox genes and the oleosins, we find those genes that gain H3K4me3 in addition to the loss of H3K27me3 to be among the most highly up-regulated for these specific classes (Table 3, Table S5).
Thus, our data supports the view that H3K4me3 and H3K27me3 are mutually exclusive marks though in general a loss in H3K27me3 is not sufficient to gain H3K4me3. However, a tightly linked interdependency between both antagonistic marks is operating for a relatively small group of genes including members of the MIKC class of major developmental regulators [42], [44], [45]. Our finding that the PRC2-antagonizing TRX-function genes ULT1/ULT2 are themselves targets of PRC2 and consequently up-regulated in fie, provides a possibility for the molecular implementation of such an interconnected control mechanism.
Here we have generated homozygous fie mutant plants overcoming a block in the analysis of PRC2 activity in the flowering plant Arabidopsis. Our approach relies on bypassing of double fertilization and circumventing the requirement for FIE during endosperm development. This has allowed us to study the genomic and developmental consequences of the complete loss of PRC2 activity during embryo and subsequent sporophyte development.
Several lines of evidence indicate that PRC2 in plants is indeed essential for depositing H3K27me3 marks similar to its function in animals. First, our ChIP-on-chip experiments showed that there is no or only a very weak H3K27me3 signal in fie and that the remaining signal shows properties that differ from the typical H3K27me3 mark. Second, at least 3 heterochromatic regions that showed H3K27me3 signal in fie in ChIP-on-chip experiments did not show a substantial level of enrichment in gene specific ChIP-qPCR assays. Third, the little residual signal of H3K27me3 in fie mutants can be further reduced in peptide competition assays with peptides that harbor H3K27me2 or H3K27me1 epitopes. Finally, H3K27me3 peptide is not effective in reducing the signal in fie further, as would be expected when the remaining signal were H3K27me3. Conversely, the H3K27me3 peptide could reduce the antibody signal in wild type to the signal strength found in fie.
Based on the by large mutually exclusive distribution of H3K27me3 and H3K4me3, we asked if genes which lost H3K27me3 in fie would in turn acquire H3K4me3. Such a reciprocal regulation has been found for AG and FLC in Arabidopsis [42], [44], [45]. Indeed, we could confirm that AG and FLC, both members of the MIKC transcription factor class, gain H3K4me3 in the absence of FIE. Moreover, 7 other MIKC transcription factors that represent important regulators during development showed a similar response. It was recently shown that the SAND domain protein ULTRAPETALA1 (ULT1) mediates the switch from H3K27me3 to H3K4me3 at the AG locus [42]. Interestingly, we found that ULT1 itself is under the control of PRC2, as it is marked by H3K27me3 in wild type and shows strong up-regulation in fie. This might explain the switch from H3K27me3 to H3K4me3 as seen for a remarkable number of the MIKC transcription factors. In animals the maintenance of trimethylated H3K4 was shown to require permanent TRX activity to counteract PRC2, as the repressive H3K27me3 state seems to be the default state for genes that are regulated by both antagonizing HMTase machineries [9]. However, global changes in H3K4me3 levels were not observed in fie, and the change from H3K27me3 to H3K4me3 was restricted to about 5.5% of the genes marked by H3K27me3.
We also identified a small group of potentially bivalently labeled genes (1.8% of reference set), i.e. harboring the activating H3K4me3 and the repressive H3K27me3 mark. The concomitance of both tags is found in more than 10% of all genes in human and mouse embryonic stem cells and Xenopus tropicalis embryos, and is thought to maintain the target genes in a “poised state”, resulting in transcriptional silencing but allowing for fast reactivation upon commitment to differentiation [46]–[50]. Bivalency has to our knowledge only been found for the AG and FT loci in Arabidopsis and its existence is also unclear for Drosophila [44], [51], [52]. However, we showed here that in contrast to Drosophila and mouse, Arabidopsis does not require the PRC2 to establish a normal body organization (see below). This renders it unlikely that animals and plants are using the same epigenetic mechanisms to set up the body plan, at least during embryogenesis.
The observation that the plant body plan can be established without PRC2 function is an unexpected result not only because PRC2 function is essential in animal embryogenesis but also regarding the strong postembryonic phenotype of clf-swn double mutants or fie mutants with a partially complementing FIE transgene [5], [6]. The overall correct body plan of fie embryos and early seedlings suggests that there is a tight network of intercellular communication presumably maintaining positional cues in the plant embryo. Indeed, research in the last decade has unraveled several patterning mechanisms in the plant embryo, for instance polar auxin distribution and non-cell-autonomously acting transcription factors [53], [54].
However, after body-plan formation, PRC2 function is key for the correct phase transition from embryonic to vegetative growth. Much progress has been made in the understanding of chromatin regulation and in particular the function of PRC2 during the phase transition from vegetative growth to flowering (for review see [55]–[57]). In contrast, the view that chromatin regulation is important for controlling the switch from mature seed to seedling is only now emerging (for review see [58], [59]), and the involvement of PRC2 in late seed development has been unclear beyond the finding that many genes implicated in seed maturation are labeled by H3K27me3 (for review see [58], [60], [61]). Defining the role of PRC2 during seed maturation has been obscured due to a prominent function of PRC2 earlier in seed development, i.e. for endosperm growth and differentiation. The combination of our genetics and genomics studies demonstrate that PRC2 is one of the major control systems of this phase change by shutting down the entire cascade of maturation genes from master regulators to a wide range of downstream targets before or at germination (see Figure 5). Moreover, our data suggest that the PRC2 mediated phase transition from seed- to seedling stage takes place primarily at the level of the embryo, as seeds with homozygous fie mutant endosperm but heterozygous mutant embryo germinate like wild type.
A wealth of genetic and physiological experiments has demonstrated that seed development is under the tight control of plant hormones and that GA triggers while ABA inhibits seed germination (for review see [1]). High ABA and low GA levels are characteristic for maturing seeds allowing the establishment of seed dormancy, while this relationship is inverted at germination. PRC2 action in the maturing seed seems to sustain the antagonistic action of the two plant hormones ABA and GA by inhibiting positive regulators in ABA and negative players in GA signaling. For example, the PRC2 target SOMNUS (SOM), a CCCH-type zinc finger protein, down-regulates GA and up-regulates ABA levels. SOM expression is seed specific and our finding that it is a PRC2 target and up-regulated in fie suggests that in wild type down-regulation of SOM is maintained by H3K27me3 to allow for high GA and low ABA levels in the germinating seed.
Besides the regulation of the ABA-GA signaling pathway, we also found that DELAY OF GERMINATION 1 (DOG1), a major regulator of seed dormancy [62], is a H3K27me3 target and significantly up-regulated in fie seedlings. Interestingly, it was recently shown that DOG1 is also regulated by HISTONE MONOUBIQUITINATION1 (HUB1), a C3HC4 RING finger protein required for histone H2B monoubiquitination [63].
In this context it will be interesting to examine if dormancy control is generally regulated at the level of chromatin, as for example different Arabidopsis accessions from diverse environmental origins show dramatic differences in seed dormancy [64]. Since PRC2 function in the perception of cold via the repression of the flowering inhibitor FLC is well established for the transition to the generative phase [57], it is tempting to speculate that a similar mechanism functions to perceive this environmental stimulus in the seed. The need of cold stratification in order to break seed dormancy in many plant species [65] and the observation that FLC plays a role in this process as well [66], might reveal a common regulatory mechanism operating in the transition from vegetative to generative phase as well as from embryonic to vegetative phase.
Interestingly another phase transition, the switch from gametophytic to sporophytic development was recently shown to be regulated by PRC2 in moss, where PRC2 represses the differentiation of the sporophyte [67], [68]. The authors correlate their observations with the transition from gametophytic to sporophytic development in flowering plants that is as well controlled by PRC2, as Arabidopsis fie mutants for example show untimely development of the gametophytic endosperm without fertilization [22], [67], [69], [70].
The reprogramming of gene activity is a mandatory step to allow for cellular differentiation processes and the stable inheritance of these differentiation states needs to be maintained for the integrity of the organism. Plants in particular have to adapt to environmental conditions and therefore need to change their developmental phase accordingly, and the phase transition from embryo to seedling stage can be considered as the earliest adaptive phase in plants. The origin of seed dormancy in land plant evolution is regarded as a major step in the successful establishment of flowering plants to sustain in non-favorable conditions and its control by PRC2 is an exciting example for the recruitment of an evolutionary conserved molecular machinery to fulfill new functions.
Unless indicated otherwise, Col-0 was used as wild type for all experiments. The cdka;1-1 (AT3G48750) allele has been previously described (SALK_106809 [22]). The SALK_042962 line was used as the standard allele for fie. As additional fie alleles the T-DNA lines GK-362D08-016994 and GK-534F01-020364 were used that both displayed the previously described typical fie mutant phenotype. One previously described fie allele in WS-0 [71] was sequenced and shown to carry a base-pair exchange mutation 5′ of the fourth splice acceptor site. The curly leaf (clf-28, SALK_139371, At2g23380) and swinger (swn-4, SALK_109121, At4g02020) alleles have been previously described [72], [73]. All seeds were sterilized using chloride gas and sown on 0.8% Phyto agar plates (½ Murashige & Skoog (MS) salts and 1% sucrose) and grown under day neutral conditions (12h light 21°C, 12 h dark 17°C). After germination, plants were transferred to either new plates for long-term observation or to soil and grown in long day conditions (16 h, 22°C light; 8 h, 18 C°dark).
To examine germination, seeds from plants that were grown under the same growth conditions and stored for at least 3 months were sterilized with Chloride-gas and stratified for at least 4 days at 4°C. Upon germination induction (light, 21°C), germination rate was monitored for the following 6 days. After approximately 14 days the plants were analyzed with respect to their phenotype to distinguish between mutants and phenotypically wild-type plants and correlated with the day of germination. Dormant seeds were dissected under a stereomicroscope using a fine needle and fine forceps and subsequently genotyped by PCR. Gibberellic acid (GA, gibberellin A3, Sigma) was dissolved in Ethanol (10 mM stock solution) and applied to the MS-plates in concentrations from 0.01 µM to 10 µM. The germination rate of fie mutants on GA-plates was analyzed 10–14 days after stratification (DAS). Abscisic acid (ABA, Sigma) was dissolved in Methanol (stock concentration 10 mM) and used in final concentration of 1 µM. Wild type germination on ABA-plates was monitored over time.
Plants were first partially dehydrated in three steps (20%, 40%, 60% isopropanole solution), then incubated for 1 hour with Fat Red solution (filtered 0.5% Sudan III in 60% isopropanole) and re-hydrated again using the same dilution series in reversed direction. Subsequently samples were additionally washed twice with water and analyzed under a dissecting microscope.
RNA extraction was performed using Qiagen-RNAesy mini-kit, following the manufacturers instruction. RNA-concentration and purity was tested using nanodrop-photometric quantification (Thermo Scientific). RNA-integrity was verified by running 1 ug of total RNA on 1.5% agarose TBE-gels to detect the 28S and 16S rRNA bands. 2 µg RNA was treated with DNAseI (MBI Fermentas) according to the manual to avoid contamination of genomic DNA and subsequently processed to obtain cDNA using polyT-primer and reverse transcriptase (Superscript III, Invitrogen) following the manufacturers instruction. After reverse transcription RNA was removed by RNAseH digest. For negative control, all steps were followed in the same manner, except for adding the reverse transcriptase. The resulting cDNA was used for Reverse Transcription(RT)-PCR or quantitative Real Time-PCR (qRT-PCR) using the Roche LightCycler 480 system. Oligonucleotides were designed using either Primer3Plus-design tool (http://www.bioinformatics.nl/cgi-bin/primer3plus/primer3plus.cgi) or QuantPrime (qPCR primer design tool: http://www.quantprime.de/main, [74] and used in final concentration of 0.25 µM each. Primers for qPCR have been tested for efficiency of >90% and are listed in Table S6. For qPCR at least two biological and three technical replicates were processed and expression was calculated relative to ACT7 (AT5G09810). Several reference genes were tested in comparison between mutant and wild type samples to confirm equal loading (see Table S6).
Microarray analysis was carried out at the Unité de Recherche en Génomique Végétale (Evry, France), using the CATMA arrays [75], [76]. Two independent biological replicates were produced. For each biological repetition and each time point, RNA samples were obtained by pooling RNA from 100 wild-type and 100 fie seedlings at stage 7 DAS and 10 wild type and 50 fie plants at 20 DAS, respectively. 7 DAS stage plants were cultivated on plates, 20 DAS material was grown on plates for 10 days, then transferred to liquid media for another 10 days in day neutral conditions (12 h light, 21°C; 12 h dark. 17°C). Total RNA was extracted using Qiagen RNeasy plant mini kit according to the supplier's instructions. The hybridization to the slides, and the scanning were performed as described in Lurin et al. (2004) [77].
Normalization and statistical analysis were based on two dye swaps (i.e. four arrays, each containing 24,576 GSTs and 384 controls) as described in Gagnot et al. (2007) [78]. The raw P-values were adjusted by the Bonferroni method, which controls the Family Wise Error Rate, (with a type I error equal to 5%) in order to keep a strong control of the false positives in a multiple-comparison context [79]. We considered as being differentially expressed the genes with a Bonferroni P-value ≤0.05, as described in Gagnot et al (2007) [78].
Microarray data from this article were deposited at Gene Expression Omnibus (http://www.ncbi.nlm.nih.gov/geo/), accession no. GSE19851, direct link: http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?token=djcjpeggkgsuirw&acc=GSE19851) and at CATdb (http://urgv.evry.inra.fr/CATdb/; Project: RS08-09_FIE) according to the “Minimum Information About a Microarray Experiment” standards.
Nuclear enriched protein extracts were prepared after thoroughly grinding in liquid nitrogen of around 1 g of plant material. All subsequent steps are carried out in the cold. The powder was dissolved in 10 ml of Lysis buffer (45 ml Low Salt Wash buffer [see below] + 0.5 ml TritonX-100 + 5 ml glycerol + 50 µl 100 mM PMSF + 20 µl β-mercaptoethanol freshly prepared on ice), vortexed and placed on a rotation wheel for 20 min at 4C. The solution was filtered using 100 µm nylon mesh and centrifuged for 20 min at 4000 rpm at 4°C, following resuspension of the pellet in 2 ml lysis buffer. The solution was transferred to a new 2 ml tube and centrifuged for 20 min, 4000 rpm, 4°C. The resulting pellet was resuspended in 200 µl 1XSDS loading buffer. Low Salt Wash buffer: 20 ml 0.5 M HEPES pH 7.5 + 6 ml 5 M NaCl + 400 µl 500 mM EDTA + H2O up to 200 ml. 15% SDS-gel page was performed according to standard protocols. After SDS page proteins were blotted on Hybond-P PVDF membrane (Amersham Biosciences) for 75 min, 140 mA in temperature controlled condition. All membrane manipulation experiments where carried out at room temperature (RT) when not stated otherwise. Membrane was blocked using incubation with 4XBlockAce (ABD Serotec) for 3 h. Throughout all experiments we used Anti-H3 1∶20,000 (Millipore, reference nr: 06-755) as loading control, Anti-trimethyl-Histone H3 (Lys27) antibody (Millipore, reference-nr: 07-449) in final concentration between 1∶10,000 and 1∶50,000, dissolved in 5%BSA in 1xTBST (1x TBS with 0.1% TritonX-100) and Anti-trimethyl-Histone H3 (Lys4) antibody (Millipore, reference nr: 07-473) at 1∶5,000–1∶10,000. The primary antibody was incubated at 4°C over night. After washing 3 times 15 min with 1xTBST the secondary antibody (Anti-Rabbit antibody, GE-Healthcare, reference-nr: NA934-100UL) was applied at 1∶50,000 in 5%BSA-1xTBST solution for 2 h. Washing was either performed stringently with 3x 30 min or less stringently 3x10 min. For detection the two-component reagent Immobilion Western Chemiluminiscent HRP substrate (Millipore) was used. For peptide competition, first the sub-saturating antibody concentration was determined. For anti-H3K27me3 this was at a concentration of 1∶50,000. Then increasing concentrations from 0.1–10 ug of H3K27me3, H3K27me2 and H3K27me1 peptide (Millipore 12-565, 12-566, 12-567) were added to a 10 ml antibody-solution and incubated under slight agitation for 4 h at RT and an additional 1 h at 4°C before hybridizing on the membrane. Subsequent hybridization and detection were performed as described above.
Chromatin immunoprecipitation (ChIP) experiments were done as described previously [74], in two biological replicates, using the following antibodies: H3K4me3, Millipore 07-473; H3K27me3, Millipore 07-449. DNA recovered after ChIP and directly from input chromatin was amplified using the Sigma GenomePlex Complete Whole Genome Amplification (WGA) Kit as directed, differentially labeled and hybridized in dye-swap experiments to a custom-made Roche-NimbleGen whole-genome tiling microarray. This microarray covers the entire forward strand of the Arabidopsis genome sequence (TAIR8) at 175 nt resolution with approx. 720K isothermal tiles (50–75 oligonucleotides). Following ANOVA normalization, raw data were analyzed using a linear regression mixture model (ChIPmix, [80]), which was adapted to handle multiple replicates simultaneously (script available on request). Lists of tiles reporting significant enrichment were converted in sets of chromatin domains by combining adjacent enriched tiles, allowing a maximal gap of 165 nucleotides. Only domains of at least 400 nucleotides were considered for further analysis. TAIR8 release was used for annotation of genes and transposable elements.
Several loci were additionally tested for H3K27me3 and H3K4me3 enrichment as compared to input. Input was diluted 1∶100 prior to qPCR application. From the diluted input material and from the ChIP-material 1 µl was applied for each triple replicate reaction in the Roche Lightcycler 480 Real Time System using Roche SYBR green reagent according to the supplier's instruction (Roche). Primers used for this assay are given in Table S6.
ChIP on chip data from this article were deposited at Gene Expression Omnibus (http://www.ncbi.nlm.nih.gov/geo/), accession no. GSE24163, direct link: (GSE24163, http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE24163).
10–14 DAS wild type and fie seedlings were fixed and processed as described previously [28], washed 3×5 min in 1× PBS before pre-incubation with BSA. Diluted rabbit polyclonal α- trimethyl H3K27 primary antibody (1∶100, Millipore 07-449) was incubated for 1 h at 37°C, washed 3×5 min in 1× PBS and incubated with diluted Alexa Fluor 488 conjugated goat anti-rabbit polyclonal secondary antibody (1∶200, Invitrogen (Molecular Probes) A-11008) for 1 h at 37°C, washed 3×5 min in 1× PBS and mounted in 1× PBS containing 1 µg/µl DAPI. Images were acquired using an Axioplan 2 Carl Zeiss Microscope with a cooled AxioCam HRc camera using a bandpass 515–565 nm emission filter (Carl Zeiss # 488010-9901-000) and a longpass 397 nm emission filter (Carl Zeiss # 488001-9901-000) for visualization of AF488 and DAPI, respectively. Fixed exposure settings for both florochromes were: AF488, 196 ms, 402 ms and 1002 ms (overexposure); DAPI, 50 ms, 100 ms and 305 ms (overexposure).
For all analyses comparing array expression and ChIP chip data a reference gene set of 24901 genes was defined that included those genes for which data from both type of experiments were available (Table S2). The Transcription factor classification was taken form the Arabidopsis transcription factor database (AtTFDB) hosted on the Arabidopsis Gene Regulatory Information Server (AGRIS, http://arabidopsis.med.ohio-state.edu/AtTFDB). Venn diagrams were generated using the VENN diagram generator designed by Tim Hulsen at http://www.venndiagram.tk/ and http://www.cmbi.ru.nl/cdd/biovenn/ (BioVenn [81]). The test for statistical significance of the overlap between two groups of genes was calculated by using software provided by Jim Lund accessible at http://elegans.uky.edu/MA/progs/overlap_stats.html. A representation factor (rf) is given and the probability (p) of finding an overlap of x genes is calculated using a hypergeometric probability formula. When p was below the calculation limits of the software (highly significant) we noted p<1.0e−99*. The representation factor is the number of overlapping genes divided by the expected number of overlapping genes drawn from two independent groups. A representation factor >1 indicates more overlap than expected of two independent groups, a representation factor <1 indicates less overlap than expected, and a representation factor of 1 indicates that the two groups by the number of genes expected for independent groups of genes. To determine which Gene Ontology (GO) categories are statistically overrepresented among the H3K27me3 targets that are up-regulated in fie, we used the BINGO 2.3 plugin for Cytoscape (http://www.psb.ugent.be/cbd/papers/BiNGO/Home.html). A custom annotation file was created using the build in annotation file for GO biological process and our reference set of 24901 genes. Other than that default parameters were used.
|
10.1371/journal.ppat.1003883 | The Arabidopsis miR472-RDR6 Silencing Pathway Modulates PAMP- and Effector-Triggered Immunity through the Post-transcriptional Control of Disease Resistance Genes | RNA-DEPENDENT RNA POLYMERASE 6 (RDR6) is a key RNA silencing factor initially characterized in transgene silencing and virus resistance. This enzyme also contributes to the biosynthesis of endogenous short interfering RNAs (siRNAs) from non-coding RNAs, transposable elements and protein-coding transcripts. One class of protein-coding transcripts that have recently emerged as major sources of RDR6-dependent siRNAs are nucleotide-binding leucine-rich repeat (NB-LRR) proteins, a family of immune-receptors that perceive specific pathogen effector proteins and mount Effector-Triggered Immunity (ETI). Nevertheless, the dynamic post-transcriptional control of NB-LRR transcripts during the plant immune response and the functional relevance of NB-LRRs in signaling events triggered by Pathogen-Associated Molecular Patterns (PAMPs) remain elusive. Here, we show that PTI is constitutive and sensitized in the Arabidopsis rdr6 loss-of-function mutant, implicating RDR6 as a novel negative regulator of PTI. Accordingly, rdr6 mutant exhibits enhanced basal resistance towards a virulent Pseudomonas syringae strain. We further provide evidence that dozens of CC-NB-LRRs (CNLs), including the functionally characterized RPS5 gene, are post-transcriptionally controlled by RDR6 both constitutively and during PTI. These CNL transcripts are also regulated by the Arabidopsis microRNA miR472 and knock-down of this miRNA recapitulates the PTI and basal resistance phenotypes observed in the rdr6 mutant background. Furthermore, both miR472 and rdr6 mutants were more resistant to Pto DC3000 expressing AvrPphB, a bacterial effector recognized by the disease resistance protein RPS5, whereas transgenic plants overexpressing miR472 were more susceptible to this bacterial strain. Finally, we show that the enhanced basal and RPS5-mediated resistance phenotypes observed in the rdr6 mutant are dependent on the proper chaperoning of NB-LRR proteins, and might therefore be due to the enhanced accumulation of CNL proteins whose cognate mRNAs are no longer controlled by RDR6-dependent siRNAs. Altogether, this study supports a model whereby the miR472- and RDR6-mediated silencing pathway represents a key regulatory checkpoint modulating both PTI and ETI responses through the post-transcriptional control of disease resistance genes.
| Virus resistance relies in some plant-viral interactions on the RNA-DEPENDANT RNA POLYMERASE 6 (RDR6), a major actor of RNA silencing that acts at the post-transcriptional level. Here, we demonstrate that RDR6 also plays a role in basal defense and race-specific resistance. RDR6 and the microRNA miR472, which targets the mRNAs of disease resistance genes of coiled-coil nucleotide-binding leucine-rich-repeats family (e.g. RPS5), act in cooperation to control post-transcriptionally these immune receptors. Induction of these resistance genes is primed in rdr6- and miR472-elicited mutants and this effect is associated with an enhanced basal and race-specific immunity in these backgrounds.
| To defend themselves against pathogens, plants have evolved potent inducible immune responses. The first line of active defense relies on the recognition of common features of microbial pathogens, such as flagellin (the major protein of bacterial flagellum), lipopolysaccharides, glycoproteins and chitin [1]. These microbial determinants are referred to as Pathogen- or Microbe- Associated Molecular Patterns (PAMPs/MAMPs) and are sensed by host-encoded Pattern-Recognition Receptors (PRRs) or surface receptors, which encode transmembrane receptor-like kinases. Upon PAMP detection, PRRs trigger a series of immune responses including, for instance, MAPK (mitogen-activated protein kinase) activation, reactive oxygen species (ROS) production, differential expression of genes, callose (β-1->3 glucose polymer) deposition and stomatal closure, which ultimately leads to basal immunity or PAMP-Triggered Immunity (PTI) [2]–[5]. To enable disease, pathogens produce a large array of divergent virulent determinants known as pathogen effectors that suppress different steps of PTI, resulting in disease susceptibility [6], [7]. As a counter-counter defense strategy, plants have evolved a repertoire of immune receptors, called disease resistance (R) proteins that can sense effector proteins and establish effector-triggered-immunity (ETI) [1]. The largest class of R proteins is composed of intracellular receptors that share structural homologies with mammalian innate immune receptors, such as NUCLEOTIDE-BINDING OLIGOMERIZATION DOMAIN-CONTAINING PROTEIN 1 (NOD1) and NOD2, which perceive bacterial PAMPs [8]. Plant NOD-like receptors (NLRs) are composed of nucleotide-binding (NB) and leucine-rich repeat (LRR) domains. They additionally contain an N-terminal domain that is composed of either a Toll/interleukin1 receptor (TIR) or a coiled-coil (CC) module, and are thus referred to as TNLs or CNLs, respectively [9]. These R proteins can directly sense pathogen effectors [1], however, in most cases they recognize indirectly these virulent determinants by detecting their effects on plant target proteins called ‘guardees’ [10]. Upon pathogen effector recognition, R proteins trigger a series of immune responses that significantly overlap with PTI responses, albeit with a stronger amplitude, and often result in a form of programmed cell death known as the hypersensitive response (HR) [1]. Importantly, constitutive expression or activation of R proteins often leads to constitutive cell death as well as severe developmental defects in the absence of pathogen [11]–[16], indicating that R genes and their products must be under tight negative control in unchallenged conditions. Consistent with this idea, transcriptional regulation, RNA processing, protein modifications, protein stability, and nucleocytoplasmic trafficking were shown to play a critical role in controlling R-mediated autoimmune responses [17].
More recently, RNA silencing has also emerged as a key regulatory mechanism that negatively regulates R gene expression [18]–[24]. RNA silencing is an ancestral gene regulatory mechanism that controls gene expression at the transcriptional (TGS, Transcriptional Gene Silencing) and post-transcriptional (PTGS, Post-transcriptional Gene Silencing) levels. The core mechanism of RNA silencing starts with the production of double stranded RNAs (dsRNAs) that are processed by RNase-III enzymes DICERs into 20–24 nt small RNA duplexes. One selected strand is subsequently incorporated into an RNA-induced silencing complex (RISC) containing an argonaute (AGO) protein, and guides these complexes onto sequence complementary RNA/DNA targets. The plant model Arabidopsis thaliana encodes 4 DICER-like proteins and 10 AGOs. DCL1 processes miRNA precursors into mature microRNAs that are mostly incorporated into the AGO1-RISC that guides mRNA degradation and/or translation inhibition of sequence complementary mRNA targets. DCL2, DCL3 and DCL4 are involved in the biogenesis of short interfering RNAs (siRNAs) from extensive dsRNAs produced from read through, convergent or overlapping transcription, endogenous hairpins as well as some miRNA precursors [25], [26]. As an example, overlapping sense and antisense transcripts that are produced at a functionally relevant disease resistance gene cluster, were found to be processed into siRNAs, leading to the down-regulation of several disease resistance gene transcripts within this cluster [18]. In addition, a large proportion of dsRNAs are produced by RNA-dependent RNA polymerases (RDRs) that convert single stranded RNAs into dsRNAs. RDR6, which is one out of six Arabidopsis RDRs, produces dsRNAs from viral and transgene transcripts as well as some endogenous transcripts [27]. These dsRNAs are processed in part by DCL4 into 21 nt siRNAs that direct PTGS of endogenous mRNA targets or exogenous RNAs derived from sense-transgenes or viral RNAs [28]–[31]. In addition to the biogenesis of primary siRNAs, plants have evolved the production of secondary siRNAs as a feed-forward amplification of silencing signals. These siRNAs are produced by the combined action of primary siRNA/miRNA-directed transcript cleavage and the activity of RDRs that use the target transcripts as template to generate dsRNAs [32]. In plants, the best-characterized endogenous secondary siRNAs are termed trans-acting siRNAs (tasiRNAs) [33], [34]. The biogenesis of these small RNA molecules is initiated by 22 nt long miRNAs that direct AGO1-mediated cleavage of a non-coding TAS primary transcript [35], [36]. One of the cleavage products is then converted by RDR6 into dsRNAs, which are processed by DCL4 into 21-nt phased siRNA duplexes. These secondary siRNAs guide an AGO protein to silence sequence complementary mRNA targets in trans. Importantly, this phenomenon is not restricted to non-coding transcripts but also targets protein-coding transcripts and both TNLs and CNLs have emerged as major targets of this silencing pathway [20], [21], [22]. For example, two 22 nt long miRNAs that initiate the production of RDR6-dependent secondary siRNAs, were found to directly control the tobacco disease resistance gene N, which recognizes the C-terminal helicase domain of the Tobacco Mosaic Virus (TMV) replicase protein [22]. These miRNAs play a functional role in N-regulation because their overexpression was shown to compromise N-mediated resistance to TMV [22]. Another recent study conducted in Solanum lycopersicum showed that miR482, a 22 nt long conserved miRNA that targets dozen of CNLs, was down-regulated in response to unrelated viruses as well as to a bacterium that encode RNA silencing suppressors [21]. Interestingly, this phenomenon was associated with the derepression of some CNLs that are targeted by miR482, suggesting that pathogen-triggered suppression of RNA silencing likely derepresses a whole repertoire of immune receptors during infection that might contribute to plant immunity [21].
Recent findings have thus revealed a critical role of miRNA-directed phased siRNA production in controlling the expression of R gene transcripts in the context of pathogen infection. Nevertheless, the interplay between the dynamic regulation of the RNA silencing machinery involved in miRNA-directed secondary siRNA production and the post-transcriptional regulation of R gene transcripts that are targeted by these small RNA species remains unknown. In addition, whereas some intracellular immune-receptors have recently been characterized in basal defense as well as plant defense against a disarmed bacterium very little is known on the functional relevance of plant NLRs in PTI [21]. The present study addresses some of these important issues by studying the regulation of RDR6 during antibacterial defense and the role of this silencing factor in the control of CNLs that are targeted by the Arabidopsis miR472, a miRNA related to miR482.
Although ARGONAUTE 1 (AGO1) and DICER-LIKE 1 (DCL1) were previously shown to contribute to PTI [37], [38], their regulation during the plant innate immune response has not been determined. To get a first insight into the regulation of components of PTGS during plant defense, we examined the expression levels of well-characterized PTGS factors in multiple conditions known to trigger PTI responses (Genevestigator database: https://www.genevestigator.com). Results from this analysis revealed that RDR6, AGO1 and SUPPRESSOR OF GENE SILENCING 3 (SGS3) mRNAs [39] were all down-regulated, with RDR6 showing the highest difference (consistently more than 2-fold in the various conditions analyzed) (Figures S1A, S1B). Accordingly, Reverse-Transcriptase Quantitative Polymerase chain reaction (RT-qPCR) analyses revealed a significant decrease in RDR6 mRNA levels in Arabidopsis leaves and seedlings treated with the flagellin-derived peptide flg22 (Figure 1), with a decrease in RDR6 transcripts starting at 10 min in Arabidopsis elicited seedlings (Figure S1C). A similar effect was observed with the type-three secretion (TTS) defective mutant Pto DC3000 hrcC−, which can elicit, but not suppress, PTI responses due to its inability to inject effector proteins within host cells (figure S1C). The PAMP-triggered dynamic regulation of RDR6 transcripts therefore suggested a potential role for RDR6 in orchestrating PTI responses. To test this idea, we first monitored the effect of the rdr6-15 loss-of-function mutation on the production of reactive oxygen species (ROS), one of the earliest cellular responses following PAMP perception, which is known to orchestrate the establishment of different defensive barriers against biotrophic pathogens [40]. We observed a more pronounced flg22-triggered oxidative burst in the rdr6 mutant as compared to WT-elicited plants (Figure 2A). However, given that the kinetics of flg22-triggered ROS production precedes the down-regulation of RDR6 transcripts in wild-type treated plants (Figure 1), these results suggest that the repression of RDR6 mRNAs is unlikely causative for this early PTI response. We also monitored the expression of PTI marker genes and found a primed induction of Flg22 RECEPTOR KINASE 1 (FRK1) in the rdr6-elicited mutant (Figure 2B, [41]). Of note, induction of FRK1 as well as the two other early PTI marker genes WRKY22 and WRKY29 was also moderately sensitized upon syringe infiltration of water in rdr6- versus WT-leaves (Figure 2B), suggesting that RDR6 may additionally repress a wounding response caused by mechanical stress.
We further monitored the flg22-triggered formation of cell wall depositions of callose, a late PTI response that plays a critical role in the establishment of basal immunity [42], [43]. An increase in flg22-induced callose depositions was observed in the rdr6-15 mutant as compared to WT plants, reinforcing a role for RDR6 in repressing this late PTI response (Figure 2C). It is noteworthy that a higher number of callose deposits were also observed in mock-treated rdr6-15 mutant versus WT plants, but not in untreated rdr6-15 mutant leaves (data not shown), suggesting that RDR6 may additionally prevent callose deposition upon wounding caused by syringe infiltration.
Natural surface openings, such as stomata, are important entry sites for bacterial plant pathogens such as Pto DC3000 and previous studies have shown that stomata closure plays an active role in limiting bacterial invasion as part of PTI responses [4]. Furthermore, fls2 mutants were found to be more susceptible to Pto DC3000 upon spray inoculation, although no discernible phenotype was observed using classical syringe infiltration assay, which bypassed basal immunity present at the leaf surface [44]. Given that the rdr6-15 mutant was sensitized for multiple flg22-triggered PTI responses, we reasoned that such silencing-deficient mutant might display enhanced resistance to Pto DC3000 upon spray inoculation. Consistent with this hypothesis, we found ∼10 times lower bacterial titer on rdr6-15 mutant as compared to WT plants spray inoculated with Pto DC3000 (Figure 2D). Collectively, these data provide evidence that the RNA silencing factor RDR6 acts as a negative regulator of basal immunity. These results also suggest that some positive regulators of plant defense are likely to be directly controlled by RDR6-dependent siRNAs.
Besides generating siRNAs directed against viral-, transgene- and transposon-derived RNAs, RDR6 is known to produce secondary 21 nt siRNAs from several endogenous loci including TAS genes [25], [26]. We thus searched for candidate defense gene transcripts that would be directly controlled by RDR6-dependent siRNAs. We used publicly available small RNA libraries derived from WT and rdr6 mutant leaves and selected candidate genes with a significant reduced amount of 21 nt siRNAs in the rdr6 as compared to the WT background. Using such criterion, we identified 75 loci that were likely targeted by RDR6-dependent siRNAs. Among those, 27 were previously annotated as TAS genes or tasiRNA targets. The remaining 48 protein-coding genes were enriched in GO categories ‘response to stress’ (http://bar.utoronto.ca/welcome.htm), (Figure S2), and include well-characterized RDR6-dependent targets such as AGO1, which is targeted by miR168-directed secondary siRNAs [45]. In addition, thirteen other candidate genes were annotated as miRNA targets and include multiple disease resistance gene transcripts that were previously identified as targets of miR472 (Figures S3, S4), a 22 nt long miRNA that is at least in part loaded into AGO1-RISC [35], [46]. These R genes are phylogenetically related to the functionally relevant disease resistance gene RPS5, which was previously characterized in ETI [47].
Given that At1g51480 and At5g43730 were among the CNL transcripts with the most matching secondary siRNAs (Figure S4), we decided to further characterize their regulation by RDR6 in both naïve and flg22-challenged conditions. These candidate genes are referred to here as Resistance Silenced Gene 1 (RSG1, At1g51480) and Resistance Silenced Gene 2 (RSG2, At5g43730). We also included RPS5 in this analysis, which was previously validated as miR472 target in Parallel Analysis of RNA Ends (PARE) datasets. We found a mild enhanced accumulation of these three candidate transcripts in unchallenged rdr6 mutant as compared to non-treated WT seedlings (Figure 3A), suggesting that these mRNAs are weakly controlled by RDR6-dependent siRNAs in naïve conditions, presumably due to their low basal transcriptional level in unchallenged conditions as previously observed for several disease resistance genes [17], [48]. We next monitored the levels of these mRNAs upon flg22 treatment in both WT and rdr6 mutant backgrounds. Whereas a mild increased induction of these transcripts was found in WT-elicited background, as observed in publicly available datasets (Figure S5), a 10- to 20-fold enhanced accumulation of these transcripts was obtained in the rdr6-elicited mutant seedlings, indicating cell priming in the absence of RDR6-dependent siRNAs (Figure 3B). These results therefore indicate that RDR6-dependent secondary siRNAs negatively regulate these CNL transcripts and that this post-transcriptional regulatory control is particularly relevant during PTI, when these disease resistance genes are presumably transcriptionally activated.
Given that miR472 was shown to target the above CNL mRNAs and to initiate the production of RDR6-dependent secondary siRNAs at these loci [20], [21], [22], we next characterized the role of this particular miRNA in the regulation of these candidate CNL transcripts as well as other orphan targets. For this purpose, we first transformed Arabidopsis with a construct containing AtmiR472 driven by the strong Cauliflower Mosaic Virus (CaMV) 35S promoter and selected a reference line (referred to as miR472OE line) exhibiting high miR472 accumulation compared to WT (Figure 4A). This line displayed a 25% and 30% reduction in the accumulation of RPS5 and RSG1 transcripts, respectively (Figure S6), providing further evidence that miR472 targets these CNL mRNAs in unchallenged conditions. Furthermore, genome-wide small RNA deep sequencing analyses revealed a drastic enhanced accumulation of secondary siRNAs at the 3′ ends of miR472 target sites for RPS5, RSG1 and RSG2 mRNAs as well as for 16 other CNL transcripts in the miR472OE line as compared to wild type seedlings (Figures 4B, 4C, S7). It is noteworthy that no siRNA were identified upstream the miR472 target site, which is in agreement with the rapid degradation of this region after miRNA-guided cleavage [49]. Furthermore, normal levels of tasiRNAs were identified in miR472OE as compared to WT seedlings (Figure S8), proving evidence that the enhanced accumulation of CNL-derived secondary siRNAs are not due to a general activation of the RDR6-dependent pathway in this transgenic line. Collectively, these results strongly reinforce a role for miR472 in initiating the biosynthesis of RDR6-dependent secondary siRNAs at our candidate CNL transcripts and revealed additional CNLs that are directly targeted by this regulatory process including the other functionally relevant disease resistance gene SUMM2 (Figure S9) [50].
We next analyzed the mRNA accumulation of two candidate CNLs in the miR472OE line challenged with flg22. Flg22-triggered induction of RPS5 and RSG1 mRNAs was significantly impaired in miR472OE line as compared to WT-elicited control (Figure 5A), supporting a role for miR472 in regulating the accumulation of these targets during flg22 elicitation. We also examined different PTI features in the miR472OE reference line by monitoring ROS production and callose deposition upon flg22 treatment. While this transgenic line displayed a normal flg22-triggered ROS production as compared to WT-elicited control (Figure 5B), we found a reduced number of flg22-induced callose deposits relative to WT-treated plants (Figure 5C), indicating that the miR472OE reference line is altered in the latter PTI response. It is noteworthy that similar PTI phenotypes were observed in another independent transgenic line overexpressing miR472 (Figure S10).
To get further insights into the role of miR472 in the regulation of CNL transcripts and PTI responses, we further characterized a transgenic line carrying a T-DNA insertion within the promoter of the AtmiR472 locus (Salk_087945, referred to as miR472m). This line displayed a drastic decrease in the accumulation of the mature form of miR472 relative to the levels of this miRNA in WT background (Figure S11). Furthermore, a primed induction of RPS5 and RSG1 transcripts was found in the miR472m line relative to WT background treated with flg22 (Figure 5A), supporting a role for miR472 in repressing mRNA accumulation of these CNL mRNAs during flg22 elicitation. Further phenotypic analyses in this line revealed a more pronounced flg22-induced ROS production and callose deposition, thereby mimicking the primed PTI responses observed in the rdr6-elicited mutant (Figures 5B, 5C). We thus conclude that miR472 and RDR6-dependent secondary siRNAs regulate PTI responses likely by targeting a whole repertoire of CNL transcripts.
Finally to determine the role of miR472 in basal resistance, we inoculated the virulent Pto DC3000 strain on miR472OE and miR472m lines and monitored bacterial titers in these genetic backgrounds as compared to WT-infected control. We found an increased Pto DC3000 titer in the miR472OE line, and, conversely, a reduced growth of this bacterium in the miR472m line as compared to WT-infected control (Figures 5D, S10). These results indicate that miR472 not only represses PTI responses but also negatively regulates basal resistance against Pto DC3000. These results also suggest that a subset of CNLs, which are targeted by miR472 and RDR6-dependent secondary siRNAs, may control basal resistance against Pto DC3000.
The effective targeting of RPS5 mRNAs by miR472 and RDR6-dependent secondary siRNAs (Figure 4B, 4C), together with the well-characterized role of RPS5 in recognizing the bacterial effector AvrPphB and mounting ETI [47], prompted us to investigate the role of miR472 and RDR6 in RPS5-mediated resistance. For this purpose, the rdr6-15 and miR472m lines were first inoculated with a Pto DC3000 strain carrying AvrPphB and bacterial titers were monitored at 4 days post-inoculation. Results from these analyses indicated a significant enhanced RPS5-mediated resistance in both rdr6 and miR472m as revealed by lower bacterial titers in these mutants as compared to WT-infected plants (Figure 6A), which is consistent with the enhanced accumulation of RPS5 transcripts in these mutant backgrounds (Figures 3A, 3B, 5A). Of note, this phenomenon was specific to RPS5-mediated resistance, because no phenotype was observed upon inoculation of rdr6 and miR472m lines with Pto DC3000 expressing AvrRpt2, a bacterial effector that is recognized by another CNL that is not targeted by miR472 (Figure S12, [13]). We next inoculated the Pto DC3000 (AvrPphB) strain on the miR472OE reference line and monitored bacterial titers as well as disease symptoms at 4 days post-inoculation. Interestingly, we found a significant enhanced Pto DC3000 (AvrPphB) titer in the miR472OE line as compared to WT-infected plants (Figure 6B), which was associated with a rescue of both chlorotic and necrotic disease symptoms in this transgenic plants (data not shown), thereby mimicking the phenotypes observed in rps5 loss-of-function mutants (Figure 6A, [47]). We conclude that overexpression of miR472 is sufficient to compromise RPS5-mediated resistance, which is consistent with the reduced levels of RPS5 mRNAs in this transgenic line (Figures 5A, S6). Collectively, these results indicate that miR472 and RDR6 negatively regulate not only PTI but also RPS5-mediated resistance, suggesting a critical role for RPS5 and other CNLs in basal and race-specific immunity.
The predicted target site of miR472 is embedded within a region encoding the P-loop domain, which is highly conserved in a large repertoire of CNL disease resistance proteins [9]. It is therefore likely that multiple CNLs are controlled by this particular miRNA and, in agreement, 19 CNL transcripts were experimentally validated as miR472 targets in Arabidopsis seedlings overexpressing miR472 (Figures 4, S7). This suggests that the enhanced basal resistance phenotype observed in the rdr6 and miR472m mutants might not only be due to the constitutive expression and/or primed induction of the few CNLs that have been characterized in these mutant backgrounds (e.g. RPS5), but also likely to multiple other relatives that are targeted by these small RNAs, rendering the functional characterization of these CNLs challenging. To circumvent this issue, we first introduced, in the rdr6 mutant background, mutations that abolish CNL-mediated signaling, and subsequently monitored Pto DC3000 titer in these double mutant backgrounds. Since several CNLs are known to trigger SA-signaling/biosynthesis [51], including RPS5 [52], we hypothesized that the SA-dependent defense response might be constitutive in the rdr6 mutant background. Consistent with this idea, we found a constitutive expression of the SA-dependent marker gene PATHOGENESIS-RELATED 1 (PR1) and the ISOCHORISMATE SYNTHASE1 (ICS1) (Figures 7A, 7B) [51], [53], as well as an enhanced resistance to Pto DC3000, in the rdr6 mutant as compared to WT control (Figures 7C, 7D, 7E, 7F). Importantly, this increased resistance to Pto DC3000 was abolished by introducing mutations that compromise SA-biosynthesis (the sid2-2 mutation, [53]) or SA-signaling (the npr1-1 mutation, [54]) in the rdr6-15 mutant background (Figures 7C, 7D). These results therefore indicate that the enhanced basal resistance achieved in the rdr6 mutant relies on the constitutive activation of the SA-dependent defense response, which might be initially triggered by the enhanced accumulation of CNLs that are no longer controlled by RDR6-dependent secondary siRNAs in this mutant background.
To get further insights into the role of these CNLs in the enhanced basal resistance phenotype observed in the rdr6 mutant, we took advantage of the property of the REQUIRED FOR MLA12 RESISTANCE (RAR1) protein. RAR1 is part of a molecular chaperone complex, containing HEAT SHOCK PROTEIN 90 (HSP90) and SUPPRESSOR OF G-TWO ALLELE OF SKP1 (SGT1), and plays a major role in NLR protein stability and activity [55]–[65]. Importantly, the steady-state accumulation of several CNL proteins, including RPS5, was shown to be dramatically impaired in rar1 loss-of-function mutants [58], [64]–[68]. We thus reasoned that by introducing a rar1 loss-of-function mutation in the rdr6 mutant background, we would destabilize CNL proteins whose cognate mRNAs are targeted by RDR6-dependent siRNAs, and therefore potentially restore disease susceptibility. Consistent with this hypothesis, we found that the increased resistance achieved in the rdr6 mutant was abolished in the rdr6-rar1 double mutant (Figure 7E). It is noteworthy that an enhanced Pto DC3000 titer was also found in the single rar1 and double rdr6-rar1 mutants as compared to WT control, indicating that RAR1 contributes to basal resistance as previously reported [64]. Given that RDR6 was found to negatively regulate RPS5-mediated resistance (Figure 6), we also monitored Pto DC3000 (AvrPphB) titer in the single rdr6 mutant as compared to the rdr6-rar1 double mutant. Results from these analyses indicated that the enhanced RPS5-mediated resistance observed in rdr6 mutants was partially compromised in the rdr6-rar1 mutant (Figure 7F). Collectively, these results indicate that the increased basal and specific resistance observed in the rdr6 mutant is dependent on the proper chaperoning of CNL proteins (e.g. RPS5), and might therefore be due to the enhanced accumulation of CNL proteins whose cognate mRNAs are no longer controlled by endogenous secondary siRNAs in this silencing-defective mutant.
RDR6 has been clearly implicated as a positive regulator of virus and viroid resistance. Indeed silencing of RDR6 in Nicotiana benthamiana results in hyper-susceptibility to some viruses and viroids [29], [30]. Moreover, in situ hybridization shows that viruses and viroids invade floral and vegetative meristems of N. benthamiana rdr6 RNAi plants [69], [70]. Here, by combining microbiological, genetic, genomic and molecular techniques, we demonstrate that RDR6 also acts as a negative regulator of PTI, basal defense as well as RPS5-mediated resistance. Indeed, we first showed that knock-out of RDR6 renders the plants more resistant to the hemibiotrophic pathogen Pto DC3000 and to the avirulent Pto DC3000 (AvrPphB) strain (Figures 2D, 6A, 7C, 7D, 7E). Furthermore, classical PTI responses such as ROS production, mRNA accumulation of PAMP-response genes as well as callose deposition were increased in rdr6 plants as compared to WT plants upon flg22 treatment (Figures 2A, 2B, 2C) [71]. Our results are thus in sharp contrast with the previously reported PTI phenotypes observed in ago1 loss-of-function mutants [38]. Why is there such a discrepancy between these PTGS-defective mutant phenotypes during PTI? One would argue that AGO1 is not only involved in the siRNA pathway but also in the canonical miRNA pathway. AGO1 impairment has thus additional consequences on the action of several miRNAs necessary for PTI [38], [72], thereby leading to the previously reported compromised PTI responses in ago1 loss-of-function mutants such as in other miRNA-defective mutants [37], [38]. It is also possible that RDR6-derived siRNAs that target disease resistance genes may not only be loaded into AGO1-RISC but also into other as-yet unknown AGO-RISCs, thereby contributing in part to the post-transcriptional regulation of CNLs in an AGO1-independent manner.
We also observed a constitutive activation of the SA defense marker gene PR1 and an enhanced expression of ICS1 in the rdr6 loss-of function mutant (Figures 7A, 7B). To examine the involvement of the SA-dependent defense in the enhanced disease resistance phenotype observed in rdr6 mutant, the rdr6-15 mutation was combined with the sid2-2, a loss-of-function mutation in ICS1 also referred to as SID2 [53]. Inactivation of ISC1/SID2 abolishes rdr6 resistance to Pto DC3000 and similar results were obtained in the npr1 mutant, which is impaired in SA signaling [54] (Figures 7C, 7D). Therefore, the SA-dependent defense pathway plays a critical role in the enhanced basal resistance phenotype observed in the rdr6 mutant. Such constitutive SA-dependent defense response might result from a derepression of a subset of CNL transcripts (e.g. RPS5 mRNAs) that are no longer regulated by secondary siRNAs in this silencing-defective mutant. Additionally, it may result from the post-translational activation of R proteins that would be constitutively present in a protein complex with RDR6 and active in the absence of this silencing factor, as observed in classical ‘guardee’ mutants [10]. Further investigations will be necessary to address these possibilities. Moreover, additional experiments will be required to determine whether the constitutive SA-dependent defense response observed in the rdr6 mutant is linked with the mild constitutive PTI responses in this silencing-defective mutant or whether both processes remain independent.
We observed a higher expression of PAMP-response marker genes in unchallenged rdr6 mutant as compared to WT seedlings and a significant hyper-induction of FRK1 in the rdr6-elicited mutant (Figure 2B). Furthermore, a more pronounced callose deposition as well as ROS production were observed in the rdr6 mutant challenged with flg22 as compared to WT-elicited seedlings (Figures 2A, 2C), indicating that this silencing-defective mutant is in a physiological situation known as “primed” state [73]. Those results also indicate that RDR6 encodes a novel negative regulator of PTI and further reinforce the idea that PTI is under a tight negative regulatory control as previously reported [2], [74], [75], [76]. Interestingly, an analogous RNA silencing-dependent regulatory phenomenon has been recently described in the transcriptional control of a disease resistance gene during PTI [24]. In this case, flg22 was shown to trigger the repression of a subset of RNA-directed DNA methylation factors and this process was associated with TGS release and with the transcriptional activation of this immune receptor, which is targeted by siRNA-directed DNA methylation in its promoter region [24]. Although RDR6 mRNAs were down-regulated in response to flg22 (Figure 1), it remains to be tested whether this molecular effect could be accompanied with a decrease in RDR6 protein levels as well as an eventual global release of RDR6-silencing as part of PTI responses.
How does RDR6 repress PTI, basal resistance and RPS5-mediated resistance? A first in silico analysis of small RNA populations derived from rdr6 mutant as compared to wild-type leaf samples allowed us to identify R gene mRNA candidates that are targeted by RDR6-dependent secondary siRNAs (Figure S4). However, the low abundance of secondary siRNAs in the majority of cases limited the identification of such miR472/RDR6 targets. By contrast, the use of our transgenic line overexpressing miR472 was instrumental in identifying with confidence 19 bona fide CNL target transcripts that contain the miR472 recognition sites as well as a large number of secondary siRNAs located downstream of their miR472-guided cleavage site (Figures 4, S9). Among these candidates, we have identified RPS5 and SUMM2 transcripts, which encode functionally relevant disease resistance proteins with well-characterized role in ETI [47], [50], [56]. These results therefore suggested that the miR472/RDR6-silencing pathway inhibits the accumulation not only of disease resistance gene transcripts encoding R proteins required for PTI and basal resistance but also of transcripts encoding immune receptors required for ETI. This implicates miR472 and RDR6 in a central regulatory pathway that modulates both ETI and PTI responses. Consistent with this, we found that RDR6 and miR472 act not only as negative regulators of PTI and basal immunity but also as repressors of RPS5-mediated resistance (Figure 6). In addition, the use of the rar1 mutant, which destabilizes disease resistance proteins including RPS5 [58], [59] was useful to provide genetic evidence that the enhanced disease resistance phenotypes observed in the rdr6 mutant is likely the result of a higher accumulation of NB-LRR proteins in this silencing-defective mutant (Figure 7).
The present work also provides genetic evidence that miR472- and RDR6-dependent secondary siRNAs efficiently control the steady state levels of three CNL transcripts. Indeed, we first showed that RPS5, RSG1 and RSG2 mRNAs were moderately up-regulated in untreated rdr6 mutant and significantly hyper-induced in this silencing-defective mutant challenged with flg22 (Figure 3B). Accordingly, a lower level of RPS5 and RSG1 mRNAs was detected in the miR472OE line (Figure S6) and a compromised induction of these CNL transcripts was also observed in this transgenic line challenged with flg22 (Figures 5A, 5B, 5C). Conversely, the miR472 knock-down line displayed higher accumulation of CNL mRNAs, which was associated with increased PTI responses (Figure 5), therefore mimicking the phenotypes observed in the rdr6 mutant (Figure 2). Collectively, these results indicate that both Arabidopsis RDR6 and miR472 negatively regulate the steady state levels of these candidate CNL transcripts in normal growth conditions and during PTI, although these effects appear more pronounced during the elicitation possibly due to the concomitant transcriptional activation of these R genes as previously demonstrated for other biotic stress responsive disease resistance genes [17], [18], [48].
Based on these results, we propose a model, which integrates the contribution of the miR472/RDR6-dependent PTGS pathway in plant immunity (Figure 8). In unchallenged conditions, both miR472 and RDR6 are constitutively expressed and negatively regulate a subset of CNL mRNAs at the post-transcriptional level (Figure 8). MiR472 guides cleavage of RPS5, RSG1, RSG2 and at least 16 other CNL transcripts that carry miR472 recognition sites and RDR6 uses 3′ cleavage products as substrates to generate dsRNAs that are presumably processed by DCL4 into 21 nt siRNAs (Figure 8). These secondary siRNAs can act in cis by guiding mRNA degradation of the CNL transcripts from which they are produced, but also likely in trans presumably by targeting CNLs as well as unrelated mRNAs that display sequence complementary to these small RNA species as was recently suggested in tomato ([21], Figure S7).
It is also likely that the genes encoding the above immune receptors remain at a transcriptionally inactive state in unchallenged conditions as demonstrated for several other disease resistance genes [17], [48]. In this case, the concomitant low basal transcriptional expression of CNLs and the miR472/RDR6-dependent post-transcriptional regulatory process would effectively deplete immune receptor mRNAs in the absence of pathogens, thus preventing an autoimmune response that would have detrimental consequences on plant fitness [1], [77]. This is reminiscent of recent findings on other 22 nt miRNAs/secondary siRNAs that target NLR transcripts in different plant species [20], [21], [22], as well as with the observation that the production of siRNAs at the disease resistance RPP4 cluster repress basal expression of several R gene transcripts within this cluster and likewise prevent constitutive activation of the SA-dependent defense pathway [18]. Our model also suggests that the mature form of miR472 is down-regulated during PTI, as a 4-fold decrease in the accumulation of this microRNA was observed in small RNA libraries generated by Li et al [38] upon flg22 treatment, which was confirmed in Arabidopsis leaves and seedlings treated with flg22 (Figure S13). We thus propose that upon pathogen detection, and perhaps also perception of non-adapted microbes, microbe-associated molecular patterns trigger the down-regulation of miR472, which in concert with the eventual transcriptional activation of CNLs, may contribute to the transient enhanced accumulation of CNL mRNAs/proteins at an early phase of the elicitation (Figure 8). This gene regulatory mechanism may also be reinforced by the down-regulation of RDR6-dependent silencing pathway as suggested by the rapid repression of RDR6 mRNAs during PTI (Figure 1). At a later phase of the elicitation, we propose that this double post-transcriptional layer of regulation mediated by miR472 and RDR6 likely trigger a robust resilencing of these CNL transcripts to prevent a sustained activation of the plant immune response.
Although R proteins have been extensively characterized in ETI [10], there is increasing evidence that these immune receptors can also contribute to basal defense as well as PTI responses in plants [78]. For example, a compromised basal resistance to virulent Pto DC3000 was previously reported in a rar1 loss-of-function mutant [64], and confirmed in the present study (Figure 7E), suggesting that plant NLRs contribute to basal immunity. More recently, a subclade of CNL proteins, characterized as ‘helper NB-LRR’, where not only required for ETI but also for basal resistance and this process was independent of their P-loop motifs [79]. Importantly, these CNLs additionally regulate PAMP-triggered SA-accumulation in response to a disarmed P. syringae strain, which provides evidence that plant NLRs contribute to PTI [79]. Nevertheless, these CNLs do not control early events of PTI responses triggered by flg22 or the elongation factor-derived peptide elf18, indicating that these immune receptors likely act downstream or independently of these early PTI signaling events [79]. In the present work, we showed that another subclade of CNLs, which are targeted by miR472 and RDR6-dependent siRNAs, possibly contribute to multiple PTI signaling events, including potentially flg22-triggered callose deposition and ROS production (Figure 3B). Interestingly, the product of one if this mRNA target, the RPS5 protein, was previously shown to reside in the same protein complex as the PTI receptor FLS2 [80], further supporting a molecular link between ETI and PTI components.
In conclusion we have established a direct link between miR472/RDR6-dependent PTGS and plant immunity. We showed that both miR472 and RDR6 act as negative regulators of PTI and ETI, presumably by repressing a subset of CNLs at the post-transcriptional level. Our data therefore sustain previous anticipations suggesting that in addition to their role in specific resistance, R proteins contribute to PTI [10], [64], [78], [81], [82]. Furthermore, given that flg22 as well as disarmed bacteria were shown to trigger Systemic Acquired Resistance (SAR), such as in response to pathogens expressing Avr products [83]–[86], we speculate that a potential release of miR472- and eventually of RDR6-dependent PTGS may also occur in distal tissues, and thereby might contribute to the transient derepression of a whole repertoire of disease resistance genes as part of the SAR response.
Arabidopsis thaliana seeds from the Col-0 accession were used as wild-type, the rdr6-15 T-DNA insertion line has been previously described in Xi et al [87]. We also used sid2-2, npr1-1 and rar1-21 mutant alleles. Plants were genotyped with the following primers and conditions: rdr6 (RDR6_LP:TGAATCCATTCCTGAACAAGC; RDR6_RP: CAATGCAACCTCATCTTGGATG; LB3: TAGCATCTGAATTTCATAACCAATCTCGATACAC), npr1(1g64280_F: AGGGGATATACGGTGCTTCAT; 1g64280_R: GAGCAGCGTCATCTTCAATTC); sid2 (sid2_F:CAGTCCGAAAGACGACCTCGAGTT;sid2_R:CTCATCATCTTCCTTCGTAAGTCTCC); rar1 (5g51700_F: AAGCAGGGAGTAAGTCAAATTTAC; 5g51700_R CAAACTGAAATCATGACTTCTTTG). All plants were grown in short days conditions subjected to a cycle of 8 h and 16 h of light and darkness, respectively, at a day/night temperature of 22.5/18.5° with 50–60% humidity for about 5–6 weeks. The plants were watered 16 h before inoculation to promote stomatal opening, thereby facilitating inoculation.
Pseudomonas syringae pv. tomato DC3000 (Pto DC3000) was grown at 28°C on NYGB medium (5 g L−1 bactopeptone, 3 g L−1 yeast extract, 20 ml L−1 glycerol) containing kanamycin (50 mg mL−1) and rifampicin (25 mg mL−1) for selection.
Pto DC3000, Pto DC3000 AvrPphB and Pto DC3000 AvrRpt2 from overnight culture were collected, washed once and resuspended in 10 mM MgCl2 at a concentration of 5×105 colony-forming units (CFU) mL−1. A. thaliana leaves were infiltrated with bacterial suspensions using a needleless syringe. Leaves were harvested immediately (0 dpi) or after 4 days. Two leaf discs (d = 0.4 mm) from two different leaves were washed in 10 mM MgCl2 and then ground with a Microfuge pestle. After grinding of the tissue, the samples were diluted 1∶10 serially. Samples were plated on NYGA solid medium (NYGB with 10 g L−1 agar) supplemented with antibiotics. Plates were placed at 28°C for 4 days and the CFU were counted. For spray inoculation bacteria were resuspended in 10 mM MgCl2 at OD600 of 0.2 (108 CFU/mL) and Silwet was added to a final concentration of 0.04%. All experiments presented were repeated three times and statistical differences were detected with a Wilcoxon test (*, P<0.05; **, P<0.01).
Reactive oxygen species released by leaf discs were assayed by H2O2-dependent luminescence of luminal [88]. Leaf discs were deposed into 96-well plate and incubated overnight in 200 µL H2O in a growth chamber. The next morning, 100 µL H2O containing 20 µM luminol and 1 µg horseradish peroxidase (Sigma) with or without 100 nM flg22 were added. Luminescence was immediately measured for 45 min using a Tristar LB 941 plate reader (Berthold technologies, Thoiry). At least 25 to 30 discs were tested by conditions.
For callose detection, leaves were infiltrated with 100 nM flg22 or water using a needleless syringe. After 15 h, about ten leaves from at least four independent plants were cleared by immersion in an alcoholic lactophenol solution by the method of Shipton and Brown [89] modified by Adam and Sommerville [90]. They were rinsed in 50% ethanol, then in water. Callose was detected by staining for 30 min in 150 mM K2HPO4 (pH 9.5) buffer containing 0.01% aniline blue (Sigma-Aldrich). After staining each leaf was mounted in 50% glycerol and examined with an Olympus Macro Zoom System Microscope MVX10 fluorescent microscope (excitation filter 365 nm and barrier filter 420 nm). Representative pictures are shown. The number of callose deposits per picture was determined using ImageJ (National Institutes of Health, Bethesda, MD, U.S.A.) and compared using a Wilcoxon test (P<0.05). We analyzed 25 to 30 pictures corresponding to more than five independent leaves for each treatment.
For RNA extraction, leaves or seedlings were collected, immediately frozen in liquid nitrogen, and then stored at −80°C. Total RNA was prepared by TRIzol (Invitrogen) extraction as recommended by the supplier (Invitrogen). For RT-PCR analysis, first-strand cDNA was synthesized using Superscript reverse transcriptase (Invitrogen,) from 1 µg of RNase-free DNaseI-treated (Promega) total RNA in a 20 µl reaction volume. Quantitative PCR reactions were performed on 1/40 of cDNA, 300 nM final concentration of each primer pair and LightCycler 480 SYBR Green I Master 2× conc. (Roche). PCR was performed in 384-well optical reaction plates heated at 95°C for 10 min, followed by 45 cycles of denaturation at 95°C for 15 s and annealing and elongation at 60°C for 30 s. A melting curve was performed at the end of the amplification by steps of 1°C (from 95°C to 50°C). Each experiment was repeated two to three times. Transcript levels were normalized to that of At2G36060, At4G29130 and At5G13440 genes. These reference genes display invariant expression over hundreds of publicly available microarray experiments. The gene-specific primers used in this analysis were listed in Figure S14.
For miR472 quantification, total RNA was isolated from plants using TRIzol reagent (Invitrogen) and treated with RNase-free DNaseI (Promega). Small RNAs were polyadenylated with ATP by poly(A) polymerase following the manufacturer's directions for the Poly(A) Tailing Kit (Ambion). After phenol-chloroform extraction and ethanol precipitation, the RNAs were reverse-transcribed with 200 U SuperScript III Reverse Transcriptase (Invitrogen) and 0.5 µg poly(T) adapter (Figure S14) according to the manufacturer's protocols (Invitrogen). The cDNAs were used for qPCR with miR472 as one primer and the reverse primer as described by Shi and Chiang [91]. 5,8S ribosomal RNA gene was used as internal control as previously described [91]. Sequences of miR472, reverse primer, poly(T) adapter and 5S primers are listed in Figure S14.
Total cellular RNA (5 µg), extracted using TRIzol reagent (Invitrogen) was processed into sequencing libraries using adapted Illumina protocols and sequenced at Fasteris (http://www.fasteris.com, Switzerland) using the Hi-seq 2000 sequencer. All next-generation sequencing data have been deposited to the NCBI Gene Expression Omnibus (GEO).
We took advantage of publicly available sRNA libraries from leaf tissue [92]. These data correspond to 2 replicates of WT and rdr6 sRNA sequenced using Illumina Genome Analyser technology. Replicates were pooled and sequence reads were matched against the Arabidopsis thaliana genome (TAIR10) using MUMmer v3.0 [93]. Only 15 to 30-nt long sRNAs reads with perfect match over their entire length were analysed further (2 434 780 and 1 753 064 for WT and rdr6 respectively). The number of 20–22 nt reads matching TAIR10 annotated protein coding genes locus or tasiRNA were then compared between WT and rdr6 libraries by differential analysis with NOISeq [94] using the parameters indicated below: k = NULL, norm = “rpkm”, long = 1000, q = 0.90, pnr = 0.5, nss = 1000, v = 0.02, lc = 1.
The WT and miR472OE sRNA libraries, containing 17 828 872 and 30 869 878 sRNA reads respectively, were processed using the same methods. Over those reads, 88.9% are 15 to 30-nt long and can be perfectly aligned to Arabidopsis genome. The miR472OE line was validated by comparing the number of reads mapping to all miRNA stem-loop loci (miRBase release 19; [95]–[98]) between WT and mutant sRNA libraries. Differential analysis of 20–22 nt reads in genes has then been done as described in the previous paragraph.
|
10.1371/journal.pbio.0050036 | Exon Silencing by UAGG Motifs in Response to Neuronal Excitation | Alternative pre-mRNA splicing plays fundamental roles in neurons by generating functional diversity in proteins associated with the communication and connectivity of the synapse. The CI cassette of the NMDA R1 receptor is one of a variety of exons that show an increase in exon skipping in response to cell excitation, but the molecular nature of this splicing responsiveness is not yet understood. Here we investigate the molecular basis for the induced changes in splicing of the CI cassette exon in primary rat cortical cultures in response to KCl-induced depolarization using an expression assay with a tight neuron-specific readout. In this system, exon silencing in response to neuronal excitation was mediated by multiple UAGG-type silencing motifs, and transfer of the motifs to a constitutive exon conferred a similar responsiveness by gain of function. Biochemical analysis of protein binding to UAGG motifs in extracts prepared from treated and mock-treated cortical cultures showed an increase in nuclear hnRNP A1-RNA binding activity in parallel with excitation. Evidence for the role of the NMDA receptor and calcium signaling in the induced splicing response was shown by the use of specific antagonists, as well as cell-permeable inhibitors of signaling pathways. Finally, a wider role for exon-skipping responsiveness is shown to involve additional exons with UAGG-related silencing motifs, and transcripts involved in synaptic functions. These results suggest that, at the post-transcriptional level, excitable exons such as the CI cassette may be involved in strategies by which neurons mount adaptive responses to hyperstimulation.
| The modular features of a protein's architecture are regulated after transcription by the process of alternative pre-mRNA splicing. Conditions that excite or stress neurons can induce changes in some splicing patterns, suggesting that cellular pathways can take advantage of the flexibility of splicing to tune their protein activities for adaptation or survival. Although the phenomenon of the inducible splicing switch (or inducible exon) is well documented, the molecular underpinnings of these curious changes have remained mysterious.
We describe methods to study how the glutamate NMDA receptor, which is a fundamental component of interneuronal signaling and plasticity, undergoes an inducible switch in its splicing pattern in primary neurons. This splicing switch promotes the skipping of an exon that encodes the CI cassette protein module, which is thought to communicate signals from the membrane to the cell nucleus during neuronal activity. We show that this induced splicing event is regulated in neurons by a three-part (UAGG-type) sequence code for exon silencing, and demonstrate a wider role for exon-skipping responsiveness in transcripts with known synaptic functions that also harbor a similar sequence code.
| Alternative pre-mRNA splicing expands protein functional diversity by directing precise nucleotide sequence changes within mRNA coding regions. Splicing regulation often involves adjusting the relative levels of exon inclusion and skipping patterns as a function of cell type or stage of development. In the nervous system, such changes affect protein domains of ion channels, neurotransmitter receptors, transporters, cell adhesion molecules, and other components involved in brain physiology and development [1,2].
There is growing evidence that various biological stimuli, such as cell excitation, stress, and cell cycle activation, can induce rapid changes in alternative splicing patterns [3,4]. These phenomena suggest that splicing decisions may be altered by communication between signal transduction pathways and splicing machineries, but such molecular links and mechanisms are largely unknown. The focus of the present study is to gain insight into these mechanisms using primary neurons as the model system.
Splicing decisions take place in the context of the spliceosome, which is the dynamic ribonucleoprotein machinery required for catalysis of the RNA rearrangements associated with intron removal and exon joining [5–7]. Spliceosomes assemble on pre-mRNA templates by the systematic binding of the small nuclear ribonucleoprotein particles, U1, U2, and U4/U5/U6, which leads to splice site recognition and exon definition. Thus, splicing decisions can be profoundly influenced by the strength of the individual 5′ and 3′ splice sites and by auxiliary RNA sequences that tune splice site strength via enhancement or silencing mechanisms. RNA binding proteins from the serine/arginine-rich (SR) and heterogeneous nuclear ribonucleoprotein (hnRNP) families play key roles in recognizing auxiliary RNA sequences from sites within the exon (exonic splicing enhancers or silencers; ESEs or ESSs, respectively) or intron (intronic enhancers or silencers; ISEs or ISSs, respectively). Despite numerous RNA motifs that have been functionally characterized as splicing enhancers or silencers, the mechanisms by which these motifs function in combination to adjust splicing patterns are not yet well understood [8,9]. The broad significance of this problem is highlighted by the observation that nearly 75% of human pre-mRNAs with multiple exons undergo alternative splicing [10]. In addition, numerous point mutations in splice sites or splicing control elements have been linked to genetic disease [11].
Insights into the molecular basis for induced effects on alternative splicing have come from a variety of biological systems. Cell excitation by KCl-induced membrane depolarization in the pituitary cell line, GH3, has been shown to induce skipping of several cassette exons, including the STREX exon of the BK slo channel, as well as the N1 and C1 exons of the glutamate NMDA receptor NR1 subunit (GRIN1 gene) [12]. Note that the CI cassette exon is called exon 21 in other studies. Calcium/calmodulin kinase (CaMK) IV was implicated in these effects based on the loss of excitation-induced exon skipping by a specific inhibitor of this enzyme, whereas the reverse effect was observed by expression of the catalytic subunit of the enzyme. In this study, a 54-nucleotide region of the 3′ splice site of the STREX exon was identified and found to be sufficient to mediate similar effects when attached to a heterologous exon. In another study, application of the drug pilocarpine induced skipping of the EN exon of clathrin light chain B, the N1 exon of c-src, and the CI exon of GRIN1 transcripts in the rat hippocampus and/or cortex of living animals [13]. In the T cell–derived cell line JSL1, phorbol ester treatment has been shown to induce cell excitation together with increased skipping of exon 4 of CD45 pre-mRNA. These effects are dependent upon a 60-nucleotide splicing silencer within exon 4, and hnRNP L has been implicated as a factor involved in mediating these effects [14,15]. In another study, phorbol ester has been shown to increase CD44 v5 exon inclusion in a manner that was dependent upon Sam68 and the MAPK/ERK pathway [16]. In HeLa cells, the cellular response to heat shock was found to generate potent and widespread splicing silencing, and SRp38 was identified as the protein factor responsible for these effects [17]. A general silencing effect on pre-mRNA splicing by SRp38 was also observed in cells undergoing mitosis [18].
Recent studies have implicated hnRNP A1 as a factor that mediates induced changes in alternative splicing in response to osmotic stress. NIH 3T3 cells that have undergone osmotic stress have a relative increase in cytoplasmic A1 compared to nuclear A1, and this effect is associated with the phosphorylation of carboxy terminal residues that extend into the M9 region [19,20]. HnRNP A1 has been shown to regulate a variety of alternative splicing events through the recognition of RNA sequences related to the consensus binding motif, UAGGGA/U [11,21,22]. Two RNA binding domains and a glycine-rich region are involved in sequence-specific recognition, and a sequence near the carboxyl terminus, M9, is required for nuclear import. HnRNP A1 has been shown to function as a substrate for protein kinase C (PKC), protein kinase A (PKA), and casein kinase [23]. HnRNP A1′s ability to silence splicing of CD44 v5 exon was reduced by the overexpression of a constitutively active form of the mitogen-activated protein/ERK kinase kinase 1, and by the small GTP-binding protein Cdc42 in NIH 3T3 and erythroleukemia cell lines [24]. Together, these results suggest that the functional properties of hnRNP A1 could be altered in complex ways through signaling pathways in response to various biological stimuli.
The CI cassette exon of the GRIN1 gene was chosen for this study based on its known hnRNP A1–mediated exon silencing mechanism, which involves two exonic UAGGs and an intronic GGGG motif [25]. Here we develop a system to study the response of the CI cassette to cell excitation in neurons of primary cultures, and we utilize this system to investigate the role of the UAGG-related exon silencing code in the induced exon-skipping response. We extend this with complementary biochemical approaches to test for binding- and expression-level effects on hnRNP A1 protein in nuclear extracts prepared from resting and excited cultures. We also take advantage of the primary neuronal cultures to explore the role of endogenous ion channels in the induced exon-skipping phenomenon using antagonists of the NMDA receptor and other physiologically relevant calcium channels. An expanded analysis of exon-skipping responsiveness of endogenous transcripts involved in synaptic and non-synaptic functions is also presented.
Primary neurons, due to their responsiveness and plasticity, should represent a useful model system to study activity-induced changes in alternative splicing, but to date, such systems have not been well developed. In this study, we have developed low-density cortical cultures from the embryonic rat forebrain as an experimental system to address the molecular mechanisms of induced changes in alternative splicing using cell-specific and biochemical approaches. Upon dissociation and plating in culture, the postmitotic neurons extend processes, establish synaptic connections, and become electrically active, whereas glial cells in these cultures play supportive roles.
Initially, changes in endogenous splicing patterns were monitored in the cortical cultures both as a function of KCl-induced depolarization, and cell differentiation using the CI cassette exon as a readout. The cortical cultures at 6, 8, 12, and 14 d in vitro (DIV) after plating were treated for 24 h with media containing 25 or 50 mM KCl, since these conditions are known to induce membrane depolarization in these cultures. Changes in the percent exon inclusion values were calculated as the difference between the value in the 50 mM KCl sample and the mock-treated control (ΔEI values).
Each stage of differentiation of the cultures showed similar trends in which CI cassette exon inclusion decreased (ΔEI, −32% to −36%; Figure 1, lanes 1–12) as a function of the KCl treatment. Based on the consistent response of each exon in this time frame, we used the cultures to ask whether KCl treatment produces transient or stable changes in the splicing patterns of the endogenous RNA transcripts. If transient changes in transcriptional or post-transcriptional events affect splicing factors involved in these mechanisms, we would expect to see a reversal to the basal splicing pattern with time after removal of the stimulus. Reversibility was tested in the 12-d cultures, again using the CI cassette splicing events as a readout. After treatment with KCl, the cultures were changed to fresh medium without KCl, and cells were harvested 0, 6, 12, and 24 h later. KCl-induced changes in the splicing of the CI cassette exon progressively lessened with time to approximately basal levels after KCl washout (ΔEI, −8%; Figure 1, lanes 13–24).
For comparison, we assayed for the effects on alternative exon 9 (E9) of the GABAA receptor γ2 (GABARG2 gene) in the same cultures, since E9 is known to undergo silencing by the polypyrimidine tract binding protein, but not hnRNP A1 (Figure S1A). Interestingly, E9 inclusion was responsive to the KCl treatment, but this effect was maximized at longer differentiation times of the cultures, and, in contrast to the CI cassette, E9 inclusion increased (ΔEI, 23%, lanes 1–12). These effects were also largely reversible after KCl washout (lanes 13–24). Thus, the KCl-induced effects on CI cassette and E9 inclusion are largely reversible, and suggest a role for mRNA homoeostasis in allowing for the readjustment of splicing patterns to basal levels.
To experimentally identify nucleotide sequence requirements for activity-induced alternative splicing in neuronal and glial cell types in the cortical cultures, we turned to cell-specific promoters that were previously characterized in transgenic mice. For this purpose, we assessed the ability of the alpha CaMKII promoter to drive neuron-specific expression in rat cortical cultures, since this promoter was shown previously to drive forebrain-specific expression in excitatory neurons of transgenic mice [26]. Fluorescent reporters were constructed in which the expression of enhanced yellow fluorescent protein (EYFP) was driven by various portions of the CaMKII promoter (Figure 2, constructs 1 and 2). Construct CaMKII_279 EYFP contained an 8.5-kilobase (kb) promoter region known to restrict expression of Cre to the forebrain of transgenic mice. For cloning purposes, a shorter promoter region was also tested, which contained the start site–proximal 2.2-kb promoter fragment (CaMKII_22 EYFP).
To determine expression patterns in the cortical cultures, confocal microscopy was used to measure the extent of overlap between EYFP expression and antibody staining with TRITC-labeled NeuN or glial fibrillary associated protein (GFAP), which served as molecular markers for neurons and glia, respectively (Figure 2B). For comparison, we also constructed a fluorescent reporter in which DsRed was fused to a 2-kb region of the Gfa2 promoter, Gfa2 DsRed (Figure 2, construct 3) [27]. The Gfa2 promoter was shown previously to drive expression predominantly in glial cells of transgenic mice [28]. As controls, the EYFP and DsRed coding sequences were also fused to the human cytomegalovirus immediate early (CMV) promoter (Figure 2, constructs 4 and 5).
The strong neuron specificity of expression of the CaMKII promoters in these cultures is shown as the percent overlap expression of each construct with NeuN or GFAP antibody staining (Figure 2A, top right). For each sample, a minimum of 200 cells with EYFP fluorescence were examined for TRITC-labeled NeuN staining. Whereas, the largest (8.5 kb) region of the promoter (CaMKII_279) showed 96% overlap with NeuN and 3% overlap with GFAP, similar results were observed with the 2.2-kb region (CaMKII_22; 95% overlap with NeuN and ∼5% overlap with GFAP). In contrast, the Gfa2 promoter was preferentially expressed in glia in these cultures (94%), and some overlap with neurons was observed (22%) in agreement with previous analysis in transgenic mice [28]. A summary of these results is also shown graphically (Figure 2C). As a reference, EYFP expressed from the CMV promoter (CMV_EYFP) showed 64% overlap with NeuN and 21% overlap with GFAP, whereas the DsRed expressed from the same promoter (CMV_DsRed) showed 54% overlap with NeuN and 26% overlap with GFAP. Due to its mixed expression profile, the CMV promoter was not used for further analysis.
Representative examples of the overlap expression of the two CaMKII EYFP reporters with NeuN is shown in Figure 2B. Overlap expression is indicated by yellow fluorescence in the nuclei of samples stained with TRITC-labeled anti-NeuN (CaMKII EYFP reporters, anti-NeuN panels), and the lack of overlap expression in samples containing the anti-GFAP antibody is clearly shown (anti-GFAP panels). The pattern of expression for samples containing the Gfa2 DsRed reporter characteristically showed overlap of DsRed fluorescence with FITC-labeled anti-GFAP, but not with FITC-labeled anti-NeuN (Gfa2 DsRed panels).
To confirm the neuron- versus glial-specific expression patterns of the promoters, we co-transfected the CaMKII_22_EYFP and Gfa2-DsRed plasmids into the same cortical cultures and determined overlap expression. No greater than 22% overlap would be expected based on the least specific of the two promoters, Gfa2. The observed overlap ranged from 14% to 19% (raw values), in contrast to 82% to 84% from the generally expressed CMV promoters. After normalizing these values to 100% transfection efficiency for the CMV samples, the co-expression of CaMKII and Gfa2 promoters ranged from 17% to 23% (corrected values). These results are in good agreement with the promoter selectivity shown above.
We next asked how these promoters affect splicing patterns in established neuronal (PC12 and ST15A) and muscle myoblast (C2C12) cell lines, in comparison to the primary cortical cultures. Similar CI cassette exon inclusion levels were observed from the two promoter types in the neuronal and non-neuronal cell types (ΔEI, <3.6%, Table 1). Thus, the increase in CI cassette exon inclusion arising from the promoter fusions in primary cortical cultures must reflect the properties of the neurons and glial cells in the cultures, and cannot be explained by the promoters alone.
Next we took advantage of the CaMKII and Gfa2 splicing reporters to characterize the KCl-induced effects on CI cassette exon silencing as a function of cell type in the primary cortical cultures. Wild-type CI cassette splicing reporters driven by the cell-specific promoters CaMKII_22 CI wt0 and Gfa2 C1 wt0 were transiently expressed in cortical cultures for 18 h, followed by treatment with KCl for 6 and 24 h. When transcripts were expressed from the CaMKII_22 promoter, CI cassette exon inclusion decreased in response to the KCl treatment, and this effect was consistent in neurons at all time points examined (Figure 3A, lanes 7–12). Similar effects were observed for the endogenous CI cassette exon when mock-transfected cultures were treated in parallel (lanes 1–6). Optimal effects were reached after 24 h for both endogenous and CaMKII_22-expressed transcripts. In contrast, when transcripts were expressed from the Gfa2 promoter, CI inclusion showed little or no change at the 6- and 24-h time points (lanes 13–18). Based on the cell specificity of the CaMKII promoter shown above, we conclude that neurons play the predominant role in the KCl-induced changes in alternative splicing in these cultures. Furthermore, the trend and magnitude of the effects for the exogenous CI cassette exon inclusion are in good agreement with those of the endogenous transcripts. As a corollary to this experiment, we measured the number of CaMKII_EYFP-expressing neurons that fail to exclude trypan blue. In mock- and depolarization-treated cultures, we observed that the viability of the neurons was high in both the mock-treated (94% viable) and depolarization-treated (93% viable) samples (Figure 3B). Thus, we conclude that the RT-PCR results shown above (Figure 3A, lanes 7–12) largely reflect splicing pattern changes in viable neurons.
To further validate these results, we transferred the E9 splicing reporter into the CaMKII and Gfa2 promoter constructs and tested their response to KCl treatment in the cortical cultures (Figure S1B). When expression was driven in neurons, E9 inclusion increased similarly to that of the endogenous transcripts (ΔEI, 17% and 15%, respectively), but this was not the case when expression was driven in glial cells (ΔEI, 3%). Thus, for a second test exon, the promoter system recapitulated the splicing response in neurons as observed for the endogenous transcripts.
Based on the well-behaved exon-skipping response of the CI cassette exon to KCl-induced depolarization in this system, we focused on this exon to identify nucleotide sequences involved in the response. A three-component code of two exonic UAGGs and an intronic GGGG motif, which was recently shown to mediate CI cassette exon silencing [29], served as the starting point for these experiments. Splicing reporters with point mutations in combinations of the three silencing motifs were subcloned under the control of the CaMKII_22 promoter, and assayed in the cortical cultures (Figure 4A).
In the absence of KCl treatment, the CI wt0 reporter with intact UAGG and GGGG motifs showed strong silencing activity (exon skipping) in the primary neurons in agreement with our previous analysis in cultured cell lines. Compared to the wild-type substrate, point mutations in a single silencer increased exon inclusion (Figure 4A, lanes 1, 4, 7, and 13), and multiple point mutations generated complete or nearly complete exon inclusion (lanes 10 and 16). In results new to this study, the UAGG and GGGG silencer motifs were found to be important for the depolarization-induced effects on exon skipping in the primary neurons. A complete disruption of KCl-induced exon skipping was associated with point mutations in two or more of the silencing motifs (substrates E17 and T8; lanes 10–12 and 16–18). In contrast, an increase in exon skipping occurred in parallel with the depolarization treatment for the wild-type substrate (ΔEI, −17%; lanes 1–3). Point mutations in single motifs showed a similar trend in which the KCl-induced effects on exon skipping were disrupted at various levels (substrates E8, E9, and D0; lanes 4–9 and 13–15). We next tested the role of exonic enhancer motifs in the same fashion, since these motifs function generally to antagonize exon silencing. The sequence, type, and position of the exonic enhancer motifs and their inactivating mutations are shown in Figure 4B. Of the six mutants, E2, E3, E4, and E5 showed reduced basal levels of CI exon inclusion in neurons, and the response of these exons to KCl-induced depolarization was either similar to (E6 and 5 m1, lanes 34–39) or reduced (E2, E3, E4, and E5, lanes 22–33) compared to the wild-type CI cassette (lanes 19–21). These results suggest that sequences outside of the UAGG motifs in the exon may also play a role in the induced splicing silencing response.
To validate these results, we asked whether a multicomponent UAGG silencing motif code is sufficient to confer sensitivity to KCl-induced depolarization; we introduced UAGG motifs into exon 5 of the gene encoding the human Dip13 beta adapter protein (DIP13B_HUMAN). This exon was chosen based on the use of algorithms that predicted strong 5′ and 3′ splice sites, a moderate number of exonic enhancer motifs (n = 10), and a lack of known exonic silencing motifs (MaxEntScan, ACEScan, and ESEFinder). Another attractive feature was the clustered arrangement of predicted enhancer motifs that allowed for the insertion of UAGG motifs at multiple discrete positions. The region containing exon 5 and its flanking splice sites was cloned into a chimeric splicing reporter by replacing the middle exon of the previously described SIRT1a plasmid [29], and the promoter was replaced with CaMKII_22 to direct expression in neurons (Figure 5A, DIP13_E5). When expressed in neurons, the DIP13_E5 splicing reporter was nearly insensitive to the excitation treatment (ΔEI, 0%; Figure 5B, lanes 1–3). In order to test the role of the silencing motif pattern in the response to excitation, we next introduced three exonic UAGG motifs and a 5′ splice site GGGG motif (DIP_3aG). Introduction of the silencing motif pattern indeed generated an exon-skipping pattern in resting neurons (61% exon inclusion), and in the presence of KCl treatment to induce excitation, exon inclusion progressively decreased to 45% (ΔEI, −16%; lanes 4–6). In another variant, DIP_E2, point mutations were introduced into the exon to destroy overlapping SC35 and ASF-SF2 motifs near the 3′ end of the exon (Figure 5A). For DIP_E2, exon inclusion decreased in resting neurons, and the response to excitation was similar to that observed for DIP_3aG (Figure 5B, lanes 7–9). Thus, we conclude that a multicomponent UAGG silencing motif code is sufficient to generate induced exon skipping in response to KCl-induced depolarization.
We reasoned that factors known to be involved in splicing silencing via UAGG silencing motifs in resting cells would be likely candidates to mediate the induced splicing silencing response in excited neurons. Alternatively, excitation might weaken the role of factors involved in antagonizing this silencing mechanism. We initially focused on hnRNP A1 to probe for changes in regulatory factors, because our previous work demonstrated direct binding of this factor to the UAGG motifs in HeLa nuclear extracts, and because its silencing effect in vivo was dependent upon the intact motif pattern.
To attempt to visualize whether changes in hnRNP A1 accompany KCl-induced depolarization, RNA-protein binding assays were used as sensitive readouts. In these experiments, 5-d cortical cultures were treated with KCl in the medium to induce depolarization, and cells were harvested for preparation of small-scale nuclear extracts (see Materials and Methods). The time of KCl treatment was shortened from 24 h to 12 h based on the expectation that any excitation-induced alterations to splicing factors should precede the splicing pattern shifts themselves.
Ultraviolet (UV) crosslinking was used to monitor direct protein binding to 5′[32P]-labeled RNA oligonucleotides under splicing conditions, and the crosslinked proteins were resolved by SDS-PAGE. A 22-mer containing three UAGG motifs showed increased crosslinking of a 34-kDa protein, suggestive of hnRNP A1, in nuclear extracts prepared from KCl-induced versus mock-treated samples (Figure 6A, lanes 1 and 2). Increased crosslinking of the 34-kDa protein in the KCl-induced extracts was observed in eight different nuclear extract preparations, and the fold increase was in the range of 1.5- to 2.5-fold. To determine whether hnRNP A1 was responsible for the observed increase in crosslinking, a size-matched control RNA with point mutations in the UAGG motifs (A1 mutant) was tested in parallel reactions. Crosslinking of the 34-kDa protein to the A1 mutant oligo was abolished, demonstrating that intact UAGGs are essential for binding. To confirm the identity of the 34-kDa protein, three UV crosslinking reactions equivalent to that shown in lane 2 were combined and immunoprecipitated with monoclonal antibody 9H10, which is specific for hnRNP A1. These results showed that, relative to the control antibody, the 34-kDa crosslinked protein was partitioned into the pellet (and depleted from the supernatant) in the presence of 9H10, supporting its identification as A1 (Figure S2). For comparison, parallel crosslinking reactions with distinct RNA oligos specific for human Tra2 and ASF/SF2 were also tested, but these proteins showed no detectable change in binding in extracts prepared from the KCl-induced versus mock-treated cultures.
Enhanced hnRNP A1 binding was also observed for the full-length CI cassette exon in UV crosslinking reactions prepared with nuclear extracts from KCl-induced cultures (Figure 6A. lanes 9 and 10). Note that the full-length CI cassette exon tested here contains three UAGGs distributed throughout the 111-nucleotide exon, which is a less-concentrated motif arrangement relative to the A1 oligo. In contrast to hnRNP A1, higher molecular weight proteins in the 57- to 100-kDa range in these samples serve as reference comparisons, since these displayed similar crosslinking levels in the two nuclear extracts (lanes 9 and 10, Ref bands). The identity of hnRNP A1 was verified by competition experiments with unlabeled A1 oligo, which reduced crosslinking to the 34-kDa protein in a concentration-dependent manner (unpublished data).
To verify the change in hnRNP A1 binding observed in the UV crosslinking assay, affinity selection was used as a complementary method. For this experiment, the full-length CI cassette exon with three UAGGs was subcloned upstream of the M3 hairpin to provide an affinity tag that was specific for the MS2-MBP fusion protein [30]. The CI cassette exon was subcloned without the M3 hairpin as a control. To assay for hnRNP A1, one half of the eluted samples were separated by SDS-PAGE and immunoblotted with the 9H10 antibody. Parallel blots of the second half of the samples were developed with an antibody specific for ASF/SF2. Binding of hnRNP A1 was dependent upon the presence of the M3 hairpin (Figure 6B, lanes 13–16), and an increase (1.5-fold) in A1 binding to the M3_E18 substrate was observed in the KCl-induced versus mock-treated extracts (lanes 13 and 14). This difference in A1 was similar in the input samples prior to RNA binding (lanes 11 and 12). The increase in the level of A1 protein was similar to the input samples prior to RNA binding. In contrast, ASF/SF2 showed no change in binding to the M3_E18 substrate in the two samples, nor was there an observable difference in ASF/SF2 in the input samples (lanes 17–22).
To confirm the difference in nuclear levels of hnRNP A1, quantitative Western blotting was used to measure A1 levels in nuclear extracts from resting and excited cultures relative to known levels of recombinant A1 (Figure 6C). These results verify that an increase in nuclear A1 protein levels in the cultures is associated with the depolarization treatment.
We would expect cell excitation initiating at the cell membrane to communicate changes to splicing factors in the nucleus via calcium-mediated signal transduction pathways. In the absence of antagonists, the NMDA receptor channel opens in response to membrane depolarization and serves as the major conduit for calcium entry into neurons. Because NMDA receptors are functionally intact in neurons in these cultures, we wished to take advantage of this property of the culture system to ask whether NMDA receptors play a role in the induced effects on splicing observed here. To address this question, we expressed the CI wt0 splicing reporter in 5-d cortical cultures under the control of the CaMKII_22 promoter, and the cultures were pre-incubated with antagonists specific for the NMDA receptor before KCl-induced depolarization. MK801 and AP5 antagonists were used for these experiments because of their known specificity and effectiveness in inhibiting the NMDA receptor calcium channel in cortical cultures [31,32]. AP5 binds competitively to the extracellular glutamate (NMDA) site of the NR2 subunit, whereas MK801 binds to the channel itself, blocking ion flow. The exon-skipping response of the CI cassette exon was strong in the control (mock treated) samples (Figure 7A, lanes 1–3), but in the presence of either antagonist, this response was attenuated in a dose-dependent manner (lanes 4–15). Relative to the control sample (ΔEI, −25%), exon skipping decreased in the presence of MK801 (ΔEI, −10%) and in the presence of AP5 (ΔEI, −4%). In contrast, when cultures were pre-incubated with bicuculline, which is an antagonist of the GABAA receptor, only slight effects were observed on induced exon skipping (ΔEI, −22%; lanes 16–21) relative to the control (ΔEI, −25%; lanes 1–3). Thus, to a first approximation, the effects of MK801 and AP5 implicate a role for the NMDA receptor in mediating the exon-skipping response of the CI cassette exon in this system.
To extend these results, we used cell-permeable inhibitors to ask whether calcium-mediated signal transduction pathways downstream of NMDA receptors are involved in the induced splicing silencing response of the CI cassette exon. Relative to the mock-treated control, the exon-skipping response was strongly inhibited in the presence of 15 μM KN93, which is an active site-based inhibitor of CaMK I, II, and IV (Figure 7B, lanes 22–24 and 40–42), and this effect was lessened when the inhibitor concentration was reduced to 3 μM (lanes 37–39). The control compound, KN92, had little or no effect in these experiments (P. An and P. J. Grabowski, unpublished data). We also found that the exon-skipping response was inhibited by H89, which has been widely used as an inhibitor of PKA (lanes 31–36). In an attempt to confirm this effect, the inhibitor KT5720, which is reported to have a higher specificity for PKA, was also tested. Interestingly, KT5720 had little or no effect at 2.5 and 10 μM concentrations (lanes 25–30), indicating that the inhibitory effect of H89 on splicing silencing observed here may involve a distinct pathway (or pathways).
To determine if additional calcium channels could play a role in the splicing response of the CI cassette exon, we tested the effects of specific antagonists of the AMPA/kainate receptor and voltage-gated calcium channels (L- and N-type) (Figure S3). Relative to control samples, the exon-skipping response was reduced, but not eliminated, in the presence of nimodipine (antagonist of L-type calcium channels) and conotoxin (N-type calcium channels), whereas CNQX (AMPA/kainate receptors) showed little or no effect. Taken together, these results are consistent with roles for multiple calcium channels in the induced splicing response of the CI cassette exon in neurons.
Next, we expanded the analysis of endogenous transcripts in the cortical cultures to test whether additional alternative cassette exons were responsive to depolarization, and if so, to determine whether exonic UAGG silencing motifs were required. We were also prompted to examine transcripts that encode protein components with synaptic functions, since the NMDA NR1 receptor and other calcium channels were implicated in the splicing response of the CI cassette exon from the experiments described above. Of 14 new exons tested, seven showed a significant exon-skipping response, five showed little or no response, and two showed an increase in exon inclusion (Figure 8).
Strong exon-skipping responders were found to contain hnRNP A1–type silencing motifs in or near the responsive exon (Figure 9). Exon 3 of hnRNP H3 and exon 2 of the RNPS1 transcripts (ΔEI, −32% and −25%, respectively) contain either one exonic UAGG and a 5′ splice site GGGG motif (H3), or two exonic UAGGs (RNPS1). Exon 35 of PLCβ4 also showed a strong response (ΔEI, −26%), and this exon contains two TAGA motifs, which reflects a four of six match to hnRNP A1 motifs reported for SMN2 exon 7 and human CD44 v5 exon [33]. These motifs cannot be the sole determinants of the response, however. Exon 4 of hnRNP H1, which is identical in size (139 nucleotides) and highly homologous to exon 3 of hnRNP H3, showed little or no response (ΔEI, −3%) in the same samples. Interestingly, there is one exonic UAGG and a GGGG tetramer near the 5′ splice site of this exon, but it lacks a third silencing motif (exonic GGGG) that is found in exon 3 of hnRNP H3. Moreover, constitutive exon 8 of the MEN1 transcript remains entirely unresponsive even though this exon contains two exonic UAGGs and a 5′ splice site proximal GGGG motif.
Of the exons that showed little or no response, exon EN of clathrin light chain B (CLCB) was the most interesting. In a previous report, this exon showed increased exon skipping in the cortex and hippocampus of rats treated with pilocarpine [13]. In cortical cultures under the conditions tested here, however, exon EN showed a stable splicing pattern even though it is a well-skipped exon (exon inclusion, 30%). Other alternative exons with a predominant skipping pattern that were unresponsive in these cultures include exon 21 of MAP4k4, exon 8 of hnRNP A1, exon N1 of GRIN1, exon 7 of Chl, and exon 7 of Agrin. These exons, including EN of CLCB, lack hnRNP A1 motifs. Thus, the exon-skipping responsiveness of certain exons to depolarization treatment cannot be explained solely by a weakening of the general splicing machinery that causes all skipped exons to undergo stronger exon-skipping responses.
Finally, two exons showed moderate increases in exon inclusion during the depolarization treatment. These included exon 9 of the GABAA receptor γ2 transcript, and the STREX exon of the c-slo transcript (ΔEI, 18%, and 12%, respectively). These results suggest that some of the biochemical changes resulting from the depolarization treatment in neuronal cells can, in principle, increase exon inclusion.
At the level of pre-mRNA splicing, the CI cassette of the NMDA R1 receptor is one of several responsive exons that undergo an increase in exon skipping in response to neuronal excitation. Despite the importance of NMDA receptors in neuronal excitability at the level of ion channel function, the mechanisms by which stimuli are communicated to the nuclear splicing machinery to affect changes in alternative splicing are poorly understood. Moreover, methodologies to address these questions in primary neurons, in which NMDA receptors are functionally intact, have not been well developed.
Here we describe a promoter-based expression assay that reports splicing changes in response to cell excitation with tight specificity for the neurons of cortical cultures derived from embryonic rat brain. Due to its neuron-specific expression in the forebrain regions of transgenic mice, the alpha CaMKII promoter was tested for its ability to provide a neuron-specific readout in a primary culture setting. In cortical cultures, full-length as well as truncated forms of this promoter displayed a strong preference for expression in neurons versus glial cells as demonstrated by co-localization of promoter-directed EYFP expression together with antibody staining for cell-specific markers. Moreover, these effects were dependent upon the primary culture context, since neuron-specific splicing patterns were eliminated when the same promoter-EYFP cassettes were expressed in neuronal versus non-neuronal cell lines.
Using splicing reporters driven by the CaMKII promoter, we show that neuronal cells play the predominant role in the acute splicing response to excitation, and these splicing pattern changes reflect the trend and magnitude of those observed for the corresponding endogenous transcripts. In neurons of these primary cultures, the exon-skipping response of the CI cassette exon was mediated by combinations of UAGG-type silencing motifs. The general importance of a strong silencing motif pattern was confirmed by a gain-of-responsiveness due to the transfer of a similar UAGG-type silencing motif pattern into a constitutive exon. In this heterologous context, the transfer of multiple exonic UAGG motifs and a 5′ splice site proximal GGGG motif generated a basal-level exon-skipping pattern in the absence of depolarization, and exon skipping was further increased, over and above this basal level, as a function of depolarization. Furthermore, by expanding this analysis to fourteen alternatively spliced exons, several exons with strong exon-skipping responses were found to contain multiple hnRNP A1 silencing motifs consistent with a plausible role for hnRNP A1 in these effects. These results extend our previous work, which identified a multicomponent UAGG and GGGG motif code for basal-level splicing silencing of the CI cassette exon in established cell lines [29]. Additional studies have documented a similar exon-skipping response of the CI cassette exon to cell excitation in rat brain [13] and in GH3 pituitary cells, but in these previous studies nucleotide sequences required for the effects were not defined.
The first RNA element shown to confer an exon-skipping response to KCl-mediated depolarization was defined for the STREX exon of the BK potassium channel. This element, termed CaRRE (CaMKIV-responsive RNA element), was defined as a 54-nucleotide pyrimidine-rich region in the 3′ splice site of the STREX exon, and a pyrimidine-rich exonic RNA element, termed D56, was also found to contribute to this splicing response in GH3 cells [12]. The CaRRE sequence was later refined by mutagenesis in primary mouse cerebellar cultures to comprise the sequence (5′) CACATNRTTAT (3′) [34]. We note that neither the CaRRE, nor the D56 element, are present in the 3′ splice site upstream of the CI cassette. Moreover, the UAGG-type silencing motifs defined here for the responsiveness of the CI cassette are absent from the STREX exon. These observations suggest that the exon-skipping responsiveness of these two exons is likely to involve distinct silencing mechanisms. The accompanying paper by Lee et al.[35] reports the identification of two types of CaRRE elements required for inducible exon-skipping in P19 cells, one of which is adjacent to a UAGG splicing silencer in the CI cassette. Whether the sequence requirements for inducible exon-skipping are more complex than these studies indicate or whether the differences in cell types or assays can account for seemingly different requirements is unknown.
In efforts to extend the results of the RNA motif analysis, we used biochemical approaches to attempt to visualize any changes in the proteins that bind directly to the identified UAGG motifs. In UV crosslinking experiments, the binding of hnRNP A1 to UAGG motifs was found to increase in nuclear extracts from the excited versus resting cells (2-fold increase), and this was verified by affinity selection and immunoblotting. In contrast, no change in the binding of splicing factors ASF/SF2 and TRA2 to their respective RNA substrates was observed in parallel reactions using these nuclear extracts. In addition, the cell-permeable PKA inhibitor H89, which was shown to block the excitation effects at the level of splicing, was also found to eliminate the increase in hnRNP A1 crosslinking in nuclear extracts that were excited in the presence of this compound. Taken together with the UAGG motif analysis, these results are in agreement with the prediction that signal transduction pathways communicate with splicing regulatory factors to cause induced effects on splicing, although at this point, many questions remain about how this communication is relayed to the nucleus, and exactly how the hnRNP A1 polypeptide is affected by the induction. Previous studies have documented an increase in the cytoplasmic distribution of A1 together with specific changes in the phosphorylation pattern upon osmotic stress [19,20]. Despite the apparent differences in the nature of the changes in hnRNP A1 during osmotic stress versus neuronal excitation, these previous studies and the results shown here, reinforce the idea that A1 is adept at responding to environmental signals.
NMDA receptors are multisubunit protein complexes concentrated in the postsynaptic membrane where they regulate calcium influx into neurons in response to the binding of glutamate and glycine. Calcium influx through NMDA receptors is believed to trigger acute biochemical changes leading to long-term changes in synaptic strength, or plasticity [36]. In this study, we have taken advantage of the cortical culture system to explore the potential role of NMDA receptors in the CI cassette exon-skipping response to neuronal excitation. If the excitation regimen used here to induce the CI cassette exon-skipping response involves calcium influx through NMDA receptors, we would expect antagonists that specifically block the NMDA calcium channel to weaken the excitation effects on splicing. Consistent with this idea, we observed a substantial, dose-dependent block in the response of this exon-skipping event when the cultures were treated with the antagonists AP5 or MK801. In contrast, when bicuculline was applied as an antagonist of the GABAA receptor, the exon-skipping response was only slightly weakened. The complete inhibition of the CI cassette exon-skipping response by KN93 is consistent with a role for calcium-mediated signaling in these effects. The effects shown here for KN93 in cortical neurons are in good agreement with previous studies from the Black laboratory, which have demonstrated a role, specifically, for CaMKIV in the exon-skipping response to excitation in GH3 cells and in cerebellum tissue from mouse CaMKIV knockout lines [12,34]. Because treatment with AP5 or MK801 largely, but not completely, blocks the CI cassette exon-skipping response to excitation, it is a viable possibility that calcium influx into neurons via AMPA receptors and/or L-type calcium channels might also be involved.
Modular changes in a variety of proteins involved in synaptic function have been documented at the level of alternative splicing. A large-scale analysis of splicing pattern changes in the neocortex of Nova2 knockout mice has revealed that approximately 34 exons with splicing defects correspond to transcripts that encode synapse-related proteins [37,38]. Of these, the CI cassette exon was found to undergo increased exon skipping (∼45% increase) in the Nova2 knockout mice, suggesting a role for Nova2 as a positive regulator of this exon in the mouse forebrain. In addition, NAPOR/CUGBP2, hnRNP H, and to a lesser extent ASF/SF2 and SC35, have been shown to function as positive regulators of the CI cassette exon at the level of splicing [29,39]. Along with the splicing silencing features of this control mechanism, a significant amount of combinatorial control of the CI cassette exon is indicated. Although it is not known how the nuclear levels of these additional factors are affected by neuronal excitation, an assumption of the present study is that selective alteration(s) to the balance of enhancement versus silencing underlies the net changes observed in the splicing patterns.
To a first approximation, our results indicate that NMDA receptors are involved in the CI cassette exon-skipping response to KCl-induced depolarization in these cultures. This raises questions about the potential biological role of this exon-skipping response. The CI cassette exon itself has been shown to be important for localization of the NMDA R1 subunit of the receptor at the plasma membrane, and the role of this exon in intracellular signaling has been documented [40,41]. Previous studies have shown that NMDA receptor recruitment to synaptic sites can be substantially affected by neuronal activity. Rao and Craig found increased accumulation of NMDA receptors at synaptic sites in hippocampal cultures when neuronal excitation was chronically blocked by treatment with AP5, although effects at the level of splicing were not detected [42]. In another study, the Ehlers laboratory showed that trafficking of NMDA receptors to synapses during activity blockade in cortical cultures was associated with spliced variants of the NMDA R1 receptor that contain the C2′ exon, but not the alternative exon, C2 [43].
The synaptic homeostasis model holds that, in order to stabilize excitability, neurons tend to downwardly adjust their synaptic strengths under conditions of chronic excitation, or upwardly adjust them under conditions of activity blockade [44]. This model, together with the results shown here, suggests that alternative splicing of the CI cassette exon might be involved in strategies by which neurons mount protective responses to an overload or lack of environmental stimuli. In agreement with the results shown here, a chronic KCl-induced depolarization of cortical cultures was shown to accompany a decrease in the NR1–1 (+CI cassette), and an increase in the NR1–2 (−CI cassette), mRNA isoform of the NMDA R1 receptor [45]. We have no direct evidence that a 2-fold increase in exon skipping of the CI cassette exon would cause a meaningful change in the number of NMDA receptors at the synapse. Nonetheless, it is apparent from other studies that subtle changes at the level of splicing can bring about important changes at the level of protein expression. In mice, a splicing defect in the sodium channel gene, Scn8a, was shown to produce viable animals with only 10% of correctly spliced mRNA in one genetic background, whereas a reduction to 5% of the correctly spliced mRNA in a different genetic background was lethal [46]. In this case, the cause of lethality was shown to be due to a mutation in the modifier gene, Scnm1. Furthermore, rescue from lethality was achieved by transgenic expression of a wild-type form of Scnm1, which brought about a 5% increase in the correctly spliced Scn8a mRNA. In another example, a bifunctional antisense oligonucleotide was introduced into fibroblasts derived from spinal muscular atrophy patients in an attempt to reprogram the splicing of exon 7 of SMN2 pre-mRNA. The SMN (survival of motor neurons) protein is a key component of nuclear gems, which are required for the biogenesis of small nuclear ribonucleoproteins. In the presence of the bifunctional oligonucleotide, exon 7 inclusion increased from 57% to 84%, and the percentage of gem-positive nuclei increased accordingly from 2%–3% to 13% [47]. Thus, subtle changes in splicing can be biologically important.
In this study, E9 of the GABAA receptor γ2 subunit transcript showed an increase in exon inclusion in response to neuronal excitation, clearly opposite to that of the CI cassette. Based on the distinctive features of these two splicing control mechanisms at the biochemical level (E9 involves silencing polypyrimidine tract binding protein), the opposite responses to excitation may be due to modifications of splicing factors that are unique for each exon. Alternatively, hnRNP A1 may be involved indirectly in the regulation of E9 inclusion. At the level of cellular function, GABAA receptors are the principal mediators of inhibitory neurotransmission, and the γ2 subunit plays important roles in clustering of GABAA receptors and the scaffold protein gephyrin at postsynaptic sites [48,49]. Although the E9 region was originally reported to confer ethanol sensitivity to the GABAA receptor [50,51], no discrete function for this region of the protein has been revealed in transgenic mice expressing the γ2S (−E9), but not γ2L (+E9) subunit isoform [52]. Thus, the possibility that the changes in splicing we observe here for GABAA receptor γ2 could be related to the adaptive response of neurons to hyperstimulation is purely speculative.
Future work will be needed to describe the full spectrum of exons that respond to neuronal excitation genome-wide, and the splicing factors and signaling pathways involved. Although this silencing motif code is relatively rare, well over 100 human exons contain two or more UAGGs that could potentially confer responsiveness in neurons. This also raises the question of whether the failure of cells to reverse acute splicing responses to environmental stimuli or stress may play roles in the misregulation of splicing that contributes to disease. It will also be of interest to explore the relationships between induced splicing effects and the function and distribution of relevant neurotransmitter receptors at the synapse.
Fluorescent reporters: CaMKII_EYFP reporters were constructed by inserting the open reading frame of EYFP and poly(A) site downstream of the 8.5 kb SalI-NotI fragment of the mouse CaMKII promoter [26] in a pBlueScript (pBS) plasmid. Promoter deletions were obtained by cleavage with FspI as shown in Figure 2. CaMKII-Cre promoter constructs were generated from corresponding Cre fusions, which were the generous gift of Joe Tsien and Zhenzhong Cui. Construction of Gfa2-DsRed involved inserting a 2.2-kb EcoRI fragment with the Gfa2 (glial fibrillary acidic protein) promoter upstream of the coding sequence of DsRed2 (red fluorescent protein) and downstream SV40 poly(A) site. The latter fragment was amplified by PCR from pDsRed2 (BD Biosciences, San Diego, California, United States).
Splicing reporters: Wild-type (wt0) and mutant CI cassette splicing reporters originally described in [29] were subcloned into promoter-specific plasmids CaMKII_22 or Gfa2. DIP13 splicing reporters were generated by insertion of the human DIP13 beta exon 5 and flanking introns into the NdeI and XbaI sites of the SIRT1a plasmid, followed by transfer of the HindIII-NotI fragment containing the complete splicing reporter sequence downstream of the CaMKII promoter. The exon 5 insert was synthesized in two parts, as two double-stranded deoxyoligonucleotides, and these were ligated together via complementary EcoRI cohesive ends. Site-directed mutations were generated with the QuikChange Site-Directed Mutagenesis Kit (Stratagene, La Jolla, California, United States) Sequences of all splicing reporters were verified by DNA sequencing. Constructs pBS_E18, and pBS-M3_E18 were generated by PCR from the corresponding full-length splicing reporters and cloned into the HindIII site of pBluescript vectors. All inserts were verified by DNA sequencing.
Approximately one dozen rat embryonic day 17 cerebral cortex tissues (Zivic Miller, Pittsburgh, Pennsylvania, United States) were treated with trypsin (0.25% trypsin, w/v, in 0.1% glucose, 2 mM L-glutamine and 20 mM HEPES in MEM) at 35 °C for 20 min. Cells were dissociated by passing the treated samples five to six times through an 18-gauge needle. The cell suspension was washed twice with ice-cold growth medium (0.5 mM L-glutamine, 2% B27 supplement in Neurabasal Medium; Invitrogen, Carlsbad, California, United States), and plated in poly-D-lysine–coated 6-well plates (BD Biosciences) at a density of 4–6 × 105 cells/well. For antibody staining, cells were plated on poly-D-lysine–coated coverslips. L-glutamic acid (25 μM final concentration) was added to the growth medium for the first 4 d of culture. Cultures were maintained at 37 °C with 5% carbon dioxide. To induce depolarization, KCl was added directly to the growth medium at the indicated concentrations. Inhibitor compounds were obtained from Sigma (bicuculline; St. Louis, Missouri, United States), or Calbiochem (H89, KN93, MK801, AP5, and KT5720; San Diego, California, United States).
For immunostaining, cortical cultures on coverslips were fixed with 3.7% formaldehyde in PBS for 7 min at room temperature, and permeabilized with 0.1% TritonX-100 in phosphate buffered saline (PBS) for 1 min at −20 °C followed by incubation for 5 min at 4 °C. Coverslips were washed thoroughly with PBS, and blocked for 2 h at room temperature in Blocking Buffer (1% BSA, 0.1% gelatin in PBS) containing 5% normal goat serum. The neuron-specific mouse anti-NeuN (1:100 dilution) or glial-specific rabbit anti-GFAP (1:1,000 dilution) antibodies were applied to coverslips and incubated overnight at 4 °C. After three washes in incubation buffer at room temperature, TRITC- or FITC-labeled goat anti-mouse or goat anti-rabbit (Chemicon, Temecula, California, United States) secondary antibodies (1:1,000 dilution) were applied. After washing in Blocking Buffer (three times for 5 min each) at room temperature, coverslips were dried and mounted on slides in 10% MOWIOL 4–88 (EMD Biosciences, San Diego, California, United States), 25% glycerol, 2.5% DABCO (Sigma), and 0.1 M Tris-HCl (pH 8.5). The number of cells with EYFP and TRITC fluorescence, or the number of cells with DsRed and FITC fluorescence, were scored by confocal microscopy (Radiance 2000; Bio-Rad, Hercules, California, United States) with FITC/EYFP (460–500 nm) or TRITC/DsRed (530–550 nm) excitation filters.
Plasmid DNA (1.25 μg) was transfected into cortical cultures using 3 μl of Lipofectamine 2000 (Invitrogen) per well of a 6-well plate. Plasmid DNA and Lipofectamine were each diluted to a final volume of 100 μl with OPTI-MEM, then combined and incubated for 20 min at room temperature. Cultures were prepared for transfection by washing once with OPTI-MEM, followed by addition of 2-ml fresh OPTI-MEM and transfection mixture to each well. Medium was replaced with fresh growth medium after 5 h.
For splicing pattern measurements, cells were harvested 24–48 h post-transfection using TRIZOL (Invitrogen). Splicing patterns were measured in triplicate by RT-PCR as described [39]. Reverse transcription (RT) reactions (20-μl total volume) contained M-MLV reverse transcriptase (Invitrogen), 2-μg RNA sample, and 0.5-μg random hexanucleotide primers (Promega). PCR reactions (10-μl volumes) contained 0.2 μM specific primers, 2 units of Taq DNA polymerase (Promega, Madison, Wisconsin, United States), 1/20th of the volume of the RT reaction, 0.2 mM dNTPs, and 1-μCi of [α-32P-dCTP]. Amplification was for 25 cycles. Primers were designed to amplify exon-included and -skipped mRNAs in each sample. Primers NR1 3021 (5′) ATGCCCGTAGGAAGCAGATGC (3′) and NR1 3255 (5′) CGTCGCGGCAGCACTGTGTC (3′) were used amplify endogenous C1 cassette mRNAs. Primers rp1 (5′) GTATGGTACCCTGCACTATTTTGTG (3′) and rp2 (5′) TTGGATCGTTGCTGATCTGGGACG (3′) were used to amplify the GABAA receptor γ2 mRNAs. CaMKII- and Gfa2-specific mRNAs were amplified with promoter-specific upstream primers: CaMKII, (5′) GCACGGGCAGGCGAGTGG (3′); Gfa2 (5′) TTGGAGAGGAGACGCATCACCTCC (3′) together with the gene-specific downstream primer. PCR products were resolved on 6% polyacrylamide/7 M urea gels. Gel images were captured with a BAS-2500 Phosphorimager (Fuji Medical Systems, Roselle, Illinois, United States), and quantitation determined with Science 2003 ImageGauge 4.0 software.
Scaled-down nuclear extracts were prepared essentially as described for rat cortical tissue [53]. Cortical cultures (five 10-cm dishes equivalent to 2 × 107 cells total) were harvested using plastic cell lifters (Corning, Corning, New York, United States) in 1× PBS (1 ml/dish). After centrifugation, cell pellets were resuspended in 4 ml of ice-cold Buffer I (10 mM Tris-HCl [pH 8.0], 0.32 M sucrose, 3 mM CaCl2, 2 mM Mg-acetate, 0.1 mM EDTA, 0.5% NP40, 1 mM DTT, and 1× protease inhibitor cocktail). Cell suspensions were transferred to pre-chilled 15-ml Wheaton glass homogenizers, and cells were broken on ice with pestle A (tight fitting) three to five times every 10 min for a total of 30 min. Cell lysates were mixed with 4 ml of Buffer II (10 mM Tris-HCl [pH 8.0], 2 M sucrose, 5 mM Mg-acetate, 0.1 mM EDTA, 1 mM DTT, and 1× protease inhibitor cocktail), and layered on top of 3.5 ml of Buffer II in 12-ml ultracentrifuge tubes. Samples were centrifuged in a SW41 rotor at 30,000 g for 45 min at 4 °C. Nuclear pellets were rinsed twice with 0.5 ml of Buffer C (20 mM Hepes [pH 7.6], 20 mM KCl, 1.5 mM MgCl2, 25% v/v glycerol, 0.5 mM DTT, 0.2 mM EDTA), resuspended in 50-μl Buffer C plus 0.25 M KCl, and transferred to 1.5-ml Eppendorf tubes. The concentration of nuclei in each sample was determined by trypan blue staining and adjusted with buffer to equal concentrations. Nuclear suspensions were incubated for 30 min on ice with occasional mixing, then centrifuged at 14,000 rpm for 10 min at 4 °C. Nuclear extracts (50 μl volumes) were collected and used for UV crosslinking experiments. Protein concentrations were typically in the range of 0.5 to 0.75 mg/ml as determined by Bradford assay.
Short RNA substrates (22-mers) were synthesized by Dharmacon and 5′ end labeled with γ-32P-ATP and T4 polynucleotide kinase. Longer RNAs were internally labeled with α32P-UTP during in vitro transcription using T7 RNA polymerase (Stratagene). RNA-protein complexes were assembled for 5 min at 30 °C in 12.5-μl reactions by combining labeled RNA substrates (100,000 disintegrations/min [dpm]) and nuclear extract (3.5 μl) under splicing conditions (20 mM Hepes [pH 7.5], 1.5 mM ATP, 5 mM creatine phosphate, 2 mM MgCl2, and 4.5 μl of 4.5 λ Buffer (25% glycerol v/v, 0.02 M Hepes [pH 7.5]). Samples were irradiated on ice with 1.8 joules at 365-nm wavelength. Samples were positioned at a distance of 8 cm from the UV light source. After adding SDS-PAGE loading buffer (volume), samples were boiled 5 min at 95 °C and resolved by electrophoresis on SDS-PAGE (10% polyacrylamide separating gels). Gels were fixed in 45% methanol, 9% acetic acid for 2 h, and dried. Radioactive images were captured as described above. For affinity selection experiments, the CI cassette exon (variant E18) was cloned upstream of the M3 hairpin and synthesized by in vitro transcription by T3 RNA polymerase. Transcripts were purified on Sephadex G50–150 spin columns and ethanol precipitated. MS2-MBP was purified as described [30].
|
10.1371/journal.pntd.0001247 | Glucose Starvation Boosts Entamoeba histolytica Virulence | The unicellular parasite, Entamoeba histolytica, is exposed to numerous adverse conditions, such as nutrient deprivation, during its life cycle stages in the human host. In the present study, we examined whether the parasite virulence could be influenced by glucose starvation (GS). The migratory behaviour of the parasite and its capability to kill mammalian cells and to lyse erythrocytes is strongly enhanced following GS. In order to gain insights into the mechanism underlying the GS boosting effects on virulence, we analyzed differences in protein expression levels in control and glucose-starved trophozoites, by quantitative proteomic analysis. We observed that upstream regulatory element 3-binding protein (URE3-BP), a transcription factor that modulates E.histolytica virulence, and the lysine-rich protein 1 (KRiP1) which is induced during liver abscess development, are upregulated by GS. We also analyzed E. histolytica membrane fractions and noticed that the Gal/GalNAc lectin light subunit LgL1 is up-regulated by GS. Surprisingly, amoebapore A (Ap-A) and cysteine proteinase A5 (CP-A5), two important E. histolytica virulence factors, were strongly down-regulated by GS. While the boosting effect of GS on E. histolytica virulence was conserved in strains silenced for Ap-A and CP-A5, it was lost in LgL1 and in KRiP1 down-regulated strains. These data emphasize the unexpected role of GS in the modulation of E.histolytica virulence and the involvement of KRiP1 and Lgl1 in this phenomenon.
| During infection, pathogens are exposed to different environmental stresses that are mostly the consequence of the host immune defense. The most studied of these environmental stresses are the response of pathogens to nitric oxide and to hydrogen peroxide, both produced by phagocytes. In contrast, the overall knowledge about the response of pathogens to metabolic stresses is scanty. Amebiasis is caused by the unicellular protozoan parasite Entamoeba histolytica, and has a worldwide distribution with substantial morbidity and mortality. During its journey in the host, the parasite is exposed to the host immune system and to variations in nutrient availability due to the host nutrition status and the competition with the bacterial flora of the large intestine. How E. histolytica responds to glucose starvation (GS) has never been investigated. Here, the authors report that the parasite virulence is boosted by GS. Paradoxically, two well accepted virulence factors, the amoebapore A and the cysteine protease A5 are less abundant in the glucose-starved parasites. This Accordingly, these proteins are not required for the boosting of the E. histolytica virulence, in contrast to KRiP1 and LgL1 that seem to be involved in this phenomenon.
| Amebiasis is a parasitic infection that is caused by the unicellular protozoa, Entamoeba histolytica. The disease has a worldwide distribution with substantial morbidity and mortality, and is one of the three most common causes of death from parasitic disease [1]. The main clinical manifestations of amebiasis are colitis and liver abscesses. Pathogenesis of E. histolytica involves adherence, penetration into host tissues, and destruction of host cells. These processes are mediated by the ameba key virulence factors galactose/N-acetylgalactosamine (Gal/GalNAc) lectin, amoebapore (AP) and cysteine proteinases (CP). Host cell destruction is initiated upon trophozoites binding of the target cells. An important molecule involved in this process is the galactose/N-acetylgalactosamine-inhibitable lectin [2]. This molecule is composed of two subunits; the 170 kDa heavy chain, which is responsible for the cell and carbohydrate binding activity of the lectin complex, and the 31–35 kDa light chain that has a structural role and participates in membrane anchoring of the complex [3]. It is believed that adherence of the parasite to the host gut cells is followed by the release of APs, which are a family of at least three small peptides capable of forming pores in lipid bilayers [4]. Other factors that play an important role in the Entamoeba pathogenesis are the CPs. These enzymes are released by the parasite, to disrupt the intestinal mucus and the epithelial barrier and to facilitate the tissue penetration by the trophozoites [5], [6]. The life cycle of the parasite consists of two stages: the infective cyst and the invasive trophozoite. During its progression through its life cycle in the host, the parasite is exposed to different environmental stresses which are the direct consequence of the host immune defence, or metabolic modifications and changes in the bacterial intestinal flora [7]. Whereas the physiological and molecular changes in E. histolytica following their exposure to oxidative and nitrosative stress(es) [8], [9], [10], [11], heat shock [12], [13], and bacterial flora [14], [15], [16] have been well investigated over the past ten years, information about the effects of metabolic stress in this parasite is lacking. E. histolytica relies solely on glycolysis and fermentation and lacks the tricarboxylic acid cycle and the mitochondrial electron chain reactions. Energy is mainly obtained from glucose fermentation, producing carbon dioxide, acetate and ethanol. Glucose starvation (GS) is a widely studied metabolic stresses in pathogens. It has been investigated in the malaria parasite Plasmodium falciparum. Interestingly, the PfEMP (var) genes, key components in malaria pathogenesis, account among the genes up-regulated by GS [17]. Recently, we found that in E. histolytica, GS leads to the accumulation of the glycolytic enzyme enolase in the nucleus and to the inhibition of the DNA and tRNA methyltransferase 2 (Dnmt2) nuclear activity [18]. In addition, GS triggers the in vitro differentiation of Entamoeba invadens trophozoites into cysts [19], a finding potentially relevant for E. histolytica. Here, we describe data on the effects of GS on the physiology of the parasite. We report that GS is a positive regulator of E. histolytica virulence. To the best of our knowledge, this is the first evidence that supports a role of a metabolic stressor in the modulation of E.histolytica virulence.
Trophozoites of the E. histolytica strain HM1:IMSS were grown under axenic conditions in Diamond's TYI-S-33 medium at 37°C. Trophozoites in the exponential phase of growth were used in all experiments. For the GS assays, trophozoites were washed three times with PBS, and then incubated for 12 hours in glucose-free Diamond's TYI-S-33 medium. Recovery was done by adding 1% glucose to the GS parasites for an additional 12 hours in the same medium.
For the construction of the antisense plasmid, KRiP1 was amplified by PCR from genomic DNA using the primers KRIP S and KRIP AS (Table1). The resulting 1500 bp PCR product was cloned into the pGEM-T Easy vector (Promega) to give pGEM-KRiP. For the construction of the control sense KRiP1 plasmid, KRiP was amplified from pGEM-KRiP with the primers KRiP Bgll II and KRIP AS and subcloned into the pGEM-T Easy vector (Promega) to give pGEM-sense KRiP. Next, BglII/NotI-digested pGEM-KRiP and pGEM-sense KRiP were subcloned into BglII/NotI-digested pSA8 plasmid [20]. The resulting plasmids contain the complete coding region of KRiP1 in the antisense or in the sense orientation, respectively, between 5′ and 3′ untranslated regions of the E.histolytica gene coding for ribosomal protein RP-L21 [20]–[21].
The rate of cultured HeLa cell monolayers destruction by trophozoites that were grown in either control or glucose-free medium was determined using a previously described protocol [22]. Briefly, E. histolytica trophozoites (2.5×105 or 105 per well) were incubated with HeLa cell monolayers in 24-well tissue culture plates at 37°C for 60 minutes. The incubation was stopped by placing the plates on ice and unattached trophozoites were removed by washing the plates with cold Phosphate Buffer Saline (PBS). HeLa cells remaining attached to the plates were stained with methylene blue (0.1% in 0.1 M borate buffer, pH 8.7). The dye was extracted from the stained cells by 0.1 M HCl, and its color intensity was measured spectrophotometrically at OD660.
The hemolytic activity of trophozoites that were grown in either control or glucose-free medium was determined using a previously described protocol [22]. Briefly, Human red blood cells were collected in heparin and then washed three times in PBS. The assay was performed by mixing 2.5×105 trophozoites that had been grown in either control or glucose-free medium with 5×108 red blood cells in 1 ml PBS at 37°C for 60 min. Samples of these mixtures were taken at 30 and 60 minutes. The cell suspension was rapidly sedimented (6000 rpm for 5 seconds), and the amount of hemoglobin in the supernatant was determined spectrophotemetrically at OD570.
The adhesion of trophozoites that were grown in either control or glucose-free medium to an HeLa cell monolayers was measured using a previously described protocol [23]. Briefly, trophozoites (2×105) were added to wells that contained fixed HeLa monolayers in 1 ml of Dulbecco's modified Eagle's medium (DMEM) without serum, and incubated at 37°C for 30 minutes. The number of adherent trophozoites was determined by counting under a light microscope the trophozoites that remained attached to the HeLa cells after gentle decanting (twice) of the non-adherent trophozoites with warmed (37°C) DMEM.
Transwell migration assays were performed in 5 mm transwell inserts (8 µm pore size Costar) suspended by the outer rim within individual wells of 24-well plates using a previously described protocol [24]. Briefly, E. histolytica trophozoites that were grown in either control or glucose-free medium, were first washed three times in Diamond's TYI-S-33 medium without serum, and then suspended at a concentration of 2×105 trophozoites/ml in serum-free medium. A 500-µl aliquot of the suspension was then loaded into the upper chamber of the transwell inserts, which were then placed in anaerobic bags (Mitsubishi Gas Chemical Company, Inc., Tokyo, Japan), and incubated at 37°C for three hours. At the end of the incubation, the inserts and media were removed, and trophozoite migration was determined by counting the number of trophozoites that were attached to the bottom of the well.
Trophozoites (1×106) were exposed to a temperature of 42°C or to 2.5 mM in hydrogen peroxide (H2O2). After 1, 1.5, 2 and 2.5 hours of incubation, an aliquot of the culture was stained with eosin (0.1% final concentration), and the number of living trophozoites were counted in a counting chamber under a light microscope (Bausch and Lomb).
Entamoeba histolytica trophozoites (107) that were grown in either control or glucose-free medium for 12 hours were harvested and lysed in 300 µL 1% octyl β-D-glucopyranoside (Sigma-Aldrich) on ice for 20 minutes with regular shaking. Crude extracts were centrifuged at 13000 rpm at 4°C for 10 minutes, supernatants were removed, and the protein amounts of in the lysate were quantified using the Bradford method [25]. Usually 4 mg of protein were recovered and 100 µg of total cell lysate was used for the labeling in two independent experiments. Urea (8M) was then added to the proteins followed by their reduction with 3 mM DTT at 60°C for 30 minutes. Trypsinization of proteins was carried out overnight at 37°C in 10 mM ammonium bicarbonate containing modified trypsin (Promega) at a 1∶50 enzyme-to-substrate ratio, overnight at 37°C. A second step of trypsinization was done by adding another portion of trypsin and incubating at 37°C for 4 hr. The tryptic peptides were desalted using C18 tips (Harvard Inc.), dried and resuspended in 50 mM Hepes (pH 6.4). Labeling by dimethylation was done in the presence of 100 mM NaCBH3 (Sterogene). Peptides from the control samples were labeled with light formaldehyde (35% Frutarom) and peptides from glucose starvation samples were labeled with heavy formaldehyde at a final concentration of 200 mM (20% w/w, Cambridge Isotope laboratories). After 1 hour incubation at room temperature, the pH was raised to 8 and the reactions were further incubated for 1 hour. Neutralization was done with 25 mM ammonium bicarbonate for 30 minutes and equal amounts of the “light” and “heavy” peptides were mixed, cleaned on C18 tips and re-suspended in 0.1% formic acid. The combined labeled peptides were separated in an on-line two-dimensional chromatography (MuDPiT). First the peptides were loaded on a 15 mm of BioX-SCX column (LC Packing) and eluted with 10 salt steps of 0, 30, 60, 80,100, 120, 160, 200, 300, 500 mM ammonium formate in 5% ACN and 0.1% formic acid, pH 3. The eluted peptides were further resolved by capillary reverse-phase chromatography (20 cm fused silica capillaries, J&W self-packed). The peptides were eluted using a 125 min gradient (5% to 40% acetonitrile containing 0.1% formic acid) followed by a wash step of 95% acetonitrile for 15 min. The flow rate was about 0.25 µl/min and the peptides were analysed using an Orbitrap mass spectrometer (Thermo-Fischer). Protein identification was performed at the Smoler Proteomics Center, Technion (http://biology.technion.ac.il/proteomics/index.htm). Mass spectrometry (MS) was done in a positive mode using repetitively full MS scan followed by collision induced dissociation (CID) of the 7 most dominant ions selected from the first MS scan. The MS data were analyzed and compared using the Sequest 3.31 (Thermo). Identification of proteins with significant changes in their abundance level was based on searching with the ProtScore program, with a cut-off at 2.0 fold, searching in the E. histolytica genome NZ_AAFB00000000 part of the NR-NCBI data base.
Nuclear and cytoplasmic fractions of E.histolytica trophozoites were prepared using a previously described protocol [21]. Proteins were separated on 12% polyacrylamide SDS-PAGE gels, and then transferred to a nitrocellulose membrane. Membranes were stained with Ponceau S (Sigma) to verify the efficiency of the transfer. The blots were then blocked with 3% skim milk in PBS, and then incubated with a monoclonal antibody against actin (1∶1000, Santa-Cruz), a polyclonal antibody against E. histolytica amoebapore A (Ap-A) (1∶500) (kindly provided by Prof. Mirelman, Weizmann Institute, Israel), and a polyclonal antibody against E. histolytica KRiP1 (1∶1000) raised in this work by three immunizations of two rabbits with a unique synthetized peptide (amino acids 82 to 108, EQSTKAPAGDKVINLD). After incubation with the first antibody, the blots were then incubated with a secondary antibody (1∶10000) (horseradish peroxide-conjugated goat anti-rabbit) antibody, Jackson ImmunoResearch), and immune complexes were detected by enhanced chemiluminescence.
Trophozoites were harvested by centrifugation, washed 3 times in PBS, resuspended in 300 µl 100 mM NaCl and subjected to four cycles of freezing at -70°C and thawing. The samples were centrifuged at 800 g for 15 minutes and supernatants were collected and centrifuged at 10000 g for 1 hour. The Pellets were resuspended in 15 µl PBS.
Total RNA was extracted from trophozoites using a TRI-reagent solution (Sigma). Reverse transcription was performed with the EZ-First Strand cDNA Synthesis Kit for RT-PCR (Biological Industries, Beit Haemek, Israel), according to the manufacturer's instructions. The primers that were used to amplify Ap-A, LgL1, CP-A5, URE3-BP, enolase encoding genes and rDNA are listed in Table 1. The PCR products were resolved on 1% agarose gels, and then stained with 0.5 µg/ml ethidium bromide (aMReSCO). Densitometry analysis of the stained PCR products was done using the TINA software (http://www.tina-vision.net).
Total RNA was prepared using the RNA isolation kit TRI Reagent (Sigma). RNA (10 µg) was size-fractionated on a 4% polyacrylamide denaturing gels containing 8 M urea and subsequently blotted to a GeneScreen membranes (NEN Bioproducts, Boston, MA). Hybridizations under stringent conditions were carried out overnight at 65°C in hybridization buffer (0.5 M Na-phosphate (pH 7.2), 7% SDS, 1 mM EDTA and 50 µg per ml hybridization buffer of salmon sperm DNA with the following DNA probes (0.004 µg/ml): Ap-A, CP-A5, URE3-BP and HSP 70. The membrane was then washed at 65°C for 20 min with washing buffer 1 (5% SDS, 40 mM Na-phosphate (pH 7.2), and 1 mM EDTA), followed by three washes of 30 min each wash at 65°C with washing buffer 2 (1% SDS, 40 mM Na-phosphate (pH 7.2), and 1 mM EDTA). Probes were randomly labeled with P32 dCTP (Amersham) using a DNA labeling kit (Biological Industries).
Animal handling and experimentation were conducted according to the European Union and the Pasteur Institute approved protocols. Four-week-old male Syrian golden hamsters (Mesocricetus auratus), with a weight ranging from 90 to 100 g, were inoculated by the intraportal route [26], [27] with 106 sense or anti-sense KRiP1 trophozoites. Two independent experiments were performed, with 3 animals for each amoeba type per experiment; thus, a total of 12 hamsters were used for this evaluation. (Six animals by experiment and conditions were used; in total, 12 hamsters were included in this evaluation. Treatment of animals and surgical procedures were done according to published methods. From 7 days post-infection, hamsters were sacrificed, livers removed). Livers were isolated 7 days after infection and treated for histological analysis. Serial (five-micron) lobe sections were stained with Harris' haematoxylin, periodic acid-Schiff reagent. Immunohistochemistry was performed on sections from different livers using an anti-Gal/GalNAc lectin monoclonal antibody CD6 (a total of 31 tissue sections were analysed).
Entamoeba histolytica trophozoites were glucose-starved for a period of twelve hours and the effect of glucose starvation (GS) on their viability was determined (Fig. 1A). Up to a period of six hours, GS has no significant effect on the viability of the trophozoites. Between 6 hours and 12 hours, a slightly lower viability (17%) was observed for the glucose-starved trophozoites when compared to the viability of parasites cultivated in the presence of glucose. A longer GS (more than 24 hours) led to significant parasite death [18]. Consequently, we used 12 hours of GS for all subsequent experiments.
The induction of stress responses is important for the adaption of cells to environmental changes following stress [28]. Consequently, we examined the resistance of glucose-starved trophozoites to heat shock and to oxidative stress. No significant difference in the resistance of starved and control trophozoites to heat shock was observed (Fig. 1B). In contrast, the glucose-starved parasites were slightly more sensitive to oxidative stress when compared to the sensitivity of control trophozoites grown in presence of glucose (Fig. 1C). In addition, the expression of heat shock protein 70 (HSP70) expression was not up-regulated by GS (Table S1). These results indicate that GS does not induce a protective response against distinct stresses.
The effect of GS on E. histolytica virulence has never been investigated. Since E. histolytica trophozoites have the capacity to lyse human erythrocytes [29], we compared the capacity of E. histolytica to lyse human erythrocytes [29] (hemolytic activity) of control and glucose-starved trophozoites as one of the measures of their virulence. We found that the hemolytic activity of glucose-starved trophozoites was nearly twice as high (1.7 fold) as that of the control trophozoites (Fig. 2A). Since the parasite is also able to disrupts monolayers of cultured mammalian cells, such as Baby Hamster Kidney (BHK) or HeLa cells [30], we compared the cytopathic activity of control and glucose-starved E. histolytica trophozoites as a second measure of virulence. We found that the cytopathic activity of glucose-starved trophozoites was significantly higher than that of the control trophozoites (Fig. 2B). Moreover, the cytopathic activity of the glucose-starved trophozoites can be restored to that of the control trophozoites upon their re-incubation in a glucose-containing culture medium for 12 hours (Fig. 2B). Since the cytopathic process is initiated by the parasite binding to its target cells [31], we then compared the ability of control and glucose-starved E .histolytica trophozoites to adhere to HeLa cells. We found that the binding of the glucose-starved trophozoites was more than twice as high as that of the controls trophozoites (Fig. 2C). During invasive amebiasis, migration is an essential process; Therefore, we compared the migration of control and glucose-starved E. histolytica trophozoites through 8 µm pores. We found that the numbers of glucose-starved trophozoites that passed through the pores was considerably greater than then number of control trophozoites (Fig. 2D). Collectively, these results indicate that GS is a positive regulator of E. histolytica virulence.
To gain insight into the mechanism by which GS enhances the virulence of E. histolytica, we compared changes in the protein composition of control and glucose-starved trophozoites, using stable isotope labeling (Table 2 and 3). In this high throughput proteomics analysis, total protein extracts from E. histolytica trophozoites grown in complete TYI media or in GS medium was compared to total lysate from E. histolytica grown in GS media (as described in Materials and methods). Proteins were identified using Mass spectrometry (MS) and identification of proteins was achieved by a minimum of three peptides. Forty-nine proteins with at least a two-fold change in their abundance were detected. Of these, 29 proteins were more abundant (Table 2) and 20 proteins were less abundant in the glucose-starved trophozoites when compared to the equivalent proteins in the controls trophozoites (Table 3). We classified the corresponding genes according to their functions using the Pfam classification (Fig. 3A). An interesting feature that we observed for glucose-starved trophozoites was a decrease in the amount of proteins that were associated with metabolism (Fig. 3A; lower panel). These proteins are mostly associated with the fatty acid pathway. The induction of the gluconeogenic pathways is a common response to GS [32]. We found that a protein that shares a short homology to glycerol-3-phosphate dehydrogenase (EHI_161020) was one of the most highly enriched in the glucose-starved trophozoites. An interesting feature of the response to GS is the increase in proteins that are involved in protein synthesis, especially the 60S and 40S ribosomal proteins (Fig. 3A; upper panel). Several other abundant proteins in the glucose-starved trophozoites may play a role in the boosting of E.histolytica virulence. One of them is the transcription factor Upstream Regulatory Element 3-Binding Protein (URE3-BP). The enrichment of URE3-BP upon GS treatment correlates with a higher amount of URE3-BP transcripts in the glucose-starved parasite compared to control trophozoites (Fig. 3B). URE3-BP is a calcium-responsive transcription factor that is known to modulate the transcription of two E. histolytica genes encoding virulence-related factors, the Gal/GalNAc lectin subunit HgL5 and ferredoxin [24]. Further proteins that may play a role in the boosting of E. histolytica virulence under GS are proteins encoded by genes up-regulated in trophozoites isolated from hamster liver abscesses [33] such as the 20 kDa antigen (EHI_056490) and the lysine-rich proteins, KRiP1 (EHI_096350) and KRiP3 (EHI_110740) (Table 2 and [33]). The increased abundance of KRiP1 observed in the proteomic analysis was confirmed by Western blot analysis (Fig. 3C). An unexpected result of the proteome analysis is the lower abundance of proteins associated with virulence in the glucose-starved E.histolytica trophozoites compared to control trophozoites (Table 3). These proteins include the key virulence factors Ap-A and CP-A5. The lower abundance of Ap-A in glucose-starved trophozoites was confirmed by Western blot analysis (Fig. 3C). This result correlates with a lower amount of Ap-A transcripts in the glucose-starved parasites compared to control trophozoites (Fig. 3B). About CP-A5, its lower abundance observed by the proteomic analysis (Table 3) correlates with a lower amount of CP-A5 transcript (Fig. 3B).
KRiP1 is one of the proteins whose abundance is the most enhanced by GS (Table 2 and Fig. 3C). Moreover, the up-regulation of KRiP1 gene expression in trophozoites isolated from hamster liver abscesses points towards an involvement of this protein in the regulation of E. histolytica virulence [33]. These observations suggest an implication of KRiP1 in the mechanism accounting for the exacerbation of amoebic virulence under GS. To test this hypothesis, the expression of KRiP1 was down-regulated using the antisense technology. For this purpose, trophozoites were transformed with a plasmid that includes the krip1 gene in the antisense orientation between the 5′ and 3′ untranslated regions (UTRs) of the E. histolytica gene coding for ribosomal protein RP-L21 (Ehg34) [20]–[21]. As a control, trophozoites were transformed with a plasmid that includes the krip1 gene in the sense orientation. It is important to note that the Ehg34 5′ UTR does not allow the translation of an mRNA that has been expressed under its control [34]. The amount of KRiP1 in trophozoites transformed with the krip1 gene sense or antisense construct was determined by Western blot analysis (Fig. 4A). The KRiP1 antisense strain contained three times less amount of KRiP1 compared with those found in the KRiP1 sense strain. No significant difference in the growth rate of KRiP1 sense and antisense trophozoites cultivated in standard TYI-33 media was observed (data not shown). In addition, no difference in the viability of KRiP1 sense and antisense transformants exposed to GS was observed (data not shown). These results indicate that KRiP1 is not essential for the parasite survival under GS.
We then compared the cytopathic activity of E. histolytica krip1 sense and antisense trophozoites cultivated in standard TYI-33 media. The cytopathic activity of E.histolytica trophozoites expressing the antisense KRiP1 vector was 25% lower compared to that of the control (Fig. 4B). Under conditions of GS, the boosting effect on the cytopathic activity was observed for the KRiP1 sense trophozoites but did not occur for the KRiP1 antisense parasites (Fig. 5). This result suggests that KRiP1 is involved in the mechanism that up-regulated the cytopathic activity of E. histolytica under GS. We further analyzed the consequences of krip1 gene expression levels on the amoebic liver abscess formation by intraportal infection of hamsters with KRiP1 sense or anti-sense trophozoites. Two independent experiments were carried out. Seven days after infection, the animals were sacrificed, morphology of liver was macroscopically observed and a histological analysis was performed. As expected, infection with trophozoites carrying the krip1 sense gene resulted in abscess development (4 of 6 animals presented well organized abscesses). Histological analysis showed that 7 of the 12 inspected lobes presented necrotic areas with parasites (Fig. 4C). In contrast, trophozoites carrying the krip1 gene antisense construct have clearly reduced their capacity to establish liver abscesses. Indeed, in 4 out of 6 hamsters, we did not observe macroscopic abscesses. From the 12 lobes examined by histology, only two presented liver lesions, whereas the other lobes presented preserved tissue architecture (Fig. 4C).
Ap-A and CP-A5 are two major E. histolytica virulence factors [35]. The reduction of their levels in glucose-starved trophozoites (Table 3) suggests that these proteins are not involved in the boosting effect of GS on the parasite's virulence. To test this hypothesis, two strains silenced for the expression of Ap-A (G3 strain) and Ap-A and CP-A5 (RB8) were studied. First, we confirmed by RT-PCR that the expression of Ap-A and CP-A5 encoding genes in the respective G3 and RB8 strains was silenced (data not shown). Then, the GS effect on the cytopathic activity of E. histolytica strains G3 and RB8 was examined (Fig. 5). We found that GS boosted the virulence of the G3 and RB8 strains indicating that Ap-A and CP-A5 are not requested for the boosting effect of GS on virulence.
E. histolytica trophozoites adhere to the host colon epithelium using the galactose/N-acetylgalactosamine (Gal/GalNAc) inhibitable lectin [36]. Monoxenic cultivation of E. histolytica with the bacterium Escherichia coli strain O55 results in down-regulation of the lectin light subunit and reduced virulence [37]. These observations suggest that the Gal/GalNAc lectin light subunits are involved in the response of the parasite to environmental changes. Therefore, and despite the fact that the light subunits were missing from the list of protein regulated by GS (Table 2 and 3), we examined more specifically their abundance in the membrane-enriched protein fraction of glucose-starved trophozoites. We observed that around three times more LgL1 was present in the protein fraction of glucose-starved trophozoites compared to the amount present in the fraction of the control strain (Fig. 3D). To determine the role of LgL1 in the boosting effect of GS on virulence, the cytopathic activity of the E. histolytica RB9 strain was studied. These parasites have been silenced for the expression of Ap-A and of LgL1 [38]. Interestingly, no boosting effect of GS on the cytopathic activity of the strain RB9 was observed (Fig. 5). These results suggest that LgL1 is involved in the mechanism that enhances the cytopathic activity of E. histolytica under GS.
Parasite survival in its host depends on its ability to react to different stress stimuli and to adapt to changing environments. Under laboratory conditions, E. histolytica is cultivated in presence of 1% glucose whereas in its human host, the parasite lives mostly in the large intestine, which is a low-glucose environment, due to the absorption of simple sugars in the small intestine [39]. We recently reported evidences that the parasite reacts to GS by modifying the cellular localization of enolase and the methylation status of its tRNAasp [18]. The results presented here indicate that GS is also a positive regulator of E. histolytica virulence.
A positive regulation of virulence by GS has also been reported in Candida albicans. In this haploid yeast, glucose depletion causes a developmental switch that allows cells to penetrate the surface of an agar medium in a process called invasive growth. This process mimics invasion of host tissue and therefore pathogenesis [40].
In many microorganisms the stress response against heat shock or starvation provides a cross-resistance to other stresses. In E.coli, glucose or nitrogen starvation lead to enhanced resistance to heat shock and oxidative stress [41]. Adaptation of C. elegans to high temperatures by the heat shock response results in a reduced sensitivity to ethanol under conditions of heat stress [42]. The mechanism acting in the cross-protection effect is the induction of heat shock proteins (HSPs) expression by the initial stress stimulus. HSPs prevent the aggregation of partially denatured proteins [43]. Our results show that GS does not protect the E. histolytica against heat shock or oxidative stress. This result is not surprising as the abundance of HSP 70 and other HSPs were not increased in glucose-starved parasites. In contrast to the lack of HSPs induction, GS has a striking effect on the abundance of other E.histolytica proteins.
The following focus of our study was directed to identify at the proteome level major E. histolytica factors modified by GS. The proteomics analysis provided several candidate(s) proteins that may be involved in the virulence boosting effect of GS.
One of the proteins whose abundance is enhanced in glucose-starved trophozoites is URE3-BP. The role of this transcription factor in the remodeling of the cell surface in response to calcium signals has been well established. Indeed, the expression of a dominant-positive mutant of URE3-BP leads to higher virulence and increased cell motility [24]; phenotypic features that we also observed in glucose-starved trophozoites. These data indicate that URE3-BP may be involved in the control of virulence exerted by GS.
Another putative regulator of the virulence under GS is KRiP1 which belongs to a group of lysine-rich proteins (KRiP and KERP) whose expression is up-regulated in virulent E.histolytica strains [33]. Some of these proteins, like KERP1, are localized to the parasite membrane and others like KRiP1 are nuclear, as we demonstrated here. Our knowledge about how these lysine-rich proteins regulate the parasite's virulence is scanty. The fact that no boosting of the cytopathic activity occurs under GS in the KRiP1 antisense transfectants suggests that this protein regulates the virulence of the parasite under stressful physiological conditions.
In an attempt to investigate the direct involvement of KRiP1 in pathogenicity, we subjected parasites modified for KRiP1 production by an antisense strategy to in vivo conditions of liver abscess formation. We demonstrate that the antisense parasites exhibit a significantly lower pathogenicity in the hamster model, indicating that KRiP1 plays a role in E. histolytica virulence in vivo.
A third candidate for the regulation of E.histolytica virulence under GS is the Gal/GalNAc lectin light subunit which is also upregulated following GS. The role of LgL1 in the virulence boosting effect of GS may be direct or the result of changes in the overall structure of the Gal/GalNAc lectin complex. In addition, the silencing of LgL1 in the RB9 strain led to the co-silencing of LgL2 and LgL3 [44]. Therefore, it is difficult at this stage to know which LgL is implicated in the GS boosting effect on E. histolytica virulence. In Saccharomyces cerevisiae, glucose sensing is achieved by two transmembrane glucose-sensing proteins, Snf3 and Rgt2 [45]. These proteins have a cytoplasmic C-terminal tail which is essential for detecting changes in glucose concentrations [46]. In addition to a role of the LgL subunits in adhesion [47], signalling in chemotaxis [48] and differentiation [19], the role of LgL in glucose sensing in E.histolytica could be a new area of the Gal/GalNAc lectin research.
An unexpected result of the proteomics analysis is the lower abundance of the virulence factors Ap-A and CP-A5 in glucose-starved E. histolytica trophozoites. Usually, in a state of energy depletion organisms synthesize virulence factors that are capable of killing host cells and degrade macromolecules to obtain sugar and energy [49], [50]. Our results suggest that AP-A and CP-A5 are not essential for the virulence boosting in glucose-starved parasites.
An interesting feature that was found in glucose-starved trophozoites is the decreased amount of proteins associated with metabolism and in particular with the fatty acid pathway. In the specific pyruvate-to-ethanol pathway in E.histolytica, acetyl-CoA is converted to acetaldehyde, which is then reduced to ethanol [51]. Reduced levels of the long chain fatty acid CoA ligase during GS may be a consequence of a mechanism by which acetyl-CoA is deviated from the fatty acid pathway into the pyruvate-to-ethanol pathway [52]. Lipid degradation as a mean to increase acetyl-CoA production was also observed in glucose-starved B. subtilis and in starved rats where fatty acids degradation could provide acetyl-CoA as substrate for the tricarboxylic acid cycle [53], [54]. Many of the ribosomal proteins were also affected by GS. These proteins are regarded as moon light proteins, thus having extra-ribosomal roles [55]. Examples for biological processes ribosomal proteins participate in include growth regulation, cell proliferation and DNA damage response [56], [57], [58]. In eukaryotic cells ribosomal proteins can auto-regulate their production, in this way synthesis of individual ribosomal protein can be controlled [59]. This mechanism may also be relevant for E. histolytica for which proteomics data provide evidence for ribosomal turnover during GS and for the accumulation of some several ribosomal proteins [60]. Alternatively, the accumulation of ribosomal components in the glucose-starved parasites may reflect their ability to immediately initiate an upshift program when the missing substrate is made available, as seen in Vibrio sp. strain S14 during carbon starvation [61]. At this stage, we cannot rule out that metabolic changes occurring in the glucose-starved parasites contribute to the modulation of their virulence.
In summary, the results of this investigation indicate that significant changes occur in the metabolism of E.histolytica exposed to GS to provide the parasite an alternative source of energy. In addition, we report for the first time that GS upregulates E. histolytica virulence and that KRiP1 and LgL1 are involved in the boosting effect. In the future, it will be interesting to determine if this specific response of the parasite to short-term glucose starvation reflects events occurring in the low-glucose environment of the large intestine.
|
10.1371/journal.pgen.1004805 | Formation of Linear Amplicons with Inverted Duplications in Leishmania Requires the MRE11 Nuclease | Extrachromosomal DNA amplification is frequent in the protozoan parasite Leishmania selected for drug resistance. The extrachromosomal amplified DNA is either circular or linear, and is formed at the level of direct or inverted homologous repeated sequences that abound in the Leishmania genome. The RAD51 recombinase plays an important role in circular amplicons formation, but the mechanism by which linear amplicons are formed is unknown. We hypothesized that the Leishmania infantum DNA repair protein MRE11 is required for linear amplicons following rearrangements at the level of inverted repeats. The purified LiMRE11 protein showed both DNA binding and exonuclease activities. Inactivation of the LiMRE11 gene led to parasites with enhanced sensitivity to DNA damaging agents. The MRE11−/− parasites had a reduced capacity to form linear amplicons after drug selection, and the reintroduction of an MRE11 allele led to parasites regaining their capacity to generate linear amplicons, but only when MRE11 had an active nuclease activity. These results highlight a novel MRE11-dependent pathway used by Leishmania to amplify portions of its genome to respond to a changing environment.
| Extrachromosomal DNA amplification is frequent in the human protozoan parasite Leishmania when challenged with drug or other stressful conditions. DNA amplicons, either circular or linear, are formed by recombination between direct or inverted repeats spread throughout the genome of the parasite. The recombinase RAD51 is involved in the formation of circular amplicons, but the mechanism by which linear amplicons are formed is still unknown in this parasite. Studies in other organisms have provided some evidence that a DNA break is required for linear amplifications, and that the DNA repair protein MRE11 can be involved in this process. In this work, we present our biochemical, cellular and molecular characterization of the Leishmania infantum MRE11 orthologue and provide evidence that this nuclease is involved in the formation of linear amplicons in Leishmania. Our results highlight a novel MRE11-dependent pathway used by Leishmania to amplify portions of its genome to respond to a changing environment.
| The protozoan parasite Leishmania is responsible for a group of diseases named leishmaniasis, affecting approximately 12 million people worldwide. No vaccine is currently available against Leishmania and treatments mainly rely on chemotherapy [1], [2]. Pentavalent antimony is the main anti-leishmanial drug although treatment failure due to resistance has been reported in most endemic regions [3]–[7]. Locus amplification is a frequent resistance mechanism allowing the parasite to modulate gene copy number and increased gene expression. Indeed, the parasite Leishmania is an early diverging eukaryotic parasite with no control of gene expression at the level of transcription initiation [8]–[10] and amplification of DNA loci is one strategy to increase the expression of resistance genes. Resistance genes can be amplified as part of extrachromosomal circular DNAs (circular amplicons) or as inverted duplications (linear amplicons) under drug pressure [11]–[16]. Gene rearrangements leading to locus amplification always occur at the level of either homologous direct or inverted repeated (IRs) sequences leading respectively to circular or linear extrachromosomal amplification [14], [15], [17]–[19]. A model for the generation of linear amplicons is shown in Figure 1.
A recent bioinformatics screen revealed that repeated sequences are widely distributed in the Leishmania genome, which is continuously being rearranged at the level of those repeated sequences. This process is adaptive as the copy number of advantageous extrachromosomal circular or linear elements increases upon selective pressure [19]. The whole genome of Leishmania is thus stochastically rearranged at the level of repeated sequences and the selection of parasite subpopulations with changes in the copy number of specific loci is used as one strategy to respond to drug pressure.
Circular or linear amplification has been observed when parasites were selected against a wide variety of drugs including the mainstay antimony [15], [20] but one drug that has proven highly useful in deciphering gene amplification mechanisms in Leishmania is the model antifolate drug methotrexate (MTX). Two loci are frequently amplified after MTX selection, one encoding the dihydrofolate reductase-thymidylate synthase (DHFR-TS) gene, the target of MTX, usually as part of circular elements [14], [21]–[23] the second encoding the pteridine reductase 1 (PTR1) gene, which is less sensitive to MTX but can reduce folates when DHFR-TS is blocked [24], [25]. The PTR1 gene is amplified as part of either circular amplicons [22], [26], [27] or linear amplicons [13], [14], [19], [28], [29].
We have recently provided mechanistic insights into the formation of circular amplicons mediated by homologous recombination between direct repeated sequences catalyzed by the RAD51 recombinase [19]. However rearrangements at IRs leading to linear amplicons were not RAD51-dependent. Studies in other organisms have provided some evidence that a DNA break is required for palindromic amplifications formed by annealing of IRs [30]–[33] and that an exonuclease activity must be recruited to perform DNA end resection after chromosomal breakage in order to allow annealing of IRs [34]–[36] (see also Figure 1). Based on these observations, we hypothesized the involvement of the nuclease MRE11 (Meiotic REcombination 11) in the generation of linear amplicons in Leishmania (Figure 1). MRE11 interacts with RAD50 and NBS1 to form the MRN complex [37], [38]. Indeed, the nuclease MRE11 is a sensor of DNA double-strand breaks in cells and is important for the DNA double-strand break repair pathway [39], [40] by homologous recombination (HR) or non-homologous end joining (NHEJ) [41]. Leishmania infantum encodes a putative MRE11 with conserved endo- and exonuclease domains as well as DNA-binding domains [41]. In this manuscript, we present our biochemical, cellular and molecular characterization of the L. infantum MRE11 orthologue and provide evidence that this nuclease is involved in the formation of linear amplicons in the parasite Leishmania.
Since the critical catalytic residues of MRE11 are conserved in Leishmania [41] and the replacement of the histidine (H) at position 217 by a tyrosine (Y) is known to abolish the nuclease activity of the human MRE11 but not its nucleic acid binding property [42], we scrutinized the amino acid alignment between the human and Leishmania sequences and found the equivalent of human H217 at position 210 of LiMRE11 (Figure 2A, upper panel and Figure S1). We therefore produced a LiMRE11 mutated at the corresponding amino acid (LiMRE11H210Y) and used a two-step affinity purification procedure to purify LiMRE11WT and LiMRE11H210Y as described in Material and Methods (Figure 2A, lower panel).
We used the electrophoretic mobility shift assay to study DNA interactions with MRE11 proteins (Figure 2B). We observed that the splayed arm (SA) and single-strand (SS) DNA structures were shifted in the presence the wild-type and mutated MRE11 protein in a dose-dependent manner while neither version of the protein were able to shift the double-strand (DS) structure in this competitive assay. The binding was quantitated and at 15 nM of either protein, 65% binding was observed with either SS and SA DNA structures while we observed only 10% binding for DS DNA (Figure 2C).
We next tested whether purified LiMRE11WT and LiMRE11H210Y displayed exonuclease activity (Figure 3A), in comparison with human MRE11WT and hMRE11H217Y proteins (Figure 3B). Our findings suggest that LiMRE11WT is enzymatically active and can perform exonucleolytic degradation with a 3′ to 5′ polarity but it is less effective than the human MRE11WT protein in cleaving DNA into smaller fragments (Figure 3A–B, lanes 1–4). As expected, LiMRE11H210Y was unable to perform DNA resection (Figure 3A, lanes 5–7), similar to its human mutated counterpart (Figure 3B, lanes 5–7). The substrate specificity was also monitored by using 25 nM of LiMRE11WT protein with DS DNA, either blunt or with 3′ or 5′ overhang DNA structures (Figure 3C). The same extensive degradation was observed with DS DNA and 5′-overhang ends (Figure 3C, lane 2 and 4) while the protein was blocked by 3′-overhang extremities (Figure 3C, lane 3). LiMRE11WT also exhibits endonuclease activity, as shown by the 13 bp band found at the bottom of the gel.
We generated a LiMRE11-GFP fusion construct that was transfected in L. infantum cells. However, we could never achieve a high copy number of the plasmid (overexpression of MRE11 can be toxic to the cell, see below) and fluorescence levels were too low for analysis. We then turned to a heterologous system to study LiMRE11 in vivo. DNA constructs encoding the fusion protein LiMRE11WT-GFP and the human counterpart hMRE11WT-GFP were transfected in human ATLD cells, which are deficient for hMRE11 [43]. After laser-induced DNA damage in these cells, we detected a localized fluorescent foci representative of the recruitment of LiMRE11WT-GFP in micro-irradiated nuclear regions (Figure 4, upper panels), similar to what was observed for the human MRE11WT-GFP fusion protein (Figure 4, bottom panels). Among 24 ATLD cells micro-irradiated, we observed a recruitment of LiMRE11 to DNA damages sites in 75% of the cases, while the human homolog was recruited in 100% of the cells. No recruitment was observed for the control GFP alone. These observations confirmed the ability of the Leishmania MRE11 protein to be recruited at DNA damage sites, in a heterologous cellular model. Altogether, these results show that LiMRE11 display similar localization properties as the human enzyme.
L. infantum MRE11 null mutant parasites were generated by replacing the entire ORF (LinJ27.1790) with genes coding for the neomycin (NEO) and hygromycin (HYG) phosphotransferases. The two resistant markers were cloned between the 5′- and 3′-MRE11 flanking regions and targeting constructs were transfected independently in two rounds by electroporation. Southern blot analysis confirmed the homologous chromosomal integration of the two antibiotic markers in the MRE11 locus (Figures 5A and 5B). Genomic DNAs of the WT and the HYG/NEO MRE11−/− null mutant were digested with XhoI, transferred onto membranes and hybridized. Hybridization with a probe recognizing the 5′UTR region of MRE11 yielded a 3 kb band in WT cells (Figure 5A and 5B-lane 1) while hybridization with a 3′UTR probe generated a 3,4 kb band as expected (Figure 5A and 5B, lane 5). In the HYG/NEO MRE11−/− strain, replacement of both MRE11 wild-type alleles by NEO and HYG led, as expected, to 4,7 kb and 4,9 kb bands respectively, with either UTR probes (Figure 5A and 5B, lanes 2 and 6).
It is standard practice to introduce episomal copies of the corresponding wild-type gene in a null mutant background to reverse a potential phenotype. However, we noticed that episomal overexpression of MRE11 as Psp72-α-PUR-α-MRE11WT in WT cells led to a growth defect (Figure S2A). This prompted us to use an alternative to generate revertants. We replaced the NEO chromosomal integrated cassette in the MRE11 null mutant by a re-expressing cassette containing either a WT or a mutated allele (H210Y) of LiMRE11 along with the PUR gene in order to generate respectively the HYG/PUR-MRE11WT and HYG/PUR-MRE11H210Y re-expressing add back strains. Hybridization of the DNAs of the add-back strains with a 5′UTR probe led, as expected, to a 3 kb band in both strains corresponding to the restoration of a WT allele at the MRE11 locus, a 4,9 kb band corresponding to the HYG chromosomal integration, and a loss of the 4,7 kb-NEO containing band (Figure 5A and 5B-lanes 3 and 4) which was replaced by the MRE11-α-PUR re-expressing cassette as supported by the hybridization of a 2,7 kb band when using a 3′UTR probe (Figure 5A and 5B-lanes 7 and 8). MRE11 expression level was assessed by quantitative real-time RT-PCR in the various cell lines generated. As expected MRE11 expression level was not detectable in the MRE11−/− null mutant while it was approximately half the level of the WT in both add back strains, consistent with one new active allele (Figure S3).
A growth defect was observed in L. infantum HYG/NEO MRE11−/− parasites compared to the WT strain. Promastigotes of the WT strain had a calculated generation time of 12 hours while the MRE11 null mutant had a generation time of 26 hours. (Figure 5C). Reintroduction of one intact WT or mutated (H210Y) allele of MRE11 into the chromosomal locus in add back strains partially rescued the growth defect with respective generation time of 15 and 22 hours.(Figure 5C).
Since the MRE11 complex is known to promote repair of DNA double-strand breaks (DSBs) [44], [45], we tested the impact of the LiMRE11 inactivation using the alkylating damaging agent methyl methanesulphonate (MMS), a compound known to induce DSBs [46]. The HYG/NEO MRE11−/− cells were significantly more sensitive to MMS compared to both WT and MRE11 add back re-expressing cells (Figure 5D). As indicated above, intriguingly, overexpressing MRE11 in WT cells led to a growth defect (Figure S2A) but also to a significant increase in MMS sensitivity (Figure S2B). However, episomal overexpression of MRE11WT in HYG/NEO MRE11−/− cells restores the growth defect and MMS susceptibility of the mutant strain to WT levels (Figures S2A and S2B).
We compared the ability of the MRE11 null mutants and WT cells to generate extrachromosomal linear amplicons. We selected clones of wild-type cells and of HYG/NEO MRE11−/− for MTX resistance in a stepwise manner (up to 1600 nM, a 16-fold increase in resistance compared to starting parent cells), a drug known to select for PTR1 linear DNA amplifications [13], [18], [28]. Leishmania chromosomes extracted from ten MTX resistant clones derived from either WT or HYG/NEO MRE11−/− parasites were separated by pulse field gel electrophoresis (PFGE) and hybridized with a PTR1 probe. Ethidium bromide stained gels already indicated that some linear amplicons smaller than the smallest chromosome were present in some resistant clones derived from WT but not in the MTX resistant MRE11−/− mutants (Figure S4). Hybridization data revealed that all ten MTX resistant clones derived from WT cells displayedPTR1 linear amplicons of varying size of 125 kb, 250 kb, 450 kb and 565 kb (Figure 6A). Clones 6 and 7 also gave rise to PTR1 circular amplicons, as suggested from the hybridizing smears (Figure 6A). The situation was drastically different in the HYG/NEO MRE11−/− parasites selected for MTX resistance. We observed only one resistant clone from the MRE11 null mutant with a faint hybridization signal corresponding to a PTR1 linear amplification (Figure 6B, clone 1), while a PTR1 circular amplification was present in clones 4 and 5 derived from the MRE11−/− mutant (Figure 6B). Clone 3 displayed a hybridization signal at around 1150 kb (Figure 6B) which could correspond to either a very large linear amplicon or to a chromosomal translocation. The difference in formation of linear amplicons between WT and MRE11−/− null mutant was found to be significant (p<0,01). We also selected the add back strains HYG/PUR-MRE11WT and HYG/PUR-MRE11H210Y for MTX resistance for testing for the specificity of the phenotype and for assessing the role of the MRE11-exonuclease activity in the generation of linear amplicons. While the MRE11−/− mutants had a decreased capacity to generate linear amplicons after MTX selection (Figure 6B), nine out of ten MTX resistant clones derived from the HYG/PUR-MRE11WT add back strain had PTR1 linear amplicons (Figure 6C, clones 1, 2, 4–10). Similar to the mutants derived from the wild-type cells (Figure 6A), four different PTR1 linear amplicons of 125, 250, 450 and 565 kb (Figure 6C) were present and four clones derived from HYG/PUR-MRE11WT had additional PTR1 circular amplicons (Figure 6C, clones 1, 2, 6 and 7). This phenotype reversion was not observed when the MRE11−/− cells were complemented with MRE11WT as part of an episomal construct. Clones derived from the latter transfectants and selected for MTX resistance were similar to the MRE11−/− mutants with no PTR1 linear amplicons (Figure S5). The results were even more surprising with the MTX resistant clones derived from HYG/PUR-MRE11H210Y. Strikingly all mutants had circular amplifications and the PTR1 hybridization intensity was in general much higher suggesting a higher copy number of the circles. Four clones derived from this add-back revertant also had a PTR1 linear amplicon (Figure 6D, clones 2, 7, 9, 10).
Previous data has indicated that linear amplicons are constituted of inverted duplications rearranged at the level of IRs with the formation of a new junction that can be amplified by PCR (see Figure 1). The diversity in size of linear amplicons observed in Figure 6A and 6C would suggest that different IRs were used and we tested for the presence of IRs in the chromosome 23 that could lead to PTR1 amplicons with size (125, 250, 450 and 565 kb) consistent with what observed in the blots. We detected a potential of 5 such IRs with size ranging from 440 to 790 bp and with a minimum of 85% identity (Figure 7A), a finding consistent with our demonstration that low copy repeated sequences are widespread throughout the genomes [19]. We performed PCR assays using five different pairs of primers recognizing the five different pairs of IRs under the principle shown in Figure 1. Amplification of the GAPDH gene was also done as a control. In the ten MTX resistant clones derived from WT cells and the HYG/PUR-MRE11WT add back strain, we detected several junctions by PCR (Figure 7B and D). The junction formed following a rearrangement at the level of IRs AA′ was the most frequently observed rearrangement, but junctions BB′, EE′ were also detected frequently while the junctions DD′ and CC′ were detected once in clones derived respectively from the WT or add back strains (Figures 7B and 7D). In clones in which we detected circles by Southern blots (Figure 6), we also obtained a positive signal for junction FF′ where the repeats are in direct orientation (Figure 7). There was a general good agreement between the number of amplicons detected by southern blots and PCR although PCR was more sensitive. For example, we could not detect a linear amplicon in clone 2 in MRE11−/− (Figure 6B) but it had a positive PCR reaction for the junction AA′ (Figure 7C).
The MRE11−/− MTX resistant mutants do not have PTR1 amplification (Figure 6B) and these cells must resist MTX by other means. Several mechanisms of resistance have been described [47], [48] including transport defects and gene amplification. We have carried out transport experiments in the mutants and observed no difference in MTX uptake between the MRE11−/− and the MRE11−/− MTX resistant mutants. We also hybridized PFGE blots with a DHFR-TS probe and while we failed in detecting circular DHFR-TS amplification, we observed an extra high molecular weight band hybridizing to DHFR-TS (Figure S6). The exact mechanism leading to this rearrangement is not known but it leads to a 2-fold increase in DHFR-TS expression (Figure S7) and it may contribute to MTX resistance.
Gene amplification as part of linear or circular extrachromosomal elements is frequently observed in the parasite Leishmania selected for drug resistance or subjected to nutritional stresses [11]. The circles or linear elements are formed at the level of homologous direct or inverted repeats, with more than 2000 repeats of more than 200 bp representing close to 5% of the Leishmania genome [19]. The genome of Leishmania is continuously being rearranged at the level of these repeats and while each cell has a core genome, they each differ by a complement of circular or linear amplicons. Upon selection the copy number of these elements increases and in the absence of selection the copy number of these elements decreases [19]. It was shown that circular elements are formed by homologous recombination between direct repeated sequences which is catalyzed by the RAD51 recombinase, known to be involved in the homologous recombination process in kinetoplastids [49], [50]. However, the rate of formation of linear amplicons was unchanged in a RAD51−/− mutant and linear amplicons must then be formed through another pathway [19]. We hypothesized that DNA repair proteins with nuclease activities may be involved since genomic DNA must be processed when inverted repeats are annealing for the formation of linear amplicons (Figure 1). We thus focused our efforts on the nuclease MRE11 that is part of the MRN complex [1], [51].
The biochemical characterization of the Leishmania MRE11 protein indicated that it has properties similar to other MRE11 orthologues. Indeed, as previously reported for MRE11 homologs in other organisms [42], [51]–[53], it binds preferentially SS and SA DNA structures in competitive assays (Figure 2); it is capable of DNA end resection and exhibited a 3′→5′ exonuclease activity on DS DNA structures, albeit with less effectiveness than the human enzyme (Figure 3). A H210Y mutant abolished its nuclease activity without impairing its DNA binding properties (Figures 2 and 3A). Finally, LiMRE11 recruitment at DNA damage loci was demonstrated in human ATLD cells, an indication that LiMRE11 detects and binds DNA breaks in these cells (Figure 4).
Having established that the Leishmania MRE11 protein bears all the hallmarks of the MRE11 family of DNA nucleases, we generated a recombinant parasite where the two alleles were inactivated (Figure 5). These parasites were viable but displayed a growth defect (Figure 5C). The parasites were also more sensitive to the DNA damaging agent MMS (Figure 5D) as also observed for Trypanosoma brucei [54]. The growth defect and susceptibility to MMS were reverted when we re-introduced one allele of MRE11 at its original chromosomal locus (Figures 5C and 5D) or as part of an episomal construct (Figure S2). Surprisingly, the expression of an episomal MRE11 gene in a WT strain severely impaired its growth rate (Figure S2A) and led to increased sensitivity to MMS (Figure 2). This result strongly suggests that in a wild-type background, the overproduction of MRE11 is somehow affecting the parasite cell growth, a phenomenon usually not observed in other cell types. The expression level of GFP-MRE11 in WT cells was very low, a phenotype also reported in the trypanosomatid parasite Trypanosoma brucei [55]. These early divergent eukaryotes may be more sensitive to exonuclease overexpression. Alternatively, overexpression of MRE11 in Leishmania may alter more acutely the stoichiometry of interactions with partner proteins such as RAD50 [56] and this could have an impact on cell growth. Interestingly and in support of the above hypothesis, we have shown in an independent study that RAD50 is essential in Leishmania WT cells but its gene can be inactivated in a MRE11−/− background (Laffitte et al., unpublished data). Possibly that recombination pathways are changed in MRE11−/− to compensate for the loss of MRE11, therefore altering the importance of the MRN complex and its components. Growth delay observed in WT cells overexpressing MRE11 may also relate to the known role of MRE11 in cell cycle regulation [37]. Indeed, MRE11 is involved in control of DNA replication initiation [57] and overexpression of MRE11 in Leishmania may have stronger effect on replication of Leishmania chromosomes.
Selection for MTX resistance often leads to linear amplifications of PTR1 in Leishmania [13], [14], [28]. We selected wild-type cells, MRE11−/− null mutants and reverted lines for MTX resistance. Amplified linear PTR1-containing amplicons were observed in all the clones derived from the WT strain but in only 1 out of 10 clones derived from the MRE11−/− null mutant (Figure 6). The capacity to generate circular amplicons was similar in the two different lines (Figure 6A and 6B). This strong phenotype was specific to MRE11, as reintroduction of MRE11 at the original chromosomal locus restored the ability of the parasites to generate PTR1 linear amplicons upon MTX selection (Figure 6C). Leishmania differs from yeast in this process since the MRN complex can prevent palindrome amplification in yeast. This process requires the interaction with the CtIP protein [36], [58] which is absent in Leishmania [41], possibly explaining the difference between the two organisms. Two others important parameters to consider are the length of the inverted repeats, which are very long in Leishmania, and the length of the sequences between these repeats. Indeed, it was previously suggested in yeast that hairpins with large loops are handled differently than hairpin with smaller loops [59]. This explanation is consistent with our study where IRs are few kb apart (Figure 7A), creating large loops in the hairpin structure, while most of the experiments done in yeast presents IRs closer to each other [58]–[61]. Further experiments could be interesting to determine whether the hairpin strength and structure influences DNA processing by the MRN complex.
The reversion of linear amplicons phenotype is dependent on MRE11 nuclease activity since reintegration of the mutated version MRE11H210Y led to parasites generating more efficiently circular PTR1 amplicons but not linear ones (Figure 6D). The mutated MRE11 therefore appears to favor homologous recombination in rescued parasites at the level of direct repeated sequences leading to circular amplicons. It is known that MRE11 is involved in initial events of homologous recombination in many organisms [37], [38], [62] and can interact with a number of nucleases and helicases, several of which are encoded in the L. infantum genome [41]. We suggest that the Leishmania MRE11H210Y is still capable of binding DNA and therefore MRN complex formation is intact, as it was previously suggested in yeast [63]. However, MRE11 lack of nuclease activity possibly makes it a better bait for recruiting HR proteins including RAD51. This facilitated recruitment could be due for example to putative longer association kinetics of the mutated MRE11 to DNA. Alternatively, the inability of LiMRE11H210Y to perform DNA resection may alter the first steps of DNA repair and possibly increase HR proteins recruitment, hence facilitating the formation of circular amplicons. This phenotype is observed in a MRE11−/− background in which we believe that recombination pathways have changed, possibly for compensating loss of MRE11. Thus, a combination of alterations in recombination pathways along with the mutated MRE11 may be responsible for the phenotype observed. It is salient to reiterate that while the episomal expression of MRE11 in the MRE11−/− null mutants reverted the growth phenotype and sensitivity to MMS (Figure S2), it did not revert the phenotype of generating linear amplicons upon MTX selection (Figure S5). This is a further demonstration of the importance of a suitable level of expression to recover proper MRE11 functions.
We have shown that gene rearrangements are continuously taking place at the level or repeated sequences, and that these rearrangements can be highlighted by PCR assays. Using PCR, we have shown that the PTR1 linear amplicons are generated at the level of 5 different inverted repeats, indicating that different rearrangements led to the linear amplicons (Figure 7). The IRs most frequently used are AA′, BB′ and EE′ and these are relatively close to one another (10 kb between A and A′ as well as between E and E′, 3 kb between B and B′) while IRs CC′ and DD′ are further apart (respectively 38 and 53 kb) and used only once. This suggests that the length of the intervening sequences between the IRs may impact the rate of annealing of the IRs and the rearrangements leading to linear amplicons. Few amplicons detected by southern blot were not observed by PCR, suggesting that either smaller inverted repeats were used or secondary rearrangements occurred. Our bioinformatics screen has revealed only one direct repeat that could entertain PTR1 circular amplification and indeed the PCR assay has revealed that in every cell in which a circular amplicon was observed in Figure 6, we observed a positive PCR signal indicative of recombination between direct repeat sequences FF′ (Figure 7).
Because of its lack of control at the level of transcription initiation, Leishmania is likely to use several mechanisms to regulate its expression. We have suggested that gene rearrangements leading to copy number variation is one such mechanism. Indeed the whole Leishmania genome is continuously and stochastically rearranged at the level of repeated sequences. We have shown that there are at least two pathways of rearrangement. One leads to circles after recombination between two direct repeated sequences and this requires RAD51. Here we have shown that linear amplicons, formed at the annealing of two IRs, is facilitated by the presence of an active MRE11. We proposed that double-strand breaks (see Figure 1) would be necessary, although this will require experimental validation, which may be challenging in Leishmania as they are no suitable inducible systems. Gene rearrangement is one main mechanism of resistance in Leishmania and a further understanding of the proteins involved in gene rearrangements may provide a strategy to circumvent the emergence of drug resistance.
Promastigotes of Leishmania infantum (MHOM/MA/67/ITMAP-263) and all recombinants were grown in SDM-79 medium at 25°C supplemented with 10% fetal bovine serum, 5 µg/ml of hemin at pH 7.0. Independent clones of all cells generated in this study were selected for methotrexate (MTX) resistance, using a stepwise selection starting from an EC50 of 100 nM up to 1600 nM of MTX. All chemical reagents were purchased from Sigma-Aldrich unless specified and were of the highest grades.
The L. infantum MRE11 gene (LinJ.27.1790) was amplified by PCR using primers 1 and 2 (Table S1) on genomic DNA template and cloned in a modified pFASTBAC1 plasmid (Invitrogen) [64] encoding the glutathione-S-transferase tag (GST) at the N-terminus of MRE11 and a 10-histidine tag at its C-terminus for protein purification. Site-directed mutagenesis (Stratagene, Quickchange) was used to generate the LiMRE11 mutant H210Y using primers 15 and 16 (Table S1). The LiMRE11WT protein and the mutated version LiMRE11H210Y were purified from baculovirus-infected SF9 cells and the GST tag was removed by PreScission cleavage as described in [64]. Full-length human MRE11 cDNAs cloned in pFASTBAC were generously provided by Tanya Paull (University of Texas, Austin). Primers 22 and 23 (Table S1) were used for site-directed mutagenesis (Stratagene, Quickchange) to generate the human MRE11 mutant H217Y. Proteins hMRE11WT and hMRE11H217Y were purified as described [65]. Full-length human MRE11 cDNAs cloned in pEYFP-C1 (Clontech) was generously provided by John Petrini (University of California, San Francisco). The fluorescence observed with pEYFP-C1 is equivalent to that from pEGFP-C1. The L. infantum gene LiMRE11WT was cloned in pEGFP-C1 plasmid (Clontech, encoding a GFP tag located at the N-terminus) for FRAP analysis.
DNA substrates were made by the annealing of the 32P-labelled primer 21 with either primer 17 for double-stranded DNA substrate (DS) or primer 20 for splayed arm (SA) (Table S1). Reactions (10 µL) contained 25 nM of 32P-labeled DNA oligonucleotides with the indicated concentration of proteins (see Figure 2) in MOPS buffer (25 mM MOPS (morpholine-propanesulfonic acid) pH 7.0, 0,2% tween-20, 2 mM CaCl2 and 2 mM DTT). After 15 minutes of incubation at 37°C, reactions were fixed at 37°C during 15 minutes with 0.2% glutaraldehyde. Samples were loaded onto a 8% TBE 1× acrylamide gel, run at 150 V for 1h30, followed by autoradiography.
Exonuclease assays were performed in MOPS/EXO buffer (25 mM MOPS (morpholine-propanesulfonic acid) pH 7.0, 60 mM KCl, 0.2% tween-20, 2 mM DTT, 2 mM ATP, 5 mM MnCl2). Double-stranded DNA substrate (DS) was generated as stated above. The indicated concentration of proteins (see Figure 3) were incubated in MOPS/EXO buffer with 200 nM of 32P-labeled DNA for 30 minutes at 37°C, followed by deproteinization in one-fifth volume of stop buffer (20 mM Tris-Cl pH 7.5 and 2 mg/mL proteinase K) for 30 minutes at 37°C. The reactions were boiled during 5 minutes at 95°C after the addition of formamide blue (50% final) then loaded on 8% acrylamide/urea gels. Gels were run at 75W for 60 minutes, dried onto DE81 filter paper, followed by autoradiography. For exonuclease assay on different DNA substrates, 32P-labeled oligonucleotide 21 (Table S1) was labeled at the 5′-end (using the terminal transferase and the New England Biolabs protocol) and hybridized to primers 17, 18 and 19 (Table S1).
ATLD human cells (kindly obtained from Yossi Shiloh, University of Tel Aviv, Israël) were maintained in DMEM medium supplemented with 20% fetal bovine serum and 1% penicillin/streptomycin (Life Technologies). ATLD cells were transfected by electroporation with 50 µg of LiMRE11-GFP or hMRE11-GFP DNA constructs. After 16 hours, we performed Fluorescence recovery after photobleaching (FRAP) analysis. Briefly, fluorescence was monitored on a Leica TCS SP5 II confocal microscope and laser-induced DNA damage was created using a 405-nm UV laser. Visualization of GFP fluorescence within the micro-irradiated nuclear region was achieved using a 488 nm excitation filter and a 63× objective. Background and photo-bleaching corrections were applied to each dataset using the Volocity-software.
The L. infantum MRE11 null mutant (MRE11−/−) cells were obtained by targeted gene replacement. MRE11 flanking regions were amplified from L. infantum wild-type genomic DNA and fused to both neomycin phosphotransferase (NEO) and hygromycin phosphotransferase (HYG) genes using a PCR fusion based-method as described previously [66]. Briefly, 5′UTR of MRE11 was amplified using primers 3 and 4 for the NEO cassette and primers 3 and 5 for the HYG cassette. The NEO gene was amplified with primers 7 and 10 and the HYG gene with primers 8 and 11. 3′UTR of MRE11 was amplified using primers 13 and 14 for both inactivation cassettes (see primer sequences in Table S1). At least 3 µg of the 5′UTR-NEO-3′UTR and 5′UTR-HYG-3′UTR linear fragments were successively transfected by electroporation (as described in [67]) into L. infantum WT to replace both MRE11 alleles. Recombinants were selected in the presence of 300 µg/ml of hygromycin B (New England Biolabs, Beverly, MA, USA) and 40 µg/ml of G418 (Geneticin; Sigma-Aldrich). After 4–5 passages, cells resistant to the drug selection were cloned in SDM-Agar plates (1%) in the presence of antibiotics at the same concentrations. Ten clones of each recombinant were picked up after 10 days and used for further analysis.
A re-expression cassette, 5′UTR-MRE11-α-PUR-3′UTR was designed to reintroduce MRE11 into its original chromosomal locus in the HYG/NEO MRE11−/− null mutant. Briefly, this cassette was obtained by PCR of the PUR gene using primers 9 and 12 on the plasmid template Psp72-α-PUR-α [68] encoding the puromycin acetyltransferase marker. This fragment was fused to the 5′UTR and coding sequences of MRE11 (amplified using primers 3 and 6) and 3′UTR fragments (amplified using primers 13 and 14 in Table S1). The cassette was then transfected by electroporation in the L. infantum HYG/NEO MRE11−/− parasites to replace the NEO allele and recombinants were selected with 100 µg/ml of puromycin (Sigma–Aldrich) and 300 µg/ml of hygromycin B (New England Biolabs). The same strategy was used to introduce MRE11 containing the mutation H210Y in the HYG/NEO MRE11−/− strain. The MRE11 ORF was also cloned in the episomal plasmid Psp72-α-puro-α, transfected in L. infantum WT and HYG/NEO MRE11−/− parasites, and stable transfectants were selected with 100 µg/ml of puromycin.
MRE11 allele replacement was confirmed by Southern blot analyses. Genomic DNAs from clones were isolated using DNAzol as recommended by the manufacturer (Invitrogen). Digested genomic DNAs or separated chromosomes were subjected to Southern blot hybridization with [α-32P]dCTP-labeled DNA probes according to standard protocols [69]. All probes were obtained by PCR (see primers in Table S1) from L. infantum genomic DNAs.
RNAs were extracted using RNeasy plus mini kit (Sigma) according to the manufacturer recommendations. The cDNA was synthesized using Oligo dT12–18 and SuperScript II RNase H-Reverse Transcriptase (Invitrogen) and amplified in SYBR Green Supermix (Bio-Rad) using a rotator thermocycler Rotor Gene (RG 3000, Corbett Research). The expression level was derived from three technical and three biological replicates and was normalized to constitutively expressed mRNA encoding glyceraldehyde-3-phosphate dehygrogenase (GAPDH, LinJ.36.2480). The sequences of the primers used in this assay are listed in Table S1.
L. infantum WT, HYG/NEO MRE11−/−, MRE11 re-expressing cells (HYG/PUR-MRE11WT and HYG/PUR-MRE11H210Y), WT Psp72-α-puro-α-MRE11 and HYG/NEO MRE11−/− Psp72-α-puro-α-MRE11 were resuspended at a concentration of 5×106 cells/ml and exposed to increasing doses of MMS (Sigma–Aldrich). Cells were counted after 72 h and reported in survival rate.
Intact chromosomes were prepared from L. infantum promastigotes harvested from log phase cultures, washed once in 1× Hepes-NaCl buffer then lysed in situ in 1% low melting agarose plugs. Briefly, cells were resuspended in HEPES-NaCl buffer at a density of 5×107 cells/ml and mixed with an equal volume of low melting-point agarose (Invitrogen). Cells were then lysed overnight at 50°C in lysis buffer (0.5M ethylenediaminetetraacetic acid (EDTA) pH 9.5, 1% sodium dodecyl sulfate (SDS), 350 ug/ml proteinase K). Leishmania intact chromosomes were separated in 1× TBE buffer (from 10× TBE: 1M Tris, 1M Acid boric, 0,02M EDTA) by Pulsed-Field Gel Electrophoresis (PFGE) using a Bio-Rad CHEF-DR III apparatus at 5 V/cm and a 120°C separation angle as described previously [70]. The range of chromosome separation was between 150 and 1500 kb.
Repeated intergenic sequences were already characterized [19]. Primers (see Table S1) used to detect new junctions created by amplicon formation (Figure 1) were designed for all putative recombination/annealing events between repeated sequences. Primers were chosen within 150 nucleotides from the repeated sequences with their orientation shown in Figure 1. Optimal primer length was 20 nucleotides and optimal melting temperature (Tm) was 64°C.
Late log phase promastigotes (30 ml) were pelleted at 3000 rpm for 5 minutes and pellets were washed once with 1× HEPES-NaCl, resuspended in suspension buffer (100 mM EDTA, 100 mM NaCl, 10 mM Tris pH 8.0), then lysed in 1% SDS and 50 µg/ml proteinase K at 37°C for 2 hours. Genomic DNAs were extracted with 1 volume phenol, precipitated with 2 volume 99% ethanol, washed with 70% ethanol twice and dissolved in 1 ml 1× TE buffer. RNAse A (Qiagen) was added at 20 µg/ml and DNAs were incubated at 37°C for 30 minutes, followed by the addition of 50 µg/ml of proteinase K and 0.1% SDS at 37°C for 30 minutes. DNAs were extracted with 1 volume of phenol, precipitated and washed in ethanol, and dissolved in DNase free-water (Millipore) at 37°C overnight.
PCR reaction mixtures consisted of 100 ng of genomic DNA isolated as described above, 1 µl of forward and reverse primers at 100 µM (Table S1), 0.5 µl dNTP mix at 10 mM, 1.25 U of FastStart Taq DNA polymerase (Roche), 1× PCR buffer+MgCl2 and 1.25 µl BSA at 66 mg/ml. The total reaction mixture was made up to 25 µl by addition of the genomic DNA. For each PCR reaction, the number of cycles was optimized to prevent saturation of the amplification. Saturation of band intensities of the amplified PCR products was determined using the AlphaImager 2000 software. The housekeeping chromosomal gene GAPDH (LinJ36.2480) was used as an internal control (primers pair gg' in Table S1) to normalize the amount of DNA loaded in each reaction.
|
10.1371/journal.pcbi.1002758 | Impact of Stoichiometry Representation on Simulation of Genotype-Phenotype Relationships in Metabolic Networks | Genome-scale metabolic networks provide a comprehensive structural framework for modeling genotype-phenotype relationships through flux simulations. The solution space for the metabolic flux state of the cell is typically very large and optimization-based approaches are often necessary for predicting the active metabolic state under specific environmental conditions. The objective function to be used in such optimization algorithms is directly linked with the biological hypothesis underlying the model and therefore it is one of the most relevant parameters for successful modeling. Although linear combination of selected fluxes is widely used for formulating metabolic objective functions, we show that the resulting optimization problem is sensitive towards stoichiometry representation of the metabolic network. This undesirable sensitivity leads to different simulation results when using numerically different but biochemically equivalent stoichiometry representations and thereby makes biological interpretation intrinsically subjective and ambiguous. We hereby propose a new method, Minimization of Metabolites Balance (MiMBl), which decouples the artifacts of stoichiometry representation from the formulation of the desired objective functions, by casting objective functions using metabolite turnovers rather than fluxes. By simulating perturbed metabolic networks, we demonstrate that the use of stoichiometry representation independent algorithms is fundamental for unambiguously linking modeling results with biological interpretation. For example, MiMBl allowed us to expand the scope of metabolic modeling in elucidating the mechanistic basis of several genetic interactions in Saccharomyces cerevisiae.
| One of the challenging tasks in systems biology is to quantitatively predict the metabolic behavior of the cell under given genetic and environmental constraints. To this end, genome-scale metabolic reconstructions and simulation tools are indispensable. The choice of the objective function to be used for simulating genome-scale metabolic models is dependent on the biological context and one of the most relevant parameters for successful modeling. Formulation of the intended objective function often requires the use of multiple fluxes, e.g. the sum of fluxes through ATP-producing reactions. We demonstrate that the existing tools confound biological interpretation of the simulations due to undesired dependence on the representation of stoichiometry and propose a new tool – Minimization of Metabolites Balance (MiMBl). MiMBl allows casting of the desired biological objective functions into linear optimization models and gives consistent simulation results when using numerically different but biochemically equivalent stoichiometry representations. We demonstrate relevance of MiMBl for addressing biological questions through improved predictions of genetic interactions within the yeast metabolic network. Genetic interactions imply functional relationship between the genes and therefore allow assessing different hypotheses for the underlying biological principles. MiMBl explains several of the genetic interactions as outcome of flux re-routing for minimal metabolite turnover adjustments.
| The fundamental role of metabolism within a living cell has become a focal point of study in many disciplines, such as cell biology, physiology, medicine and synthetic biology. The assembly of all reactions and metabolites into a genome-scale metabolic network provides a comprehensive structural framework for integrative data analysis [1], [2], as well as for quantitative modeling of cellular metabolism [3]–[6]. As the solution space for the metabolic flux state of the cell is typically very large, constraint based optimization approaches are often applied for simulating metabolic fluxes. In essence, these approaches search for an optimal flux distribution that maximizes or minimizes an appropriate biological objective function while satisfying the mass balance and metabolite exchange constraints. Among these, Flux Balance Analysis [7] is a widely used simulation tool that utilizes a linear programming formulation for maximization of growth (synthesis of biomass constituents) as biological objective function. FBA has been applied with various degrees of success, albeit mostly for “wild-type” or unperturbed metabolic networks [8], [9]. In addition to FBA, various other objective functions are frequently used, including minimization of overall intracellular flux and maximization of ATP yield, among others. An overview of various commonly used objective functions and their evaluation against experimental data for Escherichia coli can be found in Schuetz et al. [10]. In case of genetically or environmentally perturbed networks, Minimization of Metabolic Adjustment algorithm - MoMA [11] - has been reported to better represent the biological observations [11]–[14]. The hypothesis underlying MoMA is that fluxes in a perturbed cell (e.g. a mutant) will be redistributed so as to be as similar as possible to the wild-type.
The biological principles exemplified by simulation tools for both wild-type and perturbed networks are undeniably fascinating, which is confirmed by their numerous applications – including prediction of genetic interactions [2], [15], [16], metabolic engineering [13], [14], [17], microbial community modeling [18], [19] and search for evolutionary constraints in relation to different objective functions [20]. Several of the objective functions commonly used in these and other applications rely on the use of linear combination of fluxes, e.g., MoMA or minimization of overall intracellular flux (Table 1). We found that the mathematical formulation of this class of problems (i.e. where linear combination of fluxes is part of the objective function) is sensitive to the representation of the reaction stoichiometry, with results strongly dependent on the adopted scaling of the stoichiometric coefficients. Such dependency confounds the biological interpretation of simulation results, as biochemically equivalent alternative representations of the same network can lead to contradictory predictions upon a given genetic or environmental perturbation. For example, the status of a given gene may change from non-essential to essential while using biochemically equivalent representations of the stoichiometry of the metabolic network (Table S1). As the stoichiometric representation of any reaction is subjective (often scaled to have coefficient of 1 for one of the reactants/products) and a typical genome-scale modeling problem involves hundreds of reactions, there are infinitely many biochemically equivalent ways to represent a given metabolic network. Any simulation algorithm should therefore be independent of the stoichiometry representation.
We motivate the need for rethinking the problem formulation for metabolic modeling by illustrating how the current methods lead to incoherent biological predictions when alternatively representing the reaction stoichiometry. Tackling a proper problem formulation, we propose a new methodology for metabolic modeling – Minimization of Metabolites Balance (MiMBl), which accounts for reaction stoichiometry in the objective function by mapping the flux space into the metabolite turnover space. As intended, MiMBl shows robust predictions independently of the stoichiometry representation. We demonstrate the biological relevance of the new formulation with increased power for predicting genetic interactions in the metabolic network of S. cerevisiae. In a recent study reporting a large genetic interactions dataset covering the S. cerevisiae metabolic network [2], FBA was found to have limited capability for predicting the experimentally observed interactions, partially due to the lack of regulatory information. Within this study we successfully challenged MiMBl to accomplish the task of extending the range of genetic interactions that can be predicted. By combining the results from MiMBl and FBA, we probe the operating mechanisms underlying genetic interactions within metabolic networks.
Several of the biological objective functions widely used in metabolic modeling are currently formulated as linear (or quadratic) combination of fluxes. Minimization of sum of intracellular fluxes and minimization of metabolic adjustment belong to this class and are herein used as case studies of biological principles that can be robustly formulated by using MiMBl. Two different genome-scale reconstructions of the S. cerevisiae metabolic network are used, viz. iFF708 [21] and iAZ900 [22], as the choice of the appropriate metabolic reconstruction depends on the biological question to be addressed (Methods).
Minimization of the sum of intracellular flux is a routinely used objective function for estimating intracellular fluxes [10], [20], [23], [24]. By using the iFF708 S. cerevisiae genome-scale metabolic reconstruction [21] together with experimentally determined exchange rate constraints (Text S1), we illustrate how the use of this objective function leads to inconsistent predictions when using numerically different, but biochemically equivalent, reaction stoichiometry. Linear scaling of all stoichiometric coefficients of a given reaction (e.g. multiplication by a scalar θ, Methods) preserves the stoichiometry and must not affect the simulation outcome for a correct problem formulation. However, in this case, scaling of a single reaction (RPI1) results in diverting the carbon flow from glycolysis to pentose phosphate pathway, which is one of the most important metabolic branch points (Fig. 1). This deviation was verified not to be consequence of alternative optima of the same mathematical solution (Fig. S2), thus representing different biological solutions.
In order to provide insight into the nature of the problem leading to the susceptibility of the solution towards alternative representation of the stoichiometric matrix, we use a toy-model depicted in Fig. 2a. As a case study, minimization of metabolic adjustment was chosen as biological principle and formulated as minimization of Manhattan distance (a commonly used formulation of MoMA, termed lMoMA [25]). Fig. 2 also illustrates the representation dependency of the Euclidean distance formulation of MoMA (quadratic MoMA, as originally proposed in [11]). In order to provide an intuitive insight, the following discussion is centered on lMoMA. Similar explanation holds true in quadratic space in the case of quadratic MoMA. In the wild-type toy-model, flux goes from A to D via R5. The goal is to predict flux distribution in the mutant lacking R5. The biological principle of minimization of metabolic adjustment dictates rewiring of the flux through R6. However, lMoMA found contradictory optimal solutions, i.e. solutions that re-route the flux via R2–R3–R4 or R6, depending on the stoichiometric representation of R6 (Fig. 2b). Insight into the cause of this behavior can be gained by analyzing the optimal objective function values, i.e. distances, as function of θR6 (Fig. 2d). Smaller θR6 implies higher numerical value of the flux through R6, hence higher contribution of R6 to the distance. Consequently, after a certain value of θR6, the activation of the longer R2–R3–R4 pathway more than compensates the use of R6. The two solutions are not alternative optima, as the objective function value neither remains constant nor linearly scales with θR6. Such non-linear dependency of the objective function value on the scalar θR6 violates the requirement of a correct problem formulation. Indeed, we analytically demonstrate that the optimality condition for the linear programming problem after scaling is not guaranteed to be satisfied in the case of using sum of fluxes as part of the objective function (Methods). Notably, widely used FBA-like problems (max/minimization of a single flux) are perfectly robust concerning the scaling of the stoichiometric coefficients. As a single flux is used in the objective function, the relative values of all the remaining fluxes (which depend on the stoichiometry representation) does not influence the optimal solution to be found (for a theoretical proof, see Text S2).
The mathematical caveat illustrated above means that the contribution of the desired biological objective function towards the obtained solution is inseparable from that of the artifacts of stoichiometry representation. Importantly, in large metabolic networks the effects of stoichiometric representation of reactions are cumulative. As we herein show, this problem can be solved by proper normalization of the objective function variables with respect to stoichiometric representation of the reactions. To achieve such normalization, we devised two approaches, normalized lMoMA (normlMoMA) and Minimization of Metabolites Balance (MiMBl). In normlMoMA, each variable in the objective function is normalized by its value in the wild-type flux distribution. Albeit being simple, this normalization method has three major drawbacks: i) many reactions often have null fluxes in the wild-type, thus posing a problem for normalization (Methods, Text S2 and Fig. S3); ii) it requires a reference flux distribution to obtain the normalization factors, making it inappropriate to formulate objective functions such as minimization of overall intracellular flux; and iii) the influence of each flux on the metabolic adjustment would be exclusively due to its fold change, not taking into account that reactions carrying higher fluxes could have a stronger impact on the predicted flux distribution, as implied in the original concept of minimization of metabolic adjustment.
To obtain a biological meaningful and mathematically robust normalization, we propose Minimization of Metabolites Balance (MiMBl) as a new method for metabolic modeling. The objective function in MiMBl is formulated as a linear combination of metabolite turnovers (tM). The turnover of a metabolite is the sum of all fluxes producing (or consuming) it, multiplied by the corresponding stoichiometric coefficients (Methods). The objective function for minimization of metabolic adjustment is reformulated to include metabolite turnovers instead of fluxes (Fig. 2c). Because the stoichiometric coefficients are taken into account while calculating tM, MiMBl is robust to the linear scaling of the stoichiometric matrix, analytical proof of which is presented in the Methods section. In case of the toy-model (Fig. 2a, d), this robustness is illustrated by the invariant nature of the objective function as well as the flux distribution. Note that the flux through R6 linearly scales with θR6, while the turnover of all metabolites is conserved. The normalization implied in MiMBl formulation is suitable for addressing a variety of biological questions involving different objective functions, such as minimization of overall intracellular flux (by using a null vector for wild-type flux distribution) or maximization of ATP yield (by maximizing the ATP turnover for a given substrate uptake rate), among others (Table 1).
While mapping the flux space into the metabolite space for the objective function formulation, as we do for MiMBl, it is possible that, for a few cases, alternative flux distributions are found around a given metabolite. We therefore introduce a second optimization step that reinforces the proximity to the reference flux distribution. This is achieved by using a normlMoMA routine where the optimal objective function value found in the first MiMBl optimization step is used as an additional constraint (Methods). Nevertheless, highly connected metabolites ensure a degree of network connectivity, which is sufficient for decreasing the number of situations where alternative flux distributions around the same metabolite are picked by MiMBl. Indeed, we did not find any case in the simulations performed for this study where growth prediction was altered in the second optimization step. An example case where the second optimization step will be more relevant is simulations involving export of metabolites, where the choice of a particular transporter (as in the reference flux distribution) among several alternative options is desired. A more thorough analysis of MiMBl alternative optima in a genome-scale network is presented below (Fig. 3).
In order to estimate the extent to which the lack of normalization of stoichiometric coefficients within the objective function influences the biological interpretation of simulation results, we used lMoMA for simulating gene knockouts in the S. cerevisiae genome-scale metabolic model iFF708 [21]. In case of single gene knockout, three genes were found to change their status from non-essential to essential while using two biochemically equivalent matrix representations (Table S1). For instance, the mutant lacking YCR012W, coding for a 3-phosphoglycerate kinase (pPGK1), was predicted to be viable when using the as-published representation of the stoichiometric matrix S0 [21], and non-viable while using the biochemically equivalent matrix S1 (Methods). Based on such contradictory results, conclusions cannot be taken on whether YCR012W is predicted to be essential or not. As the number of deleted genes (or other network perturbations) increases, cumulative phenotypic effects related to the functional interactions between the genes are expected to take place and examples as the one mentioned above become even more striking. For triple gene knockouts, more than 200,000 triplets were found such that their predicted phenotype switched from lethal to non-lethal (or vice-versa) for the two biochemically equivalent matrix representations (Table S1). From a biotechnological perspective, predictions from genome-scale modeling have direct influence on the choice of gene targets selected for metabolic engineering. By using lMoMA, we identified metabolic engineering strategies (by simulating all possible combinations of knockouts of up to three genes, Text S1) for production of two different compounds in yeast: succinate – a native product, and vanillin-glucoside – a heterologous product. Not only a significant fraction of mutants had divergent predictions for product yield when using two biochemically equivalent stoichiometric matrices, but also several highly ranked strategies in one case were low priority targets in the other (Figs. S4, S5, S6, S7). Moreover, we also observed that the number of predicted synthetic lethal pairs differed by more than two-fold when using alternative stoichiometric matrix representations (Table S2). These inconsistencies have immediate implications on the consequent biological interpretation, as well as on the experimental design, and can be successfully overcome by using MiMBl (Fig. S4, Table S3).
The above analysis proved the robustness of MiMBl towards stoichiometric representation of metabolic reactions. However, some degree of uncertainty in the simulation results might still exist, as we shall show here, essentially arising from two main sources (Fig. S8a): i) sensitivity of the results towards the initial wild-type flux distribution used as input for minimizing the metabolic distance; and ii) potential non-uniqueness of the linear programming solution while simulating the mutant phenotype, i.e. existence of alternative optima. The sensitivity analysis for MiMBl towards both sources of uncertainty was performed using iAZ900 reconstruction of the yeast metabolic network, as the same reconstruction is subsequently used to study genetic interactions within the yeast metabolism. Both sources of variability also have impact on lMoMA simulation results (Fig. S9).
Firstly, we analyzed the sensitivity towards the wild-type (or reference) flux distribution used as input for minimization of metabolic adjustment. Using an accurate reference flux distribution is crucial for obtaining biologically meaningful simulation results. While some metabolite exchange rates are commonly available as experimentally derived constraints for the wild-type, they are usually not sufficient to uniquely estimate the corresponding intracellular fluxes, e.g. by using FBA (Figs. S8a and S1). It has been previously shown that the use of alternative optima within the reference flux distribution obtained with FBA can affect the prediction of growth upon gene deletions using quadratic MoMA [11]. We herein performed a similar analysis by using MiMBl. The growth of single gene deletion mutants was simulated with MiMBl while using alternative optimal FBA flux distributions as reference (Methods). Similarly to what was previously observed for quadratic MoMA [11], cases were found where the use of alternative FBA flux distributions, as input to MiMBl, influences the growth prediction (Fig. 3). 70% of the predictions of single gene deletion phenotypes were consistent across all FBA-alternative-optima used, while the remaining 30% showed dependence on the input reference flux distribution. Use of additional experimentally determined constraints, for instance as obtained with 13C flux analysis, will be useful for reducing the uncertainty in the input flux distribution and thereby in obtaining more robust predictions.
In order to assess the variability due to potential non-uniqueness of the optimal solution obtained with MiMBl (Fig. S8a), we performed a flux variability analysis [26]. Biologically, the alternative optima correspond to the existence of alternative pathways that result in equivalent mutant phenotypes with regards to the required metabolic adjustment. For a fixed reference flux distribution, we calculated the range of variability of intracellular fluxes upon constraining the metabolic adjustment (i.e. sum of metabolite turnover distance) to its optimal value (Methods). All of the tested fluxes were observed to have very low or no variability (vimin/vimax>0.99) across all single gene deletion phenotypes. Utility of the second step of MiMBl was seen in case of the flux through PGM1 upon deletion of YOR128C (Fig. S8b). Nevertheless, existence of a unique solution is problem dependent and it should be verified whether the possibility of alternative optima affects the prediction of fluxes of interest. Therefore, we performed an exhaustive analysis of variability of growth prediction across all single gene deletions, as well as all double gene deletions included in the genetic interactions case study. Growth was uniquely predicted in all these cases (Fig. S8c).
To what extent MiMBl contributes for increasing biological understandings gained from the application of optimization-based metabolic modeling? To address this question, we used one of the most recent and comprehensive S. cerevisiae models, iAZ900 [22], to run simulations for single and double gene knockouts and challenged MiMBl to predict the epistasis scores of all significantly interacting non-essential gene pairs reported by Szappanos et al. [2]. Genetic interaction networks are valuable resources towards deciphering the complex genotype-phenotype relationships. A genetic interaction between two genes occurs when the phenotype displayed by a double deletion mutant is different than the one expected based on the phenotypes of the single mutants. Accordingly, two genes can display positive, negative or no interaction. In order to capture most of the biological information contained in the experimental dataset, we used two different objective functions, maximization of growth (FBA) and minimization of metabolic adjustment (MiMBl). FBA is expected to cover situations where maximization of growth is the cellular objective, while MiMBl will account for regulatory effects inherent to the wild-type flux distribution, in the sense that the flux distribution in the perturbed network is kept as close as possible to that of the wild-type. This principle of proximity to the wild-type (or the reference) should partially reflect principles of flux reorganization in genetically perturbed networks. Although, both FBA and MiMBl performed equally well concerning gene essentiality predictions (∼60% sensitivity, Text S1), the benefit of using MiMBl as a biological objective function became apparent while predicting genetic interactions. This implies that the biological regulatory principle underlying MiMBl is manifested in yeast (under the investigated conditions) at larger network perturbations or less drastic phenotypes than essentiality. When applied for studying genetic interactions, FBA is a conservative method compared to MiMBl, since the parameter used to define and measure genetic interactions is also the objective of optimization, i.e., growth. Within the metabolic network, the existence of several optimal solutions theoretically satisfying maximum biomass formation is often observed. In case of a single/double gene deletion mutant where an alternative optimal pathway exists, FBA will always find such an alternative solution, even though it may not be biologically plausible due to regulatory constraints, and, thereby may miss potential genetic interactions. On the other hand, MiMBl will help in capturing more refined regulatory effects where the loss of growth is a side effect of minimizing the flux rerouting relative to the wild-type.
The subset of experimental genetic interactions involving non-essential genes from the yeast metabolic model contains 2745 interactions (939 positive and 1806 negative) connecting 520 genes (Text S1, Table S4). In order to assess the performance of the different algorithms, we carried out a sensitivity versus precision analysis. Precision was calculated as the fraction of experimentally validated interactions among all predicted interactions, while the sensitivity represents the fraction of the experimentally validated interactions captured by the predictions (Text S1). A computational epistasis score cutoff (εcutoff) was used to call a particular gene pair to be positively interacting (ε>εcutoff), negatively interacting (ε<−εcutoff) or non-interacting (|ε|<εcutoff) (Text S1). The performance of all algorithms (MiMBl, FBA, lMoMA and quadratic MoMA) is summarized as ROC (partial receiver operating characteristic) curves for both, positive and negative epistasis (Fig. 4 a, b and Fig. S10). The sensitivity and precision of the FBA predictions obtained in this study are within the same range as previously reported by Szappanos et al. [2]. MiMBl shows less precision than FBA in case of both positive (∼20% and ∼30%, respectively) and negative interactions (∼50% and ∼60%, respectively), but its sensitivity is considerably higher in both cases (∼9% vs ∼4% for positive, Fig. 4a; ∼5% vs ∼3% for negative, Fig. 4b), which reflects the conservative nature of FBA in predicting genetic interactions. Notably, for the entire range of genetic interaction cutoffs, MiMBl sensitivity and precision are considerably higher than those of lMoMA (Fig. 4a, b). The same trend was verified when the originally proposed quadratic MoMA formulation was used (Fig. S10). As previously reported by Szappanos and co-workers, lMoMA does not improve FBA predictions. This observation further emphasizes that a proper mathematical formulation of the biological principle (objective function) has a major impact on the ability to interpret in vivo observations.
We chose a strict interaction cutoff (|εcutoff| = 0.013) for further analysis of the predicted interactions (Fig. S11). For this cutoff, the correctly predicted genetic interactions map contains 142 interactions (73 positive and 69 negative) connecting 86 genes (Fig. 4f). MiMBl not only captures all interactions, except one, predicted by FBA, but also contributes with 48 additional interactions (∼34% of all accurate predictions). MiMBl predictions thus span almost all of those from FBA (Fig. 4c), which we attribute to the fact that many metabolites within the metabolic model are directly contributing to the biomass formation. Consequently, if the turnover of most metabolites is kept constant upon gene deletions, the biomass turnover (growth) will also remain constant. On the other hand, FBA is not able to capture many genetic interactions found by MiMBl (Fig. 4c). These will involve mutants where the loss of fitness upon gene deletion is caused by the change from an in vivo well-tuned pathway to an alternative pathway containing different metabolites or enzymes. For many of such cases, there are alternative pathways that sustain the same growth as the reference and FBA finds such solutions, regardless of the magnitude of the turnover adjustment that is required for the cell. Because of this feature, MiMBl is capable of capturing a part of the regulatory constraints on the operation of cellular metabolism, which lMoMA failed to capture (Fig. 4c).
The regulatory constraints imposed by MiMBl assume even stronger relevance in the case of positive interactions, where MiMBl exclusively accounts for almost 50% of all successfully predicted interactions (Fig. 4f). In fact, FBA's ability of predicting positive interactions is limited, as the maximum predicted biomass formation of a double deletion mutant would never be higher than the highest among those predicted for the two single deletion mutants. Thus, if a single deletion mutant has the maximum predicted fitness of 1 (meaning that the fitness of the mutant is the same as that of the wild-type), positive interactions involving the deleted gene will be impossible to predict. As FBA is bound to find the optimal solution that provides the highest growth, single mutants with maximum fitness are much more often predicted than the ones found by MiMBl, where minimal adjustment of the metabolic network is preferred over maintaining maximum growth. Indeed, MiMBl predicts decreased single mutant fitness for twice more gene knockouts than FBA (∼38.4 vs 18.1%). Consequently, MiMBl also displayed an improved capacity to predict both positive and negative epistasis involving the same gene. More than 80% of the genes display this feature in vivo. Interestingly, 30% of the genes involved in MiMBl predicted epistasis interact both positively and negatively, while FBA predicts that only 14% of the genes do so (Fig. 4f).
As metabolic networks are featured by several metabolites with a high degree of connectivity, interactions occur between distant pathways in the network. To assess MiMBl's ability to predict such pleiotropic effects, we calculated the network distance between each pair of genes accurately predicted to interact (Text S1). MiMBl captured interactions between genes that are significantly more distant than in case of FBA (∼40% more distant for negative epistasis, p-value = 0.022; ∼10% more distant for both positive and negative epistasis, p-value = 0.089; Fig. 4d, e).
In a metabolic network reconstruction, a group of isoenzymes is represented by a single reaction, which is associated with two or more genes. Simulation-wise, such a reaction will be inactive only when all of the corresponding isoenzyme-coding genes are deleted and deletion of any single gene will not result in a loss of fitness. Thus, in case of a reaction with two isoenzymes, when the deletion of both isoenzyme-coding genes leads to decreased fitness in silico, a negative interaction will be predicted. Our analysis captured several of such cases, for example, the negative interactions between SER3 and SER33, as well as between SAM1 and SAM2 (Fig. 4f). While this gene-deletion-centered approach allows capturing interactions between isoenzyme-coding genes, it is not suited for predicting interactions between two functionally different genes where one (or both) of them have isoenzymes. However, such interactions are often observed in vivo, since isoenzymes do not always completely compensate each other's function due to differences in kinetic and/or regulatory characteristics. Although these effects cannot be directly captured using the currently available metabolic modeling tools, we suggest evaluating the metabolic basis of genetic interactions between functionally different genes with isoenzymes by taking a reaction-centered approach. For this purpose, flux through reactions catalyzed by isoenzymes was constrained to zero when at least one of the isoenzyme-coding genes was deleted. This way, five additional genetic interactions involving isoenzymes were correctly captured: a positive interaction between the isoenzyme group TLK1 & TLK2 and the gene ARO1, as well as four negative interactions involving the isoenzyme group ALD2-ALD6 and other genes from the central carbon metabolism (Fig. 4f). These five interactions are thus likely to result from flux rerouting caused by the lack of compensation by the corresponding isoenzymes.
Use of MiMBl not only allowed us to expand the range of genetic interactions predicted by FBA, but also the combined use of these two complementary algorithms enabled finding of relevant interactions where only one or both simulation principles apply. For example, the interaction between PGK2 and GDH2, exclusively captured by MiMBl, is due to balancing of NADH and glutamate, two of the most connected metabolites in the network. As there are alternative pathways for fulfilling NADH and glutamate requirement (despite implying higher metabolic adjustments), FBA could not capture this interaction. A similar effect is observed for the negative interaction between isoenzymes SER3 and SER33. In the absence of both genes, FBA predicts the needed supply of serine to be totally fulfilled by rerouting the metabolic fluxes via the glyoxylate shunt and threonine biosynthesis. On the other hand, MiMBl predicts that the supply of serine will be shared between the two alternative pathways, but the rescue cannot be complete, because the corresponding metabolic adjustment cost overweighs the benefit of increased growth. This prediction is in very good agreement with the experimental verification that the double mutant growth is impaired and can be restored by adding glycine to the medium, which is the intermediate for serine synthesis via glyoxylate or threonine [27].
Overall, our results demonstrate that the use of optimization-based algorithms that are stoichiometry representation independent is fundamental for unambiguously linking modeling results with biological interpretation. To this end, we report a new method for formulating objective functions for metabolic modeling – MiMBl. As a biological case study, we used MiMBl to gain insights into the flux rewiring underlying genetic interactions within the yeast metabolic network. The analysis showed that the combined use of different objective functions is of primary importance in order to achieve a more complete understanding of the operating principles behind complex biological phenomena such as genetic interactions. Indeed, the number of accurately predicted genetic interactions was almost doubled owing to the use of MiMBl, highlighting the impact of metabolic adjustment constraints on the operation of perturbed metabolic networks. In conclusion, MiMBl provides a framework for consistent mathematical formulation of biological objective functions and thereby facilitates unraveling of the genotype-phenotype relations in metabolic networks.
The susceptibility of the modeling results towards the stoichiometry representation is inherent to the formulation of the objective function; and it is independent of the choice of metabolic network reconstruction. Two reconstructions were therefore selected in this study based on their suitability for addressing the biological principle in question. iFF708 [21] was the reconstruction of choice for illustrating the prediction of internal flux distribution and metabolic engineering targets, as i) this reconstruction has been successfully used for model guided metabolic engineering [13], [14] and, ii) when constrained with experimentally measured substrate and product exchange rates [28] (Text S1), iFF708 showed less flux variability at physiologically important flux nodes as opposed to more recent reconstruction iAZ900 [22] (Fig. S1). On the other hand, for studying large-scale genetic interactions in yeast, we used iAZ900 (manually curated from iMM904 [29]), as the maximum gene coverage overlap with the experimental dataset was the main criterion. Simulation conditions are provided in Text S1.
Normalized lMoMa was formulated as follows:Where N is the set of all reactions, M is the set of all intracellular metabolites, S is the stoichiometric matrix and vi is the flux for reaction i. WT stands for wild-type (or reference), vilb and viup are the lower and upper bounds for the flux of reaction i.
Metabolite turnover is defined as the sum of all fluxes producing (or consuming) it, multiplied by the stoichiometric coefficients:Nm is the subset of N producing or consuming metabolite m and αm,i is the stoichiometric coefficient of metabolite m in reaction i. Note that αm,i is always a positive number in the definition above, irrespective of m being a substrate or a product.
MiMBl was formulated as two sequential linear programming problems, as follows:
We note that MiMBl integrates reaction-to-metabolite turnover mapping into the model formulation in terms of defining biological objective functions and thereby making metabolite-usage a determinant for the prediction of metabolic phenotypes. This formulation is thus different from metabolite-centric approaches that have been proposed for interpreting simulation results [30]–[32].
Alternative stoichiometry representations were obtained by multiplying a given reaction (or a set of reactions) by a scalar θ (or a set of scalars). Consider reaction r , for which an equivalent representation is given by:where Y, X and Z represent the metabolites participating in reaction r and αYr, αXr, αZr represent the corresponding stoichiometric coefficients. Note that when the stoichiometry of reaction r is scaled by θ, the corresponding flux value will be scaled by 1/θ for the same optimal solution. For illustrating the impact of linear scaling of the reactions stoichiometry on the internal flux distribution, the reaction RPI1 of iFF708 model was divided by the scalar θ. For illustrating the impact of using alternative stoichiometry representations on the design of metabolic engineering strategies, two biochemically equivalent stoichiometric matrices were used: i) the as-published matrix from the yeast model (S0) and ii) an equivalent matrix (S1) where the stoichiometric coefficients of the reactions SERxtO, PDC6, FUR1, GAP1_21, PNP1_1, and CYSxtO were divided by 100, 100, 0.1, 0.01, 100 and 0.1, respectively. A third equivalent matrix (S2) was generated by dividing the coefficients of a single reaction (PGK1) by 0.1. The results of the comparison between S0 and S2 are presented in Fig. S6.
The impact of scaling the constraints of a given linear programming problem depends on whether such changes guarantee the optimality conditions after scaling. Consider the problem:Where is the cost coefficient of variable in the objective function. Here, a linear combination of non-normalized fluxes is used in the objective function, similarly to e.g. minimization of intracellular flux and lMoMA. Assuming that is an optimal basis matrix for the problem, the following optimization condition is satisfied:where is the index of variable in matrix , is the reduced cost of the variable , is the objective function coefficient of , is the vector containing the objective coefficients of basic variables and is the jth column of matrix [33]. Linear scaling the problem by the matrix will result in the following reduced cost for each variable:Where is a positive diagonal matrix (scaling matrix) and is the scaling factor for the jth column of matrix . In the cases of entries the corresponding columns of are accordingly scaled. Analogously, is the scaling matrix corresponding to the basic variables.
Unless all entries of are identical,Therefore the optimality condition is not guaranteed.
Corollary 1: When all (diagonal) entries of are identical (uniform scaling matrix), and therefore equal to , the optimality condition is simplified toThe same optimality condition can thus be guaranteed only when the matrix is uniformly scaled. Note that due to the nature of the biological problem, the genuine representation of might not be known, thereby cannot be guaranteed to be a uniform scaling matrix. More importantly, for metabolic modeling purposes (where flux units and ranges are problem dependent), it is nevertheless undesirable that the solution is sensitive to non-uniform scaling and thus context dependent.
Corollary 2: For any positive diagonal scaling matrix , the same optimality condition is still guaranteed if the cost coefficients vector is also scaled by . However, the choice of the appropriate for formulating a biologically meaningful problem will require existence of a unique representation of for any given network, which is not possible due to subjective nature of stoichiometry representation.
Now consider the following MiMBl-like formulated problem:Where, is cost coefficient of variable in the objective function. The new problem biologically corresponds to the previous one, after mapping the flux space into metabolite space. We term it as a MiMBl-like problem formulation.Recall that is the stoichiometric coefficient of metabolite in reaction . The objective function can be re-written as function of :Therefore, the objective function coefficient of each is a function of the stoichiometric coefficients :
Similarly to the previous problem, the following optimality condition is satisfied, so is an optimal solution.Scaling the optimality condition will result in:Unlike the previous situation (sum of fluxes in the objective function), using a MiMBl-like problem formulation guarantees that the optimality condition is always satisfied, independently of the stoichiometry representation.
The sensitivity of MiMBl and lMoMA towards the use of FBA alternative optima for wild-type flux distribution was evaluated by performing single gene deletion simulations while using 500 different flux distributions corresponding to alternative optima of the same FBA solution. FBA alternative optimal solutions were obtained following a Mixed Integer Linear Programming (MILP) routine similar to the one suggested by Lee et al. [34]. Flux variability analysis of the flux distributions obtained with MiMBl and lMoMA were performed according to the procedure suggested by Mahadevan et al. [26]: maximizing and minimizing internal fluxes after constraining the objective function to its optimal value. In case of MiMBl, this implies adding an additional constraint of the minimum Manhattan distance between the wild-type and the mutant metabolite turnovers. In case of lMoMA, the Manhattan distance between the mutant and the wild-type fluxes will have an upper bound. Growth is uniquely predicted if . Cases of = 0 were also treated as = 1, solely for the purpose of visualization (Fig. 3d).
|
10.1371/journal.ppat.1002913 | OX40 Facilitates Control of a Persistent Virus Infection | During acute viral infections, clearance of the pathogen is followed by the contraction of the anti-viral T cell compartment. In contrast, T cell responses need to be maintained over a longer period of time during chronic viral infections in order to control viral replication and to avoid viral spreading. Much is known about inhibitory signals such as through PD-1 that limit T cell activity during chronic viral infection, but little is known about the stimulatory signals that allow maintenance of anti-viral T cells. Here, we show that the co-stimulatory molecule OX40 (CD134) is critically required in the context of persistent LCMV clone 13 infection. Anti-viral T cells express high levels of OX40 in the presence of their cognate antigen and T cells lacking the OX40 receptor fail to accumulate sufficiently. Moreover, the emergence of T cell dependent germinal center responses and LCMV-specific antibodies are severely impaired. Consequently, OX40-deficient mice fail to control LCMV clone 13 infection over time, highlighting the importance of this signaling pathway during persistent viral infection.
| A robust T cell response is the hallmark of an effective immune response to a variety of invading viruses. In many acute infections, the clearance of the viral pathogen is associated with a short and vigorous T cell response followed by development of pathogen-specific immune memory. However, some viruses can establish persistent infection in their respective host, during which an ongoing T cell response is required in order to prevent overwhelming viral replication. Little is known about the factors that sustain the T cell response in the persistent phase of a viral infection. In this report, we demonstrate that ligation of the OX40 molecule, which is expressed on T cells targeting the virus, is critically required in order to sustain the anti-viral immune response. We show that virus-specific, OX40-deficient T cells fail to accumulate sufficiently and consequently, mice lacking the OX40 receptor are incapable of controlling viral replication. Collectively our data establish OX40 as a crucial signaling molecule during a persistent viral infection.
| Although persistent viral infections are typically associated with a dysfunctional and exhausted T cell signature [1], mice infected with the clone 13 (cl13) isolate of the lymphocytic choriomeningitis virus (LCMV) are able to control viral replication within 2–3 months post infection in a T cell dependent manner [2]–[7]. The inhibitory molecules involved in T cell exhaustion have been analyzed in great detail [1], [8]–[11]; yet, the signals that sustain T cell responses during persistent viral infections are not fully understood. This is a major concern, as, for example, the underlying cause for the loss of CD4 T cell responses during persistent viral infections in humans, such as chronic Hepatitis C Virus infection, is entirely unclear [12]. While recent studies demonstrated important roles for the Interleukins 21 and 6 in this context [2]–[4], [13], little is known about the role of co-stimulatory signals. OX40 (CD134) is a co-stimulatory molecule that has been shown to be important for T cell survival and function as well as establishment of T cell memory, although the degree to which OX40 influences immune responses is greatly context dependent [14]–[16]. While OX40 plays a role in driving T cell responses to several viruses [17]–[19], interestingly, it seems to be largely dispensable in the setting of acute LCMV infection. Although the LCMV-specific CD4 T cell responses in OX40-deficient mice are impaired, CD8 T cell responses, antibody titers and pathogen control are largely unaltered following acute LCMV infection [20]. However, the importance of OX40 during chronic viral infections remains unclear. Since OX40 signaling has the ability to promote long-term survival of T cells, we hypothesized that its biological relevance might be more prominent in the context of viral persistence. Indeed, in stark contrast to what has been observed during acute LCMV infection, we show that OX40 shapes both the CD4 and CD8 T cell response during persistent LCMV cl13 infection including T cell dependent antibody responses. The profound impact of OX40 expression in this context is highlighted by the observation that, in contrast to wild type mice, OX40-deficient mice are incapable of controlling viral replication.
Wild type (WT) and OX40-deficient mice (OX40−/−) on a C57BL/6 background were challenged intravenously with 2×106 PFU of LCMV cl13 [21]. LCMV cl13 infection induces severe immunopathology in infected mice, particularly within the first two weeks post infection, characterized by excessive production of pro-inflammatory cytokines and extensive weight loss [22]–[24]. Interestingly, OX40−/− mice had a much healthier appearance and lost significantly less weight following cl13 infection (Fig. 1A). Next, we analyzed the impact of OX40 on the LCMV-specific T cell response. Using MHC class I and II restricted multimers, we detected significantly lower numbers of LCMV-specific CD8 and CD4 T cells in OX40−/− mice (Figure 1, B and C). The magnitude of the anti-viral CD4 and CD8 T cell responses over time was assessed by epitope-specific intracellular cytokine staining (ICCS), focusing on the production of the key effector and immunostimulatory cytokines IFN-γ, TNF and IL-2. In MHC H-2Db mice, the CD8 T cell response during LCMV cl13 infection is characterized by functionally impaired but sustained responses against two glycoprotein (GP) epitopes (GP33 and GP276) and a loss of T cells targeting the nucleoprotein (NP)-396 epitope. While the deletion of the NP396-specific response occurred regardless of the presence of OX40 (Fig. 1F), we constantly detected greater numbers of cytokine-producing cells in WT animals compared to OX40−/− mice in response to stimulation with the GP33 and the GP276 epitope (Fig. 1, D and E). The differences were even more pronounced in the anti-viral CD4 T cell compartment, where we detected significantly lower numbers of IFN-γ and IL-2 producing CD4 T cells in the absence of OX40 (Figure 1, G–I). These findings clearly suggest that OX40 ligation has a strong role in shaping the pro-inflammatory virus-specific T cell response during LCMV cl13 infection.
Next, we wanted to assess whether the disrupted T cell response in OX40−/− mice would also impact T cell help to B cells. CD4 T cell interactions with B cells are required for the development of germinal centers (GCs) and the emergence of antibody producing plasma cells [25], [26]. Follicular T helper cells (Tfh) represent the key CD4 T cell lineage associated with germinal center responses and are characterized by the expression of high levels of CXCR5 and PD-1 [25]. Two recent studies demonstrated that the absence of Tfh cells disabled LCMV-specific antibody production in cl13 infection, resulting in an inability of these mice to control viral replication [5], [13]. In OX40−/− mice, we detected significantly reduced numbers of LCMV-specific Tfh cells as characterized by the expression of CXCR5 (Fig. 2, A–D). Consequently, numbers of germinal center B cells (Fig. 2, E and F) as well as plasma cells (Fig. 2, G and H) were profoundly decreased in OX40−/− mice, resulting in significantly lower titers of LCMV-specific antibodies (Fig. 2I).
While OX40−/− mice benefited clinically from the impaired T cell responses, as they lost markedly less weight than WT mice (Fig. 1A), we analyzed how this would affect their capability to control viral replication. To address this question, we quantified viral titers at various time points post infection in the serum and several solid organs by plaque assay. Interestingly, we found that OX40−/− and WT mice had very similar viral titers on days 10 and 20 post infection in most tissues (Fig. 3, A–D). However, while WT mice were able to control viral titers as soon as 50 days post infection in selected tissues such as the lung and serum, OX40−/− mice failed to control the virus to a similar degree. This difference was even more striking by days 80 and 120 post infection, when viral titers dropped below detection levels in most tissues in WT mice, whereas the majority of OX40−/− displayed high titers in most organs and serum (Fig. 3, A–D).
To analyze the OX40 expression kinetics on virus-specific CD4 and CD8 T cells during acute and persistent LCMV infection, we transferred naïve, congenically marked TCR transgenic (TCRtg) CD4 (smarta, smtg) and CD8 (P14) T cells into WT recipients and analyzed OX40 expression on days 3, 5, 7, and 20 post LCMV Armstrong or LCMV cl13 infection. While OX40 was not expressed on naïve P14 and smtg cells, we found that it was strongly induced on both populations 3 days following acute and persistent LCMV infection (Fig. 4, A and B). Interestingly, OX40 expression levels on both CD4 and CD8 T cells rapidly declined in acutely infected mice but were maintained in persistently infected mice. In order to analyze whether the sustained OX40 expression during persistent LCMV infection could be related to ongoing antigen recognition, we simultaneously analyzed PD-1 and OX40 expression on virus-specific T cells, since PD-1 expression has been shown to be directly linked to antigen-expression in this system [27]. Interestingly, PD-1 expression strongly correlated with OX40 expression during both acute and persistent LCMV infection suggesting that the sustained OX40 expression during LCMV cl13 infection is a consequence of ongoing antigen recognition (Fig. 4C) and in line with prior data showing that TCR signaling is sufficient to induce OX40 on T cells [28]. Although OX40 receptor expression was stronger on CD4 T cells, a subset of antiviral CD8 T cells also displayed sustained OX40 expression. Thus, we wanted to examine on which subpopulation within the antiviral CD8 T cells OX40 expression might be most relevant. Therefore, we looked at the expression of the inhibitory killer cell lectin-like receptor G1 (KLRG1) on CD8 T cells which has been associated with a short-lived effector phenotype [29], [30] and has previously been linked to OX40 [31]. Interestingly, we found that OX40 expression on CD8 T cells is largely restricted to the KLRG1-low compartment during LCMV cl13 infection (Fig. 4D), suggesting that OX40 preferentially acts to maintain higher numbers of these longer-lived effector T cells during the persistent phase of infection. Indeed, the analysis of the KLRG1-low and the KLRG1-high CD8 compartments revealed that the absence of OX40 had a significantly stronger impact on the accumulation of KLRG1-low P14 cells compared to KLRG1-high P14 cells (Fig. 4E).
Next, we analyzed which cells could be capable of engaging the OX40 receptor through expression of the OX40 ligand (OX40L). OX40L is the only ligand that is known to activate the OX40 receptor and is typically expressed on activated antigen presenting cells, but also has been visualized on lymphoid tissue inducer (LTi) cells and activated/inflamed endothelium [15]. First, we analyzed OX40L expression on splenocytes that were harvested 5 days post LCMV cl13 infection. Not surprisingly, OX40L expression was largely limited to professional antigen presenting cells, slightly detectable on B cells and almost absent on T cells (Fig. 4F). Next, we wanted to analyze the dynamics of OX40L expression following LCMV cl13 infection. Therefore we harvested and stained splenocytes from uninfected and LCMV cl13 infected mice on defined days post infection. A precise phenotypic analysis of antigen presenting cell subsets over time is very challenging in the LCMV cl13 model as those cells can dramatically change their appearance due to immune activation and direct infection of those cells. Thus, we analyzed OX40L expression independently on bulk CD11c+, CD11b+ and F4/80+ cells after exclusion of T and B cells. Interestingly, we observed that OX40L was strongly upregulated on F4/80 expressing cells following LCMV cl13 infection, whereas it was largely unaltered in CD11b and CD11c expressing cells (Fig. 4G). Given that the true extent of OX40 ligand expression is difficult to assess due to several modes of regulation typical in the TNF superfamily, including activation-induced cleavage, receptor-induced cleavage, and rapid turnover and internalization, these data suggest that macrophage/dendritic lineage cells may provide the ligand for T cell expressed OX40 but do not rule out other cell types as possible sources.
To assess the direct impact of OX40 expression on virus-specific CD4 and CD8 T cells, we crossed TCRtg CD4 (smtg) and CD8 (P14) T cells onto the OX40−/− background. We then co-transferred equal numbers of naïve WT (CD45.1+CD45.2−) and OX40−/− (CD45.1+CD45.2+) cells into C57BL/6 (CD45.1−CD45.2+) recipients prior to LCMV cl13 infection or LCMV Armstrong infection. Strikingly, WT CD4 smtg cells accumulated to much higher numbers compared to OX40−/− smtg cells (Fig. 5A). The number of WT smtg cells within the transferred population was approximately 4 fold increased compared to OX40−/− smtg cells as soon as day 8 p.i. (Fig. 5B). Following the initial contraction phase around day 10 p.i. this difference was approximately 10 fold (Fig. 5B). On day 20 p.i, the OX40−/− smtg compartment was almost entirely lost (Fig. 5, A and B). To address whether OX40 signals are dose dependent in this setting, we transferred equal numbers of WT (OX40+/+), heterozygous (OX40+/−) and OX40-deficient (OX40−/−) smtg cells into separate WT hosts. Indeed, OX40+/+ smtg cells expressing the highest levels of OX40 following LCMV cl13 infection accumulated significantly better than OX40+/− smtg cells and OX40−/− smtg cells (Fig. 5, C–E). When we performed similar co-transfer experiments with WT and OX40−/− CD8 TCRtg P14 cells, consistent with our findings on the CD4 level, we repeatedly harvested higher numbers of WT P14 cells compared OX40−/− P14 cells from spleens of persistently infected mice (Fig. 5F).
In order to directly compare the role of OX40 during acute and persistent LCMV infection, we performed co-transfer experiments with WT and OX40−/− TCRtg CD4 and CD8 T cells prior to LCMV Armstrong and cl13 infection. In agreement with the observation of reduced OX40 receptor expression during acute infection (Fig. 4, A and B), the percentage of OX40−/− CD4 and CD8 TCRtg cells within the ‘transferred cells’ gate was significantly lower in cl13 infected animals than in Armstrong infected animals (Fig. 5, G and H).
Signals through OX40 can influence several aspects of T cell biology, such as survival, proliferative capacities and the ability to secrete cytokines [15]. In order to define the precise role of OX40 on antiviral T cells during persistent infection, we first examined the functional capacities of WT and OX40-deficient cells in co-transfer experiments. We performed CFSE- and BrdU- labeling experiments to study the ability of WT and OX40−/− CD4 and CD8 TCRtg cells to proliferate. Those experiments revealed that T cell proliferation occurred independently of OX40 signals over the initial course of infection (Fig. 6, A–C), demonstrating that the lack of accumulation is unlikely to be a consequence of impaired cell division in this model. Secondly, we analyzed cytokine production of antiviral T cells since we found drastically reduced numbers of cytokine producing CD4 and CD8 T cells in OX40−/− mice (Fig. 1, D–I). Analysis of IL-2, IL-21 and IFN-γsecretion by WT, Het and KO smtg cells and IFN- γ secretion by WT and OX40−/− P14 cells revealed modest increases in the percentage of cytokine producing cells in the WT compared to Het and KO TCRtg T cells. However, these differences failed to consistently reach statistical significance, suggesting that OX40 signals are not required for direct anti-viral T cell function (Fig. 6, D and E). Similarly, we did not detect differences in IFN- γ secretion by WT and OX40−/− P14 cells (Fig. 6F). Lastly, we investigated whether OX40−/− cells were more prone to apoptosis, since OX40 has been shown to antagonize apoptotic cell death in T cells [14], [15]. Indeed, OX40-deficient T cells displayed significantly stronger Annexin V binding, suggestive of a higher number of apoptotic cells within the OX40-deficient compartment (Fig. 7, A and B). To examine what pathways might be involved, we focused on CD4 T cells and found significantly elevated levels of the pro-apoptotic Fas-receptor on days 5, 8 and 20 post infection on OX40-deficient cells (Fig. 7C) suggestive of negative regulation of Fas by OX40. Moreover, this was associated with higher levels of active caspase 3 in those cells (Fig. 7D). OX40 has also been shown to upregulate anti-apoptotic Bcl-2 family members, such as Bcl-2 and Bcl-xL in T cells [14]. OX40-deficient T cells fail to maintain Bcl-2/Bcl-xL levels and showing the significance of this, retrovirally introducing those proteins back into OX40-deficient CD4 and CD8 T cells restores their survival capacities [14], [32]. Expression of Bcl-2 family members is directly dependent on OX40 mediated NF-κB1 and Akt activity as OX40 deficient T cells display reduced activation of both NF-κB1 and Akt and restoring these activities by introducing active IκB kinase or active Akt rescues survival and Bcl-2/Bcl-xL expression in OX40 deficient T cells [33]–[35]. Indeed, when analyzing Bcl-2 and Bcl-xL expression on virus-specific CD4 T cells by tetramer and intracellular antibody-staining, we found significantly higher levels of Bcl-2 and Bcl-xL on GP66-specific CD4 T cells from WT animals compared to OX40−/− mice (Fig. 7E). Collectively, these findings indicate that the failure of OX40-deficient effector T cells to accumulate in high numbers is a result of reduced survival of these cells, rather than an inability to divide sufficiently or secrete anti-viral or immunostimulatory cytokines.
Co-stimulation of T cells is a central component of adaptive immunity. Numerous co-stimulatory pathways have been described and it has become evident that the biological relevance of each of those pathways is greatly dependent on the immunologic context [36], e.g. it has been shown that signaling through the co-stimulatory CD27/CD70 pathway can impair T cell responses during persistent infection while it positively regulates T cells during acute infection [37], [38]. Thus, it is of great interest to define those co-stimulatory pathways that dominantly influence adaptive immunity in a given disease model. In order to do so, we chose to use the murine LCMV system as it offers the possibility to directly compare an acute viral infection that is effectively cleared within 7–10 days (LCMV Armstrong) to a protracted infection that persists for several weeks in various tissues (LCMV cl13). Our findings demonstrate that OX40-deficiency profoundly impacted the anti-viral immune response during persistent LCMV cl13 infection.
Although the role of OX40 has never been studied in a persistent viral infection with ongoing viral replication, much is known about the mechanisms by which OX40 can influence T cell responses. It has been demonstrated that OX40 promotes T cell survival, division and function in various immune models, including cancer models [39]–[41], viral [17], [18] and bacterial [31] infection as well as autoimmunity [42], [43]. Interestingly, during persistent infection, the role of OX40 seems to be largely restricted to promoting T cell survival, as we have not observed differences in T cell proliferation or cytokine production in the absence of OX40. In previous publications it has been shown that the ability of OX40 to promote T cell survival is based on its capacity to directly recruit and activate various intracellular signaling pathways, such as PI3K and Akt/PKB or NF-κB [33]–[35], [44]. These pathways control the expression of anti-apoptotic members of the Bcl-2 family and T cells that lack OX40 express reduced levels of Bcl-2, Bcl-xL, and Bfl-1 [14], [32]. Restoration of NF-κB activity or Akt activity through retroviral introduction of active IκB kinase or active Akt rescues survival and Bcl-2/Bcl-xL expression in OX40 deficient T cells [33]–[35]. We show here that OX40-deficient cells fail to accumulate, are more prone to apoptosis and display reduced levels of Bcl-2 and Bcl-xL, strongly suggesting that the OX40 induced signaling axis that controls the expression of anti-apoptotic molecules is functional in the context of persistent viral infection.
Although OX40 has primarily been associated with CD4 T cell function, it became evident in recent years that it can also strongly influence CD8 T cells [15], [17]. We observed much higher levels of OX40 expression on the anti-viral CD4 T cells compared to CD8 T cells, suggesting that OX40 might predominantly shape the CD4 response in this context. Consequently, the impaired CD8 T cell and B cell responses in the OX40-deficient mice could have been a result of insufficient CD4 T cell help as CD4 T cells are key players in providing help to both CD8 T cells and B cells. However, our adoptive cell-transfer model clearly demonstrates a crucial requirement of OX40 on both CD4 and CD8 T cells in order to accumulate sufficiently, implying direct OX40 signaling to CD8 T cells is likely important.
Another observation of our study that is noteworthy is how OX40 deficiency impacts the antiviral immune response already at early stages post infection, however, loss of viral control in those mice occurs in a delayed fashion. The notion that changes in the immune response can precede differences in viral titers in the LCMV cl13 system has been described before [4], [5], [45]. Indeed, even though anti-viral T cell numbers and LCMV-specific antibody-titers were markedly different as soon as day 10 and 20 post infection, this had no immediate impact on viral control and viral titers initially declined in both WT and OX40-deficient mice. Moreover, OX40-deficient mice had a clinical benefit from the weak immune response and displayed reduced signs of immunopathology. Thus, it seems that the amount of T cells that accumulates in WT mice in an OX40-dependent manner during early stages exceeds the number that is required for initial virus control. And while this over-abundance of pro-inflammatory effector T cells mediates immunopathology, it does enable stronger T cell pressure on viral replication over time, which eventually facilitates virus control.
While OX40 has previously been shown to positively regulate antigen-specific immune responses, the degree to which OX40 influences adaptive immunity and, importantly, facilitates virus control in the context of persistent infection is remarkable. Indeed, the strong impact of OX40 on the cellular and humoral immunity in the persistent LCMV-system is in contrast to previous findings in the acute LCMV system, where the absence of OX40 primarily affected the magnitude of the CD4 T cell response but did not have an impact on the antiviral CD8 T cell response, antibody titers and virus control [20]. Since our data suggest that OX40 expression is associated with antigen recognition, the rapid elimination of antigen in the acute infection model might be responsible for the loss of OX40 expression and thus, the reduction of biological relevance of OX40 signaling during acute infection. Those observations demonstrate different OX40 requirements for T cell accumulation during acute vs. persistent LCMV infection and, in addition, support our previous finding that OX40 usage may be dictated by the virulence of an invading pathogen [16].
While CD8 T cell responses have long been known to be critical for control of persistent LCMV-infection, the relevance of antibody responses in this context may have been underappreciated. Importantly, Pinschewer and colleagues showed that mice that were unable to produce LCMV-specific antibodies failed to control infection [46] and more recently, striking defects in control of LCMV cl13 were observed when antibody responses were reduced due to defective follicular T helper cell responses [5], [13]. These studies have established a strong role for T cell dependent humoral immunity in the control of persistent viruses and one study has demonstrated an important role of IL-6 in this context [13]. Our work, which we report in this manuscript, extends those observations and clearly establishes a critical role for OX40 in sustaining follicular T helper cell responses in the context of persistent viral infection.
Importantly, these findings could open new paths in the understanding of T cell failure during persistent viral infections in humans, i.e. chronic HCV infection. Particularly HCV-specific CD4 responses are very weak in persistently infected individuals [47], [48] and recently, it has been shown that CD4 T cell priming and accumulation occurs in the acute phase of HCV infection. However, in many cases, those CD4 responses rapidly disappear, severely increasing the risk of developing persistent viremia [12]. The rapid loss of CD4 responses in OX40 deficient mice is somewhat reminiscent of this observation, suggesting that OX40 might be an interesting molecule to study in those patients. The findings regarding the role of OX40 in the antibody response to a persisting virus might also be relevant to HIV infection in humans, as there is an intense interest in broadly neutralizing antibodies and why they only develop in a small subset of HIV-infected individuals [49], [50].
Collectively, our findings establish OX40 as a key factor in sustaining the cellular and humoral immunity during viral persistence and have important implications for the study of T cell dysfunction in persistent viral infections in humans.
C57BL/6 and B6.SJL mice were purchased from The Jackson Laboratory and housed at the La Jolla Institute (LIAI) as well as OX40−/− mice on a C57BL/6 background [14]. LCMV-GP61–80-specific CD4 TCRtg (smtg) and LCMV-GP33–41-specific CD8 TCRtg mice (P14) were bred on an OX40−/− background. 5–8 week old, age and sex matched mice were infected intravenously with 2×106 PFU of LCMV cl13 or intraperitoneally with 2×105 PFU of LCMV Armstrong. All animal experiments were performed in strict accordance with the PHS Policy on Humane Care and Use of Laboratory Animals of the National Institutes of Health. All procedures were approved by and comply with the regulations of the La Jolla Institute Animal Care Committee (AP152-MvH6).
Surface staining was performed on splenic single cell suspension with fluorescently-labeled or biotinylated antibodies against CD4, CD8, CD19, B220, CD150 (SLAM), CXCR5, PD-1, KLRG1, FAS, IgD, CD138, CD45.1, CD45.2, OX40L and fluorescently-labeled PNA as well as GP33-pentamers and GP66-tetramers. Streptavidin- APC and -PE were used to stain biotinylated antibodies. Bcl6, Bcl-2, Bcl-xL and active caspase 3 were stained intracellularly after permeabilization with Cytofix/Cytoperm (BD). For most experiments, LIVE/DEAD (Invitrogen) viability dye was used to exclude dead cells.
For ICCS, cells were stimulated with CD8- (GP33, GP276) and CD4-restricted (GP61) LCMV-epitopes (10 µg/ml) for five hours at 37°C in supplemented RPMI-medium (Invitrogen). Cells were permeabilized using Cytofix/Cytoperm (BD) and stained with fluorescently-labeled antibodies against IL-2, IFN-γ, and TNF. IL-21 staining was performed as described previously [51] using an IL-21R-FC chimeric protein (R&D Systems) and a PE-labeled anti-human IgG (Jackson Immunoresearch). All samples were acquired on a LSRII (BD) using DIVA software (BD) and analyzed using FlowJo software (Tree Star). Antibodies were obtained from BD, BioLegend or eBioscience.
Spleens were harvested from TCRtg mice (smtg or P14) on a WT and OX40−/− background. CD4 (smtg) and CD8 (P14) isolation was performed using purified rat antibodies against B220, CD11c, CD11b, CD16/32, I-A/I-E and CD4 or CD8 and sheep anti-rat Dynabeads (Invitrogen). For co-transfer experiments into the same or separate hosts, equal numbers of OX40+/+(WT), OX40+/−(Het) and/or OX40−/−(KO) cells (2,000 P14, 5,000 smtg; 1,000,000 for d3 analysis) were transferred into CD45.1/2 mismatched recipients. Cells were labeled with CFSE (BioChemika) prior to transfer in selected experiments. For analysis of proliferation at later time points we used the BrdU-Proliferation kit from BD.
Monolayers of Vero cells (ATCC #CRL-1587) were exposed to tissue homogenate or serum of individual mice in 1/10 dilutions. Plaque counts were performed 5 (organs) or 6 (serum) days post infection. Vero cells were cultured and maintained in supplemented DMEM (Invitrogen) at 37°C.
Elisa plates (Nunc) were coated overnight with LCMV infected cell lysate. LCMV-specific IgG was detected in serum samples using HRP goat anti-mouse IgG (Invitrogen). SureBlue Reserve TMB Kit (KPL) functioned as substrate. Absorbance was detected by a Spectra Max M2e (Molecular Devices).
Statistical analyses were performed and line art figures were designed using GraphPad Prism 5 software (GraphPad). If not specified otherwise, the unpaired, two tailed Student's t-test was used. For viral loads, the log rank test was applied with a defined endpoint of viral titers below the limit of detection. All error bars are SD. Values of p<0.05 were considered significant. *p<0.05, **p<0.01 and ***p<0.001.
|
10.1371/journal.ppat.1006689 | Co-opting ATP-generating glycolytic enzyme PGK1 phosphoglycerate kinase facilitates the assembly of viral replicase complexes | The intricate interactions between viruses and hosts include exploitation of host cells for viral replication by using many cellular resources, metabolites and energy. Tomato bushy stunt virus (TBSV), similar to other (+)RNA viruses, induces major changes in infected cells that lead to the formation of large replication compartments consisting of aggregated peroxisomal and ER membranes. Yet, it is not known how TBSV obtains the energy to fuel these energy-consuming processes. In the current work, the authors discovered that TBSV co-opts the glycolytic ATP-generating Pgk1 phosphoglycerate kinase to facilitate the assembly of new viral replicase complexes. The recruitment of Pgk1 into the viral replication compartment is through direct interaction with the viral replication proteins. Altogether, we provide evidence that the ATP generated locally within the replication compartment by the co-opted Pgk1 is used to fuel the ATP-requirement of the co-opted heat shock protein 70 (Hsp70) chaperone, which is essential for the assembly of new viral replicase complexes and the activation of functional viral RNA-dependent RNA polymerase. The advantage of direct recruitment of Pgk1 into the virus replication compartment could be that the virus replicase assembly does not need to intensively compete with cellular processes for access to ATP. In addition, local production of ATP within the replication compartment could greatly facilitate the efficiency of Hsp70-driven replicase assembly by providing high ATP concentration within the replication compartment.
| Positive-strand (+)RNA viruses build replication organelles with the help of many co-opted cellular factors and by usurping cellular metabolites and energy, which frequently lead to disease states in plants, animals and humans. The authors discovered that tomato bushy stunt virus (TBSV) co-opts the glycolytic ATP-generating Pgk1 phosphoglycerate kinase to facilitate the intracellular assembly of new viral replicase complexes. The ATP generated by co-opted Pgk1 within the replication compartment is used to fuel the ATP requirement of co-opted cellular heat shock protein 70 chaperone, which is essential for the assembly of new TBSV replicase complexes. Direct recruitment of Pgk1 by TBSV might provide easy access to ATP. Metabolic reprogramming of the glycolytic pathway might be a widespread phenomenon during various viral infections.
| Positive-strand (+)RNA viruses build robust viral replication machineries, called replication organelles, with the help of many co-opted cellular factors. In addition, viruses also obtain metabolites and energy from the infected cells. Overall, (+)RNA viruses induce major metabolic and structural changes in the infected cells, which frequently lead to disease states [1,2,3,4,5].
One of the best characterized (+)RNA virus is TBSV, which induces large replication compartments consisting of aggregated peroxisomal and ER membranes. The replication compartments contain numerous vesicle-like structures in the limiting membrane of peroxisomes, which harbor the viral replicase [6,7,8]. TBSV also manipulates the cytoskeleton and endosomal trafficking [9,10]. TBSV co-opts numerous host proteins to support various viral functions, including the heat shock protein 70 (Hsp70), the endosomal sorting complex required for transport (ESCRT) machinery, translation factors and DEAD-box RNA helicases [11,12,13,14,15,16,17]. TBSV also induces enrichment of sterols at the viral replication sites via co-opting oxysterol-binding proteins and membrane contact sites and retargets phosphatidylethanolamine to the replication sites to build suitable membranous subcellular environment for replication [8,9,18]. TBSV-induced major changes in the cell require energy, but how TBSV can obtain the energy to fuel these energy-demanding processes is not known.
Glycolysis is an essential and highly conserved energy-producing pathway in the cytosol. Glucose is converted into pyruvate by a ten-enzyme catalyzed reaction that produces ATP and NADPH. The two ATP-generating enzymes in the glycolytic pathway are Pgk1 phosphoglycerate kinase and Cdc19 pyruvate kinase (PKM2/PKLR in humans). Pgk1 produces ATP from ADP and DPG using substrate-level phosphorylation in the cytosol [19]. Pgk1 is highly conserved and present in every organism. In humans, Pgk1 is also important for tumor growth and DNA replication/repair and mutations in Pgk1 are associated with myopathy, mental disorders and hemolytic anemia [20,21,22]. Pgk1 is also important for the activation of nucleotide-based anti-HIV drugs [23].
Previous genome-wide screens in yeast model host have identified several components of the glycolytic pathway that affected TBSV replication [24,25,26]. Proteomic screens also identified Pgk1 phosphoglycerate kinase, which was pulled-down with the tombusviral p92 RdRp and interacted with p33 replication protein [27,28]. Therefore, we decided to characterize the putative role of Pgk1 in tombusvirus replication in this work. We discovered that TBSV co-opts the glycolytic Pgk1 through direct interaction with the viral replication proteins. Altogether, we provide evidence that the ATP generated locally in the viral replication compartment by the co-opted Pgk1 is used to fuel the ATP requirement of the co-opted cellular Hsp70 molecular chaperones. The functions of co-opted Hsp70 are to facilitate the assembly of new viral replicase complexes and the activation of the viral RNA-dependent RNA polymerase. The local production of ATP by Pgk1 within the replication compartment could greatly facilitate the efficiency of Hsp70-driven replicase assembly by providing high ATP concentration within the replication compartment.
Among the surprises from high-throughput screens with TBSV and yeast model host were the identification of the glycolytic Pgk1 as a putative co-opted host factor in TBSV replication [24,25,26,29]. To demonstrate interaction between Pgk1 and the viral replication proteins, first we performed co-purification experiments from yeast replicating TBSV repRNA. The isolated membrane fraction was solubilized with detergent and the FLAG-tagged p33 and FLAG-p92pol replication proteins were affinity-purified, followed by Western-blot analysis. We found that the His6-tagged Pgk1 expressed from a plasmid was co-purified with the tombusvirus replicase (Fig 1A, lane 2). In contrast, Pgk1 was not co-purified with Flag-GFP control from yeast (S1 Fig). To examine if Pgk1 is a permanent component of the tombusvirus replicase, we halted the formation of new tombusvirus replicase complexes by shutting down p33 and p92pol expression and stopping ribosomal translation by adding cycloheximide [7]. FLAG-affinity-purification of the tombusvirus replicase from the membrane fraction of yeast at various time-points showed the rapid release of Pgk1 from the replicase (Fig 1B, lanes 3–4 versus 2). Thus, Pgk1 seems to be a temporarily co-opted factor in the tombusvirus replicase complex.
To obtain additional evidence of subversion of Pgk1 into the tombusvirus replication compartment, we performed confocal laser microscopy experiments in yeast expressing GFP-p33 and BFP-Pgk1. Interestingly, we observed the robust recruitment of the cytosolic Pgk1 into the tombusvirus replication compartment, which was marked by Pex13-RFP peroxisomal marker protein (Fig 2A). Similar experiments with ectopically expressed BFP-NbPgk1 and p92-YFP showed the enrichment of Pgk1 in the tombusvirus replication compartment in Nicotiana benthamiana cells replicating the TBSV repRNA (Fig 2B). In addition, bimolecular fluorescence complementation (BiFC) assay further confirmed the recruitment of the glycolytic NbPgk1 via interaction with p33 into the peroxisomal TBSV replication compartment (Fig 2C). Similarly, we observed that the interaction between TBSV p92pol replication protein and the glycolytic NbPgk1 takes place in the large replication compartment, which was marked by RFP-SKL peroxisomal luminal marker protein (Fig 2C, lower panels). Therefore, we conclude that TBSV recruits the cytosolic Pgk1 through direct interaction with the viral replication proteins into the viral replication compartment in both yeast and plant cells.
To test if the recruitment of Pgk1 into the viral replication compartment plays a role in TBSV replication, we have made a haploid yeast strain, in which the wt PGK1 gene was replaced with the HA-tagged PGK1. In addition, expression of HA-PGK1 was placed under the regulation of GALS promoter, thus allowing induction by the addition of galactose and repression by addition of glucose to the culture media [30]. Induction of TBSV repRNA replication in GALS::PGK1 yeast with suppressed PGK1 expression showed ~4-fold reduction in TBSV repRNA replication (Fig 3A, lanes 7–9 versus 10–12, and S2 Fig). Expression of Pgk1 from a plasmid in GALS::PGK1 yeast with suppressed Pgk1 expression from chromosome increased TBSV repRNA replication by ~2-fold in glucose-containing media, while did not increase it on the non-fermentable glycerol-media (Fig 3B, lanes 7–10). These data indicate that Pgk1 plays a pro-viral function during TBSV replication in yeast.
Over-expression of either yeast Pgk1 (Fig 3C) or the N. benthamiana cytosolic NbPgk1 (Fig 3D) led to ~3-fold and a ~2.5-fold increase, respectively, in TBSV repRNA accumulation in wt yeast, confirming a pro-viral function for Pgk1 in TBSV replication.
Estimation of PGK1 mRNA levels in TBSV-infected N. benthamiana leaves indicated ~2-fold increased level of expression (Fig 4A). Similarly, Western blotting demonstrated a ~50% increase in accumulation of Pgk1 protein in yeast cells replicating TBSV repRNA (Fig 4B), suggesting that a tombusvirus can mildly induce the accumulation of a glycolytic enzyme in cells.
To confirm the importance of Pgk1 in TBSV replication in a plant host, we knocked-down via virus-induced gene silencing (VIGS) the expression of the cytosolic Pgk1 in N. benthamiana and the Pgk1-silenced leaves were inoculated with the full-length infectious TBSV. TBSV genomic RNA accumulation was inhibited by ~4-fold in these Pgk1 knock-down plants (Fig 4C). Inoculation of Pgk1-silenced N. benthamiana with the similar peroxisome-replicating cucumber necrosis virus (CNV) (S3 Fig) and the mitochondrial-replicating carnation Italian ringspot virus (CIRV) showed a ~6-fold decrease in the accumulation of genomic RNAs (Fig 4D and 4E). Thus, the cytosolic Pgk1 is a critical pro-viral host factor for various tombusviruses in a plant host.
Over-expression of NbPgk1 in N. benthamiana replicating TBSV RNA showed a close to 3-fold enhanced TBSV accumulation (Fig 4F), suggesting that Pgk1 level affects tombusvirus replication. Altogether, these data have confirmed the pro-viral function of Pgk1 in tombusvirus replication in plants.
Based on the above results, we assumed that the co-opted Pgk1 likely generates ATP within the viral replication compartment. Therefore, we estimated the local ATP level within the replication compartment by using the p33 replication protein tagged with ATeam, a cellular ATP-sensor module. ATeam module can measure ATP levels via FRET due to the conformational change in the enhanced ATP-binding domain of the bacterial ATP synthase upon binding to ATP without ATP hydrolysis (Fig 5A) [31]. In the ATP-bound form, the ATP-sensor module brings the CFP and YFP fluorescent tags into proximity, increasing FRET, which can be detected by confocal laser microscopy. In the ATP-free stage, the extended conformation of the ATP-sensor module places CFP and YFP at distal position, leading to low FRET signal. The ATeam-tagged p33 localizes to the aggregated peroxisomes that represent the sites of replication (Fig 5B).
We found by intracellular expression of p33-ATeam that the local ATP level within the TBSV replication compartment was decreased by ~6-fold in Pgk1 knock-down in N. benthamiana leaves in comparison with control leaves (Fig 5B). Interestingly, the mitochondria-replicating CIRV also accumulated a ~10-fold lower level ATP within the replication compartment in Pgk1 knock-down leaves than in control leaves (Fig 5C). Therefore, we conclude that Pgk1 is recruited to the sites of tombusvirus replication to generate high concentration of ATP within the replication compartment.
To decipher the function of the co-opted Pgk1 during TBSV replication, we tested viral RNA levels in yeast replicating TBSV repRNA. Down-regulation of Pgk1 expression reduced both (+)- and (-)RNA levels by ~4-fold (Fig 6A). These data suggest that Pgk1 likely plays an early role, prior to viral RNA synthesis, possibly during replicase assembly steps.
To test this scenario, we performed various in vitro assays. First, to further examine if Pgk1 was required for viral RNA replication, we obtained cell-free extracts (CFE) from wt yeast and GALS::PGK1 yeast with suppressed Pgk1 expression to reconstitute the TBSV replicase in vitro based on purified recombinant TBSV proteins. The yeast CFE-based assay, which supports a single full cycle of viral RNA replication [32], showed ~7-to-10-fold reduced repRNA production when low level of Pgk1 was produced in GALS::PGK1 yeast (Fig 6B, lanes 3 versus 1 and 4, also S4 Fig). Interestingly, both (+)RNA product (Fig 6B) and dsRNA replication intermediate obtained with CFE from GALS::PGK1 yeast with depleted Pgk1 in comparison with the wt yeast CFE were decreased by~7-fold (Fig 6C, lanes 7–8 versus 3–4). Therefore, the in vitro replication results support the important role of Pgk1 in viral RNA replication.
We also obtained purified replicase preparations from GALS::PGK1 yeast with suppressed Pgk1 expression versus comparable preparations from wt yeast, which were programmed with viral RNA transcripts [33]. In spite of having comparable amounts of replication proteins, the purified replicase preparations from Pgk1 depleted yeast showed ~3-fold reduced activity in comparison with the similar preparations obtained from wt yeast (Fig 6D).
Since the activity of the purified replicase preparations mostly depends on the efficiency of replicase assembly and replication protein activation, we compared the efficiency of p92pol activation by the supernatant fraction of CFEs prepared from Pgk1 depleted or wt yeasts. These assay revealed ~3-fold less RdRp activity of the purified recombinant p92pol mutant by the preparation from Pgk1 depleted than from wt yeasts (compare lanes 1–2 and 5–6 in Fig 6E). The major active component of the supernatant fraction of CFEs is Hsp70 molecular chaperone, which is essential for the activation of p92pol RdRp in vitro [29]. Accordingly, addition of purified Ssa1 (yeast Hsp70) to the WT fraction has increased the efficiency of p92pol RdRp mutant activation by ~5.5-fold, whereas the comparable assay with Ssa1 involving the preparation from Pgk1 depleted yeast showed only ~2-fold enhancement (compare lanes 3–4 and 7–8 in Fig 6E). Therefore, the in vitro results from multiple assays support the model that Pgk1 has a major role in tombusvirus RNA replication and Pgk1 likely fuels the energy requirements of co-opted Hsp70 molecular chaperones to facilitate the efficient replicase assembly and replication protein activation within the viral replication compartment.
TBSV co-opts Vps4p AAA+ ATPase, which is an ATP-dependent host protein, into the viral replicase complex to facilitate replicase assembly [7]. Vps4p is an ESCRT protein involved in membrane deformation, which is necessary for formation of TBSV-induced individual spherules (vesicle-like structures) supporting viral RNA replication [7]. The spherule formation protects the viral dsRNA replication intermediate from the innate RNAi machinery during infection [34]. To test if Pgk1-produced ATP might support Vps4p ATPase function during replicase assembly, we utilized a recently developed intracellular probe based on a reconstituted RNAi system in yeast (Saccharomyces cerevisiae), which lacks the RNAi machinery. The reconstituted RNAi machinery from S. castellii, which consists of the two-component DCR1 and AGO1 genes [35], is a simple, and easily tractable system [34]. We found that the induction of RNAi activity had only slightly more inhibitory effect on TBSV RNA accumulation in Pgk1-depleted yeast than in wt yeast (Fig 6F). Based on these results, the structure of the assembled tombusvirus replicase in Pgk1 depleted yeast might not be different from those assembled in wt yeast. Therefore, likely the co-opted ESCRT machinery and the Vps4p ATPase activity might not be affected when Pgk1 is depleted. Alternatively, the residual Pgk1 still present in GALS::PGK1 yeast could provide enough ATP for Vps4p ATPase activity during the replicase assembly process.
TBSV replication, similar to other (+)RNA viruses, is a very intensive and robust process that likely depends on consumption of a large amount of ATP. By recruiting the glycolytic Pgk1 into the virus replication compartment, Pgk1 could efficiently supply ATP to facilitate various steps in the replication process, including the assembly of the viral replicase complex. Accordingly, we show that TBSV actively recruits the cytosolic Pgk1 to the sites of viral replication through direct interaction with the viral replication proteins in both yeast and plant cells. It seems that a portion of recruited Pgk1 is getting released from the sites of replication when the formation of new viral replicases is halted via inhibition of translation. This observation suggests that the ATP produced by the co-opted Pgk1 is likely used up locally to fuel early steps in virus replication, such as viral replicase assembly.
The proposed role of the co-opted Pgk1 during the early steps in virus replication is further supported by additional observations. For example, down regulation of Pgk1 level in yeast affected both (-) and (+)-strand RNA levels and also decreased the in vitro activity of the purified replicase preparations obtained from yeast. Moreover, the in vitro assembly of functional TBSV replicase in CFEs from yeast with depleted Pgk1 level was inefficient, resulting in reduced production of both dsRNA intermediate and new (+)RNA progeny. All these results point at deficiency in viral replicase assembly when Pgk1 is depleted. We have shown previously that the assembly of the tombusvirus replicase requires the co-opted cellular Hsp70 that uses ATP for its molecular chaperone function [32,36,37]. Based on the above observations, we propose that the ATP generated by the recruited Pgk1 serves the energy need of co-opted Hsp70 to drive efficient replicase complex assembly within the elaborate replication compartment (Fig 6G).
In addition, we obtained evidence that the functional activation of the TBSV p92 RdRp, which also requires Hsp70 molecular chaperone, depends on the ATP generated by the recruited Pgk1. The initially inactive p92 RdRp becomes functional during replicase assembly in the presence of p33 replication protein, the viral (+)RNA, ER membrane and the ATP-dependent Hsp70 [27,29]. In a simplified RdRp activation assay, we found that CFEs obtained from yeast with depleted Pgk1 were inefficient in promoting p92 RdRp activity. The addition of purified recombinant Hsp70 to the above assay could stimulate p92 RdRp activity to lesser extent in case of depleted Pgk1 than in the presence of wt yeast CFEs. Therefore, we propose that a major function of the Pgk1-generated ATP is to provide fuel to the co-opted Hsp70 during p92 RdRp activation and the assembly of viral replicase complexes (Fig 6G). The advantage of co-opting Pgk1 to the replicase complex could be that the energy hungry virus replicase assembly process does not need to intensively compete with cellular processes for access to plentiful ATP. Also, local production of ATP within the replication compartment could greatly facilitate the efficiency of Hsp70-driven replicase assembly by providing high ATP concentration within the replication compartment. There is previous evidence that Pgk1 could provide ATP to enhance the chaperone activity of Hsp90 that leads to multistress resistance [21].
In addition to Hsp70 molecular chaperone, TBSV also co-opts additional ATP-dependent host proteins into the viral replicase complex, including Vps4p AAA+ ATPase, involved in replicase assembly [7], and two DEAD-box helicases [38,39]. Using a reconstituted RNAi-based molecular probe, we found only a slightly less RNA protection in Pgk1-depleted yeast than in wt yeast (Fig 6F), which is in contrast with the poor TBSV RNA protection provided by VRCs assembled in the absence of ESCRT components [34]. Therefore, we propose that Vps4p AAA ATPase still functions efficiently enough in VRC assembly in yeast with depleted Pgk1 and ATP levels. On the other hand, the DEAD-box helicases, however, selectively affect (+)-strand synthesis, while Pgk1 affects both (-) and (+)-strand synthesis in CFE-based assay and in yeast to a similar extent. TBSV also affects the actin network, which requires ATP for functions, but it is currently on open question if the co-opted Pgk1 supplies ATP for the subverted actin network. Therefore, based on the collected data, we propose that the ATP-generated by Pgk1 within the replication compartment is primarily used by tombusviruses to provide ATP for the co-opted Hsp70 chaperone to support efficient assembly of the numerous viral replicase complexes.
Similar to our current findings, cancer cells also use glycolysis to efficiently generate ATP, even in the presence of oxygen [22]. Up-regulation of glycolytic machinery promotes rapidly proliferating cancer cells survival. Up-regulation of glycolytic pathway also takes place in effector T cells and it is required for immune cells activation [20,40]. Apparently, the relatively inefficient glycolytic pathway can provide enough ATP when up-regulated in these cells.
Several viruses are also known to reprogram the glycolytic pathway during infections based on metabolomic profiling [41,42,43], which unraveled enhanced glucose uptake into the infected cells. The hexokinase activity is increased in hepatitis C virus (HCV) or Dengue virus-infected cells [41,44]. ATP was shown to accumulate at the sites of HCV replication, likely to satisfy the energy demand of virus replication [45]. Whereas TBSV exploits glycolytic enzymes, such as PGK1 (this work) and Glyceraldehyde 3-phosphate dehydrogenase (GPDH, called Tdh2p and Tdh3p in yeast) for pro-viral functions [13,15], replication of a plant potexvirus is inhibited by GAPDH due to its RNA-binding function [46]. Altogether, metabolic reprogramming of the glycolytic pathway might be a widespread phenomenon during various viral infections.
In addition to its role in ATP production during glycolysis, Pgk1 seems to have additional roles in virus replication. For example, Sendai virus, a negative-strand RNA virus, co-opts Pgk1 to promote viral mRNA synthesis, albeit its enzymatic activity is not required for the stimulation of RNA synthesis [47]. A partial/recessive resistance in Arabidopsis against potyviruses is due to a single amino acid mutation in the conserved N-terminal portion of the chloroplast phosphoglycerate kinase (cPGK2), which is a nuclear gene encoding cPgk2 targeted to the chloroplast [48,49]. In addition, cPgk1 was found to bind to the 3’UTR of the viral RNA and it is involved in the localization of potexvirus RNA to the chloroplasts, which is important for virus accumulation [50,51].
TBSV RNA replication in yeasts and plants was analyzed after total RNA extraction with Northern blot analyses as described previously [52]. Briefly, BY4741 and GalS::PGK yeast strains were co-transformed with pHisGBK-CUP1- Hisp33/ADH-DI-72 and pGAD-CUP1-His92 [24]. The transformed yeast strains were grown at 23°C in SC-HL− (synthetic complete medium without histidine and leucine) media supplemented with 2% raffinose with or without 2% galactose and BCS for 24 h at 23°C. Then, yeast cultures were re-suspended in SC-HL− medium supplemented with 2% raffinose and with or without 2% galactose and 50 μM CuSO4. Yeasts were grown at 23°C for 16 h before being collected for total RNA extraction.
Yeast cells were grown as described above for Northern analysis. Total proteins were isolated by the NaOH method as described previously [53]. The total protein samples were analyzed by SDS-PAGE and Western blotting with anti-His and anti-PGK antibodies, followed by alkaline phosphatase-conjugated anti-mouse secondary antibody (Sigma) as described previously [54].
To examine the subcellular localization of Pgk1 in plants, N. benthamiana leaves were co-infiltrated with Agrobacterium carrying plasmids pGD-p92-YFP and pGD-BFP-PGK (OD600 of 0.5, each) together with pGD-35S::p19, pGD-DI-72 and pGD-p33. After 48 h, the agroinfiltrated leaves were subjected to confocal microscopy (Olympus America FV1000) using 405 nm laser for BFP and 488 nm laser for YFP. Images were taken successively and merged using Olympus FLUOVIEW 1.5 software.
To identify interactions between NbPgk1 and TBSV p33 replication proteins, the plasmids pGD-p33-cYFP, pGD-nYFP-PGK and pGD-nYFP-MBP were transformed to Agrobacterium strain C58C1. These Agrobacterium transformants were used to co-agroinfiltrate the leaves of four weeks-old N. benthamiana plants. Transformed leaves were subjected to confocal laser microscopy after 48 h using Olympus FV1000 microscope as described previously [13].
To detect the ATP levels within the tombusvirus replication compartments in plant cells, in the case of TBSV, PGK-silenced plants or control plants were agroinfiltrated with plasmids pGD-p33-ATeamYEMK, pGD-DI-72, pGD-35S::RFP-SKL, pGD-35S::p19 and pGD-p92. In the case of CIRV, the leaves were agroinfiltrated with pGD-p36-ATeamYEMK, pGD-35S::AtTim21-RFP, pGD-35S::p19, pGD-DI-72 and pGD-p95. The images were taken at 2.5 or 3.5 days post-agroinfiltration and analyzed with the method described previously [31]. Confocal FRET images were obtained with an Olympus FV1000 microscope (Olympus America). Cells were excited by a 405 nm laser diode, and CFP and Venus were detected at 480–500 nm and 515–615 nm wavelength ranges, respectively. Each YFP/CFP ratio was calculated by dividing pixel-by-pixel a Venus image with a CFP image using Olympus FLUOVIEW software or ImageJ software.
Additional standard experimental procedures are presented in the supporting information S1 Text.
|
10.1371/journal.pcbi.1005604 | Adjudicating between face-coding models with individual-face fMRI responses | The perceptual representation of individual faces is often explained with reference to a norm-based face space. In such spaces, individuals are encoded as vectors where identity is primarily conveyed by direction and distinctiveness by eccentricity. Here we measured human fMRI responses and psychophysical similarity judgments of individual face exemplars, which were generated as realistic 3D animations using a computer-graphics model. We developed and evaluated multiple neurobiologically plausible computational models, each of which predicts a representational distance matrix and a regional-mean activation profile for 24 face stimuli. In the fusiform face area, a face-space coding model with sigmoidal ramp tuning provided a better account of the data than one based on exemplar tuning. However, an image-processing model with weighted banks of Gabor filters performed similarly. Accounting for the data required the inclusion of a measurement-level population averaging mechanism that approximates how fMRI voxels locally average distinct neuronal tunings. Our study demonstrates the importance of comparing multiple models and of modeling the measurement process in computational neuroimaging.
| Humans recognize conspecifics by their faces. Understanding how faces are recognized is an open computational problem with relevance to theories of perception, social cognition, and the engineering of computer vision systems. Here we measured brain activity with functional MRI while human participants viewed individual faces. We developed multiple computational models inspired by known response preferences of single neurons in the primate visual cortex. We then compared these neuronal models to patterns of brain activity corresponding to individual faces. The data were consistent with a model where neurons respond to directions in a high-dimensional space of faces. It also proved essential to model how functional MRI voxels locally average the responses of tens of thousands of neurons. The study highlights the challenges in adjudicating between alternative computational theories of visual information processing.
| Humans are expert at recognizing individual faces, but the mechanisms that support this ability are poorly understood. Multiple areas in human occipital and temporal cortex exhibit representations that distinguish individual faces, as indicated by successful decoding of face identity from functional magnetic resonance imaging (fMRI) response patterns [1–10]. Decoding can reveal the presence of face-identity information as well as invariances. However, the nature of these representations remains obscure because individual faces differ along many stimulus dimensions, each of which could plausibly support decoding. To understand the representational space, we need to formulate models of how individual faces might be encoded and test these models with responses to sufficiently large sets of face exemplars. Here we use representational similarity analysis (RSA) [11] to test face-coding models at the level of the representational distance matrices they predict. Comparing models to data in the common currency of the distance matrix enables us to pool the evidence over many voxels within a region, obviating the need to fit models separately to noisy individual fMRI voxels.
Many cognitive and neuroscientific models of face processing do not make quantitative predictions about the representation of particular faces [12,13]. However, such predictions can be obtained from models based on the notion that faces are encoded as vectors in a space [14]. Most face-space implementations apply principal components analysis (PCA) to face images or laser scans in order to obtain a space, where each component is a dimension and the average face for the training sample is located at the origin [15,16]. In such PCA face spaces, eccentricity is associated with judgments of distinctiveness, while vector direction is associated with perceived identity [17–19]. Initial evidence from macaque single-unit recordings and human fMRI suggests that brain responses to faces are strongly modulated by face-space eccentricity, with most studies finding increasing responses with distinctiveness [20–23]. However, there has been no attempt to develop a unified account for how a single underlying face-space representation can support both sensitivity to face-space direction at the level of multivariate response patterns and sensitivity to eccentricity at the level of regional-mean fMRI activations. Here we develop and evaluate several face-space coding models, which differ with respect to the proposed shape of the neuronal tuning functions across face space and with respect to the distribution of preferred face-space locations over the simulated neuronal population.
Face-space coding models define high-level representational spaces, which are assumed to arise through unspecified low-level featural processing. An alternative possibility is that some cortical face-space representations can be explained directly by low-level visual features, which typically covary with position in PCA-derived face spaces. To explore this possibility, we evaluated a Gabor-filter model, which receives stimulus images rather than face-space coordinates as input, and has previously been used to model response preferences of individual voxels in early visual cortex [24].
We found that cortical face responses measured with fMRI strongly reflect face-space eccentricity: a step along the radial axis in face space results in a much larger pattern change than an equal step along the tangential axis. These effects were consistent with either a sigmoidal-ramp-tuning or a Gabor-filter model. The performance of these winning models depended on the inclusion of a measurement-level population-averaging mechanism, which accounted for local averaging of neuronal tunings in fMRI voxel measurements.
In order to elicit strong percepts of the 3D shape of each individual face, we generated a set of photorealistic animations of face exemplars. Each 2s animation in the main experiment featured a face exemplar in left or right half profile, which rotated outward continuously (S2 Movie, Materials and methods). The animations were based on a PCA model of 3D face shape and texture [25]. Each frame of each animation was cropped with a feathered aperture and processed to equate low-level image properties across the stimulus set (Materials and methods). We generated 12 faces from a slice through the PCA face space (Fig 1b). Euclidean distances between the Cartesian coordinates for each face were summarized in a distance matrix (Fig 1a), which served as the reference for comparisons against distances in the perceptual and cortical face spaces (Fig 1c–1h). We generated a physically distinct stimulus set with the same underlying similarity structure for each participant by randomizing the orientation of the slice through the high-dimensional PCA space (for examples, see S1 Fig). This served to improve generalizability by ensuring that group-level effects were not strongly influenced by idiosyncrasies of face exemplars drawn from a particular PCA space slice. In formal terms, group-level inference therefore treats stimulus as a random effect [26].
In order to sample human cortical and perceptual face representations, human participants (N = 10) participated in a perceptual judgment task followed by fMRI scans. The perceptual judgment task (S1 Movie, 2145 trials over 4 recording days) involved force-choice judgments of the relative similarity between pairs of faces, which were used to estimate a behavioral distance matrix (Materials and methods). The same faces were then presented individually in a subsequent rapid event-related fMRI experiment (S2 Movie, 2496 trials over 4 recording days). Brain responses were analyzed separately in multiple independently localized visual regions of interest. We compared representational distance matrices estimated from these data sources to distances predicted according to different models using the Pearson correlation coefficient. These distance-matrix similarities were estimated in single participants and the resulting coefficients were Z-transformed and entered into a summary-statistic group analysis for random-effects inference generalizing across participants and stimuli (Materials and methods).
We observed a strong group-average correlation between distance matrices estimated from perceptual dissimilarity judgments and Euclidean distances in the reference PCA space (mean(r) = 0.83, mean(Z(r)) = 1.20, standard error = 0.05, p<0.001, Figs 1c and 5a, S1 Table). Correlations between the reference PCA space and cortical face spaces were generally statistically significant, but smaller in magnitude (Fig 5b and 5c, S5 Fig, S1 Table). Distances estimated from the fusiform face area were weakly, but highly significantly correlated with the reference PCA space (mean(r) = 0.17, mean(Z(r)) = 0.17, standard error = 0.04, p<0.001, Fig 5c). Distances estimated from the early visual cortex were even less, though still significantly, correlated with the reference PCA space (mean(r) = 0.07, mean(Z(r)) = 0.07, standard error = 0.04, p = 0.044, Fig 5b). These smaller correlations in cortical compared to perceptual face spaces could not be attributed solely to lower functional contrast-to-noise ratios in fMRI data, because the effects generally did not approach the noise-ceiling estimate for the sample (shaded region in Fig 5). The noise ceiling was based on the reproducibility of distance matrices between participants (Materials and methods). Instead, these findings indicate that the reference PCA space could not capture all the explainable dissimilarity variance in cortical face spaces.
We quantified the apparent warps in the cortical face spaces by constructing a multiple-regression RSA model, with separate distance-matrix predictors for eccentricity and direction, and for within and across face viewpoints (Fig 2a, Materials and methods). These predictors were scaled such that differences between the eccentricity and direction parameters could be interpreted as warping relative to a veridical encoding of distances in the reference PCA face space. We also observed strong viewpoint effects in multiple regions, including the early visual cortex (Fig 1e and 1f). Such effects were modeled by separate constant terms for distances within and across viewpoint.
Eccentricity changes had a consistently greater effect on representational patterns than direction changes, suggesting that a step change along the radial axis resulted in a larger pattern change than an equivalent step along the tangential axis (Fig 2c and 2d). The overrepresentation of eccentricity relative to direction was observed in each participant, and in both cortical and perceptual face spaces, although the effect was considerably larger in cortical face spaces. A repeated-measures two-factor ANOVA (eccentricity versus direction, within versus across viewpoint) on the single-participant parameter estimates from the multiple-regression RSA model was consistent with these apparent differences, with a statistically significant main effect of eccentricity versus direction for perceptual and cortical face spaces (all p<0.012, S2 Table). Thus, compared to the encoding in the reference PCA space, cortical face spaces over-represented the radial, distinctiveness-related axis compared to the tangential, identity-related axis.
Although these findings suggest a larger contribution of face-space eccentricity than direction in visual cortex, we also observed clear evidence for greater-than-chance discrimination performance among faces that differed only in face-space direction. Group-average cross-validated discriminant distances for faces that differed in direction but not eccentricity exceeded chance-level performance (p<0.05) for typical and caricatured faces in both the early visual cortex and the fusiform face area (S3 Fig). Sub-caricatured faces were less consistently discriminable. Indeed, direction discrimination increased with eccentricity both within (mean = 0.013, standard error = 0.007, p = 0.036) and across (mean = 0.012, standard error = 0.005, p = 0.016) viewpoint in the fusiform face area (linear effect of sub>typical>caricature), suggesting that direction discrimination increased with face-space eccentricity in a dose-dependent manner (S3 Table). By contrast, within viewpoint discrimination performance in the early visual cortex scaled with eccentricity (mean = 0.039, standard error = 0.008, p<0.001), but distances that spanned a viewpoint change did not vary with eccentricity (mean = 0.009, standard error = 0.019, p = 0.297). This is consistent with a view-dependent representation in early visual cortex. Thus, cortical regions discriminate identity-related direction information even in the absence of a difference in distinctiveness-related eccentricity information, suggesting that cortical face representations cannot be reduced to a one-dimensional code based on distinctiveness alone. In summary, cortical coding of face-space position is systematically warped relative to the reference PCA space, with a substantial overrepresentation of eccentricity and a smaller, but reliable contribution of face-space direction.
Cortical face-space warping could not be explained by regional-mean activation preferences for caricatures. We performed a regional-mean analysis of responses in each cortical area, which confirmed previous reports that fMRI responses increase with distinctiveness across much of visual cortex (Fig 6, S6 Fig) [23]. In order to test the influence of such regional-mean activation effects on representational distances, we adapted our discriminant distance metric to remove additive and multiplicative overall activation effects (Materials and methods). Distance matrices estimated using this alternative method were highly similar to ones estimated without removal of overall activation effects (all r = 0.9 or greater for the Pearson correlation between group-average distance matrices with and without mean removal, S4 Fig). Thus, although eccentricity affected the overall activation in all visual areas, the warping of the cortical face spaces could not be attributed to overall activation effects alone.
We developed multiple computational models, each of which predicts a representational distance matrix and a regional-mean activation profile. These models can be divided into three classes: the sigmoidal-ramp tuning and exemplar tuning models receive face-space coordinates as input, while the Gabor-filter model receives gray-scale pixel intensities from the stimulus images as input. We evaluated each of these three model classes with and without a measurement-level population-averaging mechanism, which approximates how fMRI voxels locally average underlying neural activity.
The sigmoidal-ramp-tuning model proposes that the representational space is covered with randomly oriented ramps, each of which exhibits a monotonically increasing response along its preferred direction in face space (Fig 3a). This model is inspired by known preferences for extreme feature values in single units recorded from area V4 and from face-selective patches in the macaque visual cortex [20,27,28]. We modeled the response along each model neuron’s preferred direction using a sigmoidal function with two free parameters, which control the horizontal offset and the saturation of the response function (Materials and methods). A third parameter controlled the strength of measurement-level population averaging by translating each individual model unit’s response toward the population-mean response. The way this accounts for local averaging by voxels is illustrated for the fusiform face area in Fig 3a and 3b. It can be seen that measurement-level population averaging introduces a substantial U-shape in the individual response functions, with only a minor deflection in favor of a preferred face-space direction. At the level of Euclidean distances between population response vectors evoked by each face, this leads to exaggerated distances for radial relative to tangential face differences (Fig 3c). In summary, measurement-level population averaging provides a simple means to interpolate between two extreme cases: A value of 0 corresponds to the case where the model’s response is perfectly preserved in the fMRI voxels, whereas a value of 1 corresponds to the case where the model’s response to a given stimulus is reduced to the arithmetic mean over the model units.
In the exemplar model, each unit prefers a location in face space, rather than a direction, and its tuning is described by a Gaussian centered on the preferred location. The representational space is covered by a population of units whose preferred locations are sampled from a Gaussian centered on the norm face (Fig 3d, Materials and methods). We fitted the Gaussian exemplar-tuning model similarly to the sigmoidal ramp-tuning model, using two parameters that controlled the width of the Gaussian tuning function and the width of the Gaussian distribution from which preferred faces were sampled. We also evaluated a variant of the exemplar model where the distribution of preferred faces followed an inverted-Gaussian distribution (Fig 3e). Population averaging was modeled in the same way as for the sigmoidal-ramp-tuning model using a third parameter.
The Gabor-filter model differs from the previous model classes in that it receives gray-scale image intensities as input, rather than PCA space position (Fig 4a). Such models have previously been used to account for response preferences of individual voxels in early visual cortex [24]. The model comprises Gabor filters varying in orientation, spatial frequency and phase, and spatial position. The filters are organized into banks, each corresponding to a spatial frequency and comprising a different number of spatial positions (coarser for lower spatial frequencies; Fig 4b, Materials and methods). We assumed that all orientations and spatial positions are equally represented. For the spatial frequencies, however, we let the data determine the weighting. We fitted a weighted representational model with one weight for each spatial-frequency bank (5 free parameters, Fig 4c). Local averaging in fMRI voxels was modeled using two stages of measurement-level population averaging: First, filters with tuning centers on either side of the vertical meridian were translated separately toward their respective hemifield-specific population averages. Second, a global-pool averaging was performed similarly to the other models. The contribution of these two population-average signals to the measured responses was modeled by 2 additional parameters (Fig 4d). The additional hemifield-specific averaging stage was necessary to account for strong view-specific effects in early visual cortex, but did not materially contribute to the fit in ventral temporal regions.
We fitted each of the computational models so as to best predict the representational distance matrices from cortical regions and perceptual judgments using a grid search over all free parameters. We used a leave-one-participant-out cross-validation approach, in which model performance was evaluated on participants and face identities not used in fitting the parameters (Materials and methods). Model performance was summarized as the Fisher-Z-transformed Pearson correlation coefficient between the model distances and the data distances. We performed statistical inference on the average Fisher-Z-transformed correlations over all train-test splits of the data using T tests. Our cross-validation scheme tests for generalization across participants and face identities and ensures that models that differ in complexity (number of free parameters) can be compared. In order to investigate the effect of local averaging in fMRI voxels on representational similarity, we fitted two variants of each model: the full model and a variant that excluded measurement-level population averaging.
We found that all evaluated models explained almost all the explainable variance for the perceptual face space (Fig 5a), with only negligible differences in cross-validated generalization performance (for all pairwise model comparisons, see S4 Table). The inclusion of measurement-level population averaging had little effect on performance. This is expected because perceptual judgments, unlike fMRI voxels, are not affected by local averaging across representational units. Thus, the behavioral data was ambiguous with regard to the proposed models, which motivates model selection by comparison to the functional imaging data.
Unlike the perceptual judgments data, the cortical face spaces exhibited substantial differences between the model fits, with a robust advantage for measurement-level population averaging in most cases. In the following, we focus on generalization performance for fits to the early visual cortex and the fusiform face area (for fits to other regions, see S5 Fig).
The early visual cortex was best explained by the Gabor-filter model, which beat the alternative computational models (p<0.001 for all pairwise model comparisons, S4 Table) and came close to explaining all explainable variance given noise levels in the data. Generalization performance for this model was slightly, but significantly better (p = 0.009, Fig 5b) when population averaging was enabled. As a control, we also tested raw pixel intensities as the representational units. We found no significant difference in performance between the 0-parameter pixel-intensity model and the fitted Gabor-filter model with population averaging (p = 0.380).
The fusiform face area was also well explained by the Gabor-filter model, but in this region we observed similar performance for the sigmoidal-ramp-tuning model. Both models, with population averaging, reached the lower bound of the noise ceiling (Materials and methods; Fig 5c, S4 Table), suggesting that these models were able to explain the variance in the dataset that was consistent between participants. We also observed comparable generalization performance for the Gaussian exemplar model. Measurement-level population averaging improved generalization performance in the fusiform face area for both the Gabor-filter (p<0.001) and sigmoidal-ramp-tuning models (p = 0.032), but did not improve either of the exemplar models (p = 0.263 for Gaussian exemplar, p = 0.096 for negative Gaussian exemplar). In summary, the representation in the fusiform face area could be explained by multiple models, and in most cases measurement-level population averaging improved the quality of the fit.
Multiple computational models provided qualitatively similar fits to our cortical data at the distance-matrix level. However, we might still be able to adjudicate between them at the level of regional-mean activation profiles. To this end, we obtained activation-profile predictions from each model by averaging over all model units. We then correlated the predicted population-mean activation profile with the regional-mean fMRI activation profile for each participant and performed an activation profile similarity analysis analogously to the distance matrix similarity analysis above. We found that the activation profiles from the computational models were predictive of cortical activation profiles, even though these models were fitted to distance matrices rather than to regional-mean fMRI responses (Fig 6, for other regions see S6 Fig). In particular, the sigmoidal-ramp and Gabor-filter models both predicted increasing population-mean responses with face-space eccentricity, while the Gaussian-exemplar model predicted decreasing responses with eccentricity (S5 and S6 Tables for pairwise comparisons). This constitutes evidence against the Gaussian-exemplar model, under the assumption that neuronal activity is positively associated with regional-mean fMRI response in visual regions [29–33]. The preference for faces closer to the PCA-space origin (sub-caricatures) in the Gaussian-exemplar model arises as a necessary consequence of the Gaussian distribution of preferred faces, which is centered on the average face. We also tested a Gaussian exemplar-tuning model with an inverted Gaussian distribution of preferred faces. In this model, more units prefer faces far from the norm (caricatures) than faces close to the norm (sub-caricatures). However, the inverted-Gaussian exemplar model’s generalization performance was considerably worse than the standard-Gaussian exemplar model’s (Fig 5).
In summary, exemplar models accurately predicted representational distances for cortical face spaces when the preferred-face distribution was Gaussian, but such distributions led to inaccurate predictions of regional-mean fMRI activation profiles. Thus, analysis of regional-mean fMRI responses enabled us to adjudicate between models that made similar predictions at the distance-matrix level, and specifically indicated that the Gaussian exemplar model is unlikely to be the correct model for cortical face-space representation, despite a good fit at the distance-matrix level.
We found that models that included measurement-level population averaging generally outperformed models that did not. This advantage appeared to originate in how models with population averaging captured two effects in the cortical face spaces: symmetric view-tolerance and over-representation of face-space eccentricity relative to direction.
First, population averaging enabled the image-based Gabor-filter model to exhibit symmetric-view-tolerant responses to the face exemplars. Two mirror-symmetric views will drive a mirror symmetric set of Gabor features. Thus, while the pattern of activity differs, the population-mean activity is similar. Measurement-level pooling over filters centered on distinct visual field locations therefore renders symmetric views more similar in the representation. For instance, the face space in the fusiform face area exhibited little sensitivity to viewpoint, and the Gabor-filter model fit to this region was greatly improved by the inclusion of measurement-level population averaging (Fig 5c, S2 Fig). Indeed, with population averaging, generalization performance for the image-based Gabor-filter model was similar to the sigmoidal-ramp-tuning and exemplar models, for which view tolerance is assumed at the input stage. General view tolerance, beyond the symmetric views we used here, is computationally more challenging. However, for our stimulus set, it was not necessary to posit any intrinsic view-invariant computations in the fusiform face area to explain how its face spaces come to exhibit symmetric view tolerance.
Second, measurement-level population averaging increased the degree to which both the sigmoidal-ramp and the Gabor-filter model over-represented face-space eccentricity relative to direction, which improved the fit for multiple cortical regions. To isolate this smaller effect from the larger symmetric-view-tolerance effect, we collapsed viewpoint in the first-level single-participant fMRI linear model and re-estimated the cortical face spaces and all model fits for the resulting simplified 12-condition design matrix, where each predictor coded appearances of a given face identity regardless of its viewpoint (S7 Fig). Even after collapsing across viewpoints at the first level in this way, measurement-level population averaging still improved the generalization performance of the sigmoidal-ramp and the Gabor-filter model in nearly all cases (p<0.05, see S7 Table for descriptive statistics and S8 Table for all pairwise comparisons), including the early visual cortex (p<0.001 for sigmoidal ramp tuning, p = 0.004 for Gabor filter) and the fusiform face area (p = 0.001 for sigmoidal ramp tuning, p = 0.007 for Gabor filter). Thus, the advantage for measurement-level population averaging could not be accounted for by the fact that it helps explain symmetric view tolerance. In sum, the addition of population averaging to the model improved model generalization performance, and this advantage appeared to originate in accounts for two distinct observed phenomena.
This study investigated human face processing by measuring how a face space of individual exemplars was encoded in visual cortical responses measured with fMRI and in perceptual judgments. Relative to a reference PCA model of the 3D shape and texture of faces, cortical face spaces from all targeted regions systematically over-represented eccentricity relative to direction (i.e., the radial relative to the tangential axis). Cortical regions varied in their sensitivity to face viewpoint. We fitted multiple computational models to the data. Considered collectively, the cortical face spaces in the fusiform face area were most consistent with a PCA-space-based sigmoidal-ramp-tuning model and an image-based Gabor-filter model, and less consistent with models based on exemplar coding. As expected, effects in the early visual cortex were consistent primarily with the Gabor-filter model. In all cases, the winning models’ performance depended on the inclusion of a measurement-level population-averaging mechanism, which approximates how individual model units are locally averaged in functional imaging measurements.
Out of the models we considered, the best accounts for the fusiform face area were a PCA-space-based sigmoidal-ramp-tuning model and an image-based Gabor-filter model. Exemplar-coding models exhibited relatively lower generalization performance, or made inaccurate predictions for regional-mean fMRI activation profiles. Importantly, the advantage for both the sigmoidal-ramp and Gabor-filter model depended on the measurement-level population averaging mechanism. The key contribution of our modeling effort is to narrow the set of plausible representational models for the fusiform face area to two models that can explain both representational distances and the regional-mean activation profile.
It may appear surprising that a PCA-space coding model based on sigmoidal-ramp tuning and an image-based model based on Gabor filters should perform so similarly when fitted to face spaces in the fusiform face area. However, the sigmoidal-ramp-tuning model captures continuous variation in face shape and texture, which covaries with low-level image similarity. For instance, local curvature likely increases with face space eccentricity and is encoded in a ramp-like manner at intermediate stages of visual processing in macaque V4 [27]. Conversely, the Gabor-filter model likely possesses sensitivity to face-space direction because the contrast of local orientation content varies with major face features such as eyebrow or lip thickness. Similarly, the Gabor-filter model’s ability to account for regional-mean activation profiles likely arises because local contrast increases with face-space eccentricity, which results in an overall greater activation over the filter banks. There are multiple ways to parameterize face space, not all of which require domain-specific face features. This might also clarify why scene-selective areas such as the parahippocampal place area exhibited somewhat similar representational spaces as face-selective regions in the current study. Such widely-distributed face-exemplar effects are consistent with previous decoding studies [2,4]. The Gabor-filter model provides one simple account for how such widely distributed face-exemplar effects can arise. It is likely that the face-space effects we report are driven at least in part by a general mechanism for object individuation in visual cortex rather than the engagement of specialized processing for face recognition.
The models we evaluate here raise the provocative possibility that in some cases, fMRI effects that might conventionally be attributed to high-level featural coding could instead arise from the neuroimaging measurement process. Such an explanation appears possible for two effects in our data. First, even though the Gabor-filter model is a single-layer network with limited representational flexibility, this model nevertheless exhibited near-complete tolerance to mirror-symmetric viewpoint changes, when coupled with measurement-level population averaging. Second, both this model and the sigmoidal-ramp-tuning model showed greater over-representation of eccentricity when measurement-level population averaging was enabled, suggesting that this over-representation in the fMRI data might also plausibly arise through local averaging in voxels.
Previous studies have tended to interpret view-tolerant fMRI effects in terms of cortical processing to support invariant object recognition [2,34,35]. The Gabor-filter model suggests a mechanism by which functional imaging measures can exaggerate apparent view-tolerance through spatial pooling over neuronal responses. This result does not contradict previous reports of view-tolerant coding for faces in neuronal population codes measured with single-unit recording [36–39], but rather demonstrates that the type of tolerance to symmetric viewpoint changes that we observed in the current study can be explained without resorting to such intrinsic view-tolerant mechanisms (see also Ramirez et al. [40]). Such findings may go some way toward reconciling apparent discrepancies between single-unit and functional imaging data. For instance, tolerance to symmetrical viewpoints is widespread in human visual cortex when measured with fMRI [34], but appears specific to a subset of regions in the macaque face-patch system when measured with single-unit recordings [36]. These results are only contradictory if the measurement process is not considered. In summary, we demonstrate that measurement effects can produce apparent view tolerance in fMRI data. This finding does not suggest that fMRI cannot detect view-tolerant coding (see also [41]). For example, population averaging may not account for all cases of non-symmetric view tolerance. However, our results do suggest that modeling of the measurement process is important to correctly infer the presence of such mechanisms from neuroimaging measurements.
The winning models in this study exemplify how sensitivity to face-space eccentricity at the regional-mean activation level can arise as an artifact of averaging, with no individual neuron encoding distinctiveness or an associated psychological construct. Previous functional imaging studies often interpreted response modulations with face eccentricity as evidence for coding of distinctiveness or related social perception attributes [21–23,42]. However, both the PCA-space-based sigmoidal-ramp-tuning model and the image-based Gabor-filter model exhibited increasing population-average responses with eccentricity, even though neither model encodes eccentricity at the level of its units. Although one could, of course, construct a competing model that explicitly codes eccentricity, the models used here are more consistent with single-unit recording studies, where cells generally are tuned to particular features, with a preference for extreme values, rather than responding to eccentricity regardless of direction [20,27,28]. Here we demonstrate that when the local averaging of such biologically plausible neuronal tunings is modeled, eccentricity sensitivity emerges without specialized encoding of this particular variable. Related effects have been reported in attention research, where response-gain and contrast-modulation effects at the single-neuron level may sum to similar additive-offset effects at the fMRI-response level [43]. In summary, direct interpretation of regional-mean fMRI activations in terms of neuronal tuning can be misleading when the underlying neuronal populations are heterogeneous.
A simple model of measurement-level population averaging was sufficient here to substantially improve the generalization performance of multiple computational models for multiple cortical regions. The precise way that fMRI voxels sample neuronal activity patterns remains a topic of debate [30,31,33,44]. However, under the simple assumption that voxels sample random subsets of neurons by non-negatively weighted averaging, the effect on the measured fMRI distance matrix will be a uniform stretching along the all-one vector (representing the neuronal population average). To appreciate this point, consider the case of measurement channels that are unlike fMRI voxels in that they sample with random positive and negative weights. Under such conditions, we expect neuronal distances to be approximately preserved in the measurement channels according to the Johnson-Lindenstrauss lemma (for further discussion of this point, see [45]). Intuitively, the measurement channels re-describe the space with randomly oriented axes (without any directional bias). However, fMRI voxels are better understood as taking local non-negative weighted averages of neuronal activity, since the association between neural responses and fMRI response is generally thought to be positive. In such representational spaces the axes have orientations that are biased to fall along the all-1 vector. In practice, this measurement model assumes that the fMRI distance matrix for a given region of interest will over-represent distances to the extent that those distances modulate the neuronal population-average response.
Here we approximated measurement effects for models with nonlinear parameters by mixing the population average into the predicted representational feature space. Despite the simplicity of this method, our noise-ceiling estimates indicate that the winning models captured nearly all the explainable variance in the current dataset. For model representations without nonlinear parameters, this measurement model can be implemented more easily by linearly combining the model’s original distance matrix and the distance matrix obtained for the population average dimension of the space (using squared Euclidean distances estimates, see also 46–48). Thus, the measurement-level population averaging mechanism we propose here is widely and easily applicable to any case where a computational model is compared to neuroimaging data at the distance-matrix level.
The distance-matrix effects of local pooling of neuronal responses in fMRI voxels is correctly accounted for by our measurement model under the assumption that neurons are randomly intermixed in cortex (i.e. voxels sample random subsets of neurons). This simplifying assumption is problematic for early visual areas, where there is a well-established retinotopic organization with a strong contralateral response preference. For the retinotopic Gabor filter model, we therefore added a hemifield-specific pooling stage, which helped account for strong view-sensitivity in occipital regions of interest. Accounting for measurement effects in the presence of topographic organization is likely to prove more challenging for naturalistic stimulus sets. One solution is to account for local averaging in fMRI voxels by local averaging of the model’s internal representational map [45]. This local-pooling approach can be thought of as providing a further constraint on the comparison between model and data, because smoothing the model representation is only expected to improve the fit if the model response topography resembles the cortical topography. This may provide a means of adjudicating between topographically organized models, even when the models predict similar distance matrices in the absence of measurement-effect modeling. In summary, the global population average is a special dimension of the representational space, which is overrepresented in voxels that pool random subsets of neurons. This effect accounts for much variance in the representational distances in the current study, and is easy to model. Modeling the overrepresentation of the global average, as we did here, is suitable for models that do not predict a spatial organization (e.g. PCA-space coding models). For models that do predict a spatial organization, it may be more appropriate to simulate fMRI voxels by local averaging of the model’s representational map [45].
This study demonstrates the importance of considering multiple alternative models to guide progress in computational neuroimaging. In particular, the finding that practically every model we evaluated exhibited significantly greater-than-zero generalization performance strongly suggests how studies that only evaluate a limited set of candidate models can arrive at misleading conclusions (see e.g. [49]).
Representational similarity analysis has two key advantages for model comparison relative to alternative approaches. First, competing model predictions can be easily visualized at the level of the best-fitting distance matrix for a given cortical region (e.g., S2 Fig). By contrast, models that are fitted to individual voxels are harder to visualize because the number of voxels per region typically exceeds what can be practically plotted. Furthermore, individual time-points in a rapid event-related fMRI experiment cannot easily be labeled according to experimental stimuli or conditions, which complicates interpretation of fitted time-courses. Second, RSA makes it possible to compare model fits and estimated parameters across data modalities, for instance, between fMRI responses and psychophysical similarity judgments. Such comparisons are challenging when the data is modeled at a lower level, because modality-specific parameters must be added to each model (e.g., parameters controlling the hemodynamic response function for fMRI, decision-threshold parameters for behavioral judgments). The presence of these non-shared parameters makes it difficult to attribute any apparent modality differences to the data rather than to the model specification. In summary, RSA is a particularly attractive analysis approach for studies that emphasize model comparison.
Although the central goal of model comparison is to select the best account of the data, the finding that some models are not dissociable under the current experimental context also has important implications for the design of future studies. Here we demonstrated that Gaussian-distributed exemplar-coding models are less likely to account for human face coding, while accounts based on sigmoidal ramp tuning and Gabor filter outputs perform very similarly. This suggests the need to design stimulus sets that generate distinct predictions from these winning models. For example, presenting face stimuli on naturalistic textured backgrounds may be sufficient to adjudicate between the two models, because the Gabor-filter model lacks a mechanism for figure-ground separation. In conclusion, our study exemplifies the need to test and compare multiple models and suggests routes by which the sigmoidal-ramp-tuning model of face-space coding could be further evaluated.
All procedures were performed under a protocol approved by the Cambridge Psychology Research Ethics Committee (CPREC). Human participants provided written informed consent at the beginning of each data recording day.
10 healthy human participants participated in a similarity judgment task and fMRI scans. The psychophysical task comprised 4 separate days of data collection which were completed prior to 4 separate days of fMRI scans. Participants were recruited from the local area (Cambridge, UK) and were naïve with regard to the purposes of the study. Five additional participants participated in data collection up to the first MRI data recording day, but were not invited to complete the study due to difficulties with vigilance, fixation stability, claustrophobia and/or head movements inside the scanner. The analyses reported here include all complete datasets that were collected for the study.
We generated faces using a norm-based model of 3D face shape and texture, which has been described in detail previously [15,25]. Briefly, the model comprises two PCA solutions (each trained on 200 faces), one based on 3D shape estimated from laser scans and another based on texture estimated from digital photographs. The components of each PCA solution are considered dimensions in a space that describes natural variation in facial appearance. All stimulus generation was performed using the PCA solution offered by previous investigators, and no further fitting was performed for this study. We yoked the shape and texture solutions in all subsequent analyses since we did not have distinct hypotheses for these.
We developed a method for sampling faces from the reference PCA space in a manner that would maximize dissimilarity variance. This is related to the concept of design efficiency in univariate general linear modeling [50], and involves maximizing the variance of hypothesized distances over the stimulus set. Because randomly sampled distances in high dimensional spaces tend to fall in a narrow range of distances relative to the norm [51], we reduced each participant’s effective PCA space to 2D by specifying a plane which was centered on the norm of the space and extended at a random orientation. The face exemplars constituted a polar grid on this plane, with 4 directions at 60 degrees separation and 3 eccentricity levels (scaled at 30%, 100% and 170% of the mean eccentricity in the training face set). The resulting half-circle grid on a plane through the high-dimensional space is adequate for addressing our hypotheses concerning the relative role of direction and eccentricity coding under the assumption that the high-dimensional space is isotropic. The use of a half-circle also serves to address a potential concern that apparent eccentricity sensitivity might arise as a consequence of adaptation to the experimental stimuli [52]. Such adaptation effects are only collinear with eccentricity (ie, prototype) coding if the prototype is located at the average position over the experienced stimulus images. For our stimulus space, this average position would fall approximately between sub- and typical faces and between the second and third direction in the PCA space. Contrary to an adaptation account, there was no suggestion in our data that cortical distances were exaggerated as a function of distance along this axis. The orientation of the PCA-space slice was randomized between participants and model fits were based on cross-validation over participants. Under these conditions, any non-isotropicity is only expected to impair generalization performance. In preliminary tests we observed that this method yielded substantially greater dissimilarity variance estimates than methods based on Gaussian or uniform sampling of the space.
We used Matlab software to generate a 3D face mesh for each exemplar. This mesh was rendered at each of the orientations of interest in the study in a manner that centered the axis of rotation on the bridge of the nose for each face. This procedure ensured that the eye region remained centered on the fixation point throughout each animation in order to discourage eye movements. Renders were performed at sufficient increments to enable 24 frames per second temporal resolution in the resulting animations. Frames were converted to gray-scale and cropped with a feathered oval aperture to standardize the outline of each face and to remove high-contrast mesh edges from the stimulus set. Finally, we performed a frame-by-frame histogram equalization procedure where the average histogram for each frame was imposed on each individual face. Thus, the histogram was allowed to vary across time but not across faces. Note that histogram matching implies that the animations also have identical mean gray-scale intensity and root-mean-square contrast.
A potential concern with these matching procedures is that they could affect the validity of the comparison to the reference PCA space. However, we found that the opposite appeared to be true: distances in the reference PCA space were more predictive of pixelwise correlation distances in the matched images than in the original images. Thus, the matching procedure did not remove features that were encoded in the PCA space and may in fact have acted to emphasize such features.
We used a pair-of-pairs task to characterize perceptual similarity (S1 Movie). Participants were presented with two vertically offset pairs of faces on a standard LCD monitor under free viewing conditions, and judged which pair was relatively more dissimilar with a button press on a USB keyboard (two-alternative force choice). Each face rotated continuously between a leftward and a rightward orientation (45 degrees left to 45 degrees right of a frontal view over 3 seconds). Ratings across all possible pairings of face pairs (2145 trials: all pairings of the 66 possible pairs of the 12 faces) were combined into a distance matrix for each participant, where each entry reflects the percentage of trials on which that face pair was rated as relatively more dissimilar. The behavioral data was collected over 16 runs (135 trials for the first 15 runs, 120 in the final run). Each participant completed 4 runs in each of 4 data recording days.
We measured brain response patterns evoked by faces in a rapid event-related fMRI experiment (S2 Movie). Participants fixated on a central point of the screen where a pseudo-random sequence of face animations appeared (7 degrees visual angle in height, 2s on, 1s fixation interval). We verified fixation accuracy online and offline using an infrared eye tracking system (Sensomotoric Instruments, 50Hz monocular acquisition). The faces rotated outward in leftward and rightward directions on separate trials (18 to 45 degrees rotation left or right of a frontal view), and participants responded with a button press to occasional face repetitions regardless of rotation (one-back task). This served to encourage attention to facial identity rather than to incidental low-level physical features. Consistent with a task strategy based on identity recognition rather than image matching, participants were sensitive to exemplar repetitions within viewpoint (mean d’+-1 standard deviation 2.68+-0.62) and to exemplar repetitions where the viewpoint changed (2.39+-0.52).
The experiment was divided into 16 runs where each run comprised 156 trials bookended by 10s fixation intervals. Each scanner run comprised two experimental runs, which were modeled independently in all subsequent analyses. The data was collected on 4 separate MRI data recording days (2 scanner runs per recording day). The trial order in each run was first-order counterbalanced over the 12 faces using a De Bruijn sequence [53] with 1 additional repetition (diagonal entries in transfer matrix) added to each face in order to make the one-back repetition task more engaging and to increase design efficiency [50]. The rotation direction in which each face appeared was randomized separately, since a full 24-stimulus De Bruijn sequence would have been over-long (576 trials). Although the resulting 24-stimulus sequences were not fully counter-balanced, we used an iterative procedure to minimize any inhomogeneity by rejecting rotation direction randomizations that generated off-diagonal values other than 0 and 1 in the 24-condition transfer matrix (that is, each possible stimulus-to-stimulus transfer in the sequence could appear once or not at all). These homogeneous trial sequences served to enhance leave-one-run-out cross-validation performance by minimizing over-fitting to idiosyncratic trial sequence biases in particular runs. We modeled the data from each run with one predictor per face exemplar and viewpoint.
Functional and structural images were collected at the MRC Cognition and Brain Sciences Unit (Cambridge, UK) using a 3T Siemens Tim Trio system and a 32-channel head coil. There were 4 separate MRI data recording days for each participant. Each recording day comprised 2 runs of the main experiment followed by 2 runs of the functional localizer experiment. All functional runs used a 3D echoplanar imaging sequence (2mm isotropic voxels, 30 axial slices, 192 x 192mm field of view, 128 x 128 matrix, TR = 53ms, TE = 30ms, 15° flip angle, effective acquisition time 1.06s per volume) with GRAPPA acceleration (acceleration factor 2 x 2, 40 x 40 PE lines). Each participant’s functional dataset (7376 volumes over 8 scanner runs for the main experiment) was converted to NIFTI format and realigned to the mean of the first recording day’s first experimental run using standard functionality in SPM8 (fil.ion.ucl.ac.uk/spm/software/spm8/). A structural T1-weighted volume was collected in the first recording day using a multi-echo MPRAGE sequence (1mm isotropic voxels)[54]. The structural image was de-noised using previously described methods [55], and the realigned functional dataset’s header was co-registered with the header of the structural volume using SPM8 functionality. The structural image was then skull-stripped using the FSL brain extraction tool (fmrib.ox.ac.uk/fsl), and a re-sliced version of the resulting brain mask was applied to the fMRI dataset to remove artifacts from non-brain tissue. We constructed design matrices for each experimental run by convolving the onsets of experimental events with the SPM8 canonical hemodynamic response function. Slow temporal drifts in MR signal were removed by projecting out the contribution of a set of nuisance trend regressors (polynomials of degrees 0–4) from the design matrix and the fMRI data in each run.
We estimated the neural discriminability of each face pair for each region of interest using a cross-validated version of the Mahalanobis distance [56]. This analysis improves on the related Fisher’s linear discriminant classifier by providing a continuous metric of discriminability without ceiling effects. Similarly to the linear discriminant, classifier weights were estimated as the contrast between each condition pair multiplied by the inverse of the covariance matrix of the residual time courses, which was estimated using a sparse prior [57]. This discriminant was estimated separately for the concatenated design matrix and fMRI data in each possible leave-one-out split of the experimental runs, and the resulting weights were transformed to unit length and projected onto the contrast estimates from each training split’s corresponding test run (16 estimates per contrast). The 16 run-specific distance estimates were averaged to obtain the final neural discriminability estimate for that participant and region. When the same data is used to estimate the discriminant and evaluate its performance, this algorithm returns the Mahalanobis distance, provided that a full rather than sparse covariance estimator is used [56]. However, unlike a true distance measure, the cross-validated version that we use here is centered on 0 under the null hypothesis. This motivates summary-statistic random-effects inference for above-chance performance using conventional T tests.
We developed a variant of this discriminant analysis where effects that might be broadly described as region-mean-related are removed (S4 Fig). This control analysis involved two modifications to how contrasts were calculated at the level of forming the discriminant and at the level of evaluating the discriminant on independent data. First, each parameter estimate was set to a mean of zero in order to remove any additive offsets in response levels between the conditions. Second, for each pair of mean-subtracted parameter estimates, the linear contribution of the mean estimate over the pair was removed from each estimate before calculating the contrast. This corrects for the case where a single response pattern is multiplicatively scaled by the conditions. The resulting control analysis is insensitive to effects driven by additive or multiplicative scaling offsets between the conditions.
We used a multiple regression model to estimate the relative contribution of eccentricity and direction to cortical and perceptual face-space representations. Multiple regression fits to distance estimates can be performed after a square transform, since squared distances sum according to the Pythagorean theorem. We partitioned the squared distances in the reference PCA space into variance associated with eccentricity changes by creating a distance matrix where each entry reflected the minimum distance for its eccentricity group in the squared reference PCA matrix (that is, cases along the group’s diagonal where there was no direction change). The direction matrix was then constructed as the difference between the squared reference PCA matrix and the eccentricity matrix (Fig 2a). These predictors were vectorized and entered into a multiple regression model together with a constant term. This partitioning of the dissimilarity variance in the reference PCA matrix is complete in the sense that a multiple regression RSA model where the reference PCA matrix is used as the dependent variable yields parameter estimates of [1,1,0] for eccentricity, direction and constant, respectively, with no residual error. These three predictors were then split according to viewpoint, with separate sets of predictors for distances within and across viewpoint. The absolute values of the cortical and perceptual distance matrices were squared and then transformed back to their original sign before being regressed on the predictor matrix using ordinary least squares. Finally, the absolute values of the resulting parameter estimates were square-root transformed and returned to their original signs.
We used a conventional block-based functional localizer experiment to identify category-selective and visually-responsive regions of interest in human visual cortex. Participants fixated a central cross on the screen while blocks of full-color images were presented (36 images per block presented with 222ms on, 222ms off, 16 s fixation). Participants were instructed to respond to exact image repetitions within the block. Each run comprised 3 blocks each of faces, scenes, objects and phase-scrambled versions of the scene images. Each participant’s data (8 runs of 380 volumes, 2 runs collected for each of 4 MRI data recording days) was smoothed with a Gaussian kernel (6mm full width at half maximum) and responses to each condition were estimated using a standard SPM8 first-level model. Regions of interest were identified using a region-growing approach, where a peak coordinate for each region was identified in individual participants, and a region of interest was grown as a contiguous set of the most selective 100 voxels extending from this coordinate. We defined the face-selective occipital and fusiform face areas with the minimum-statistic conjunction contrast of faces over objects and faces over baseline, and the scene-selective parahippocampal place area and transverse occipital sulcus as the minimum-statistic conjunction contrast of scenes over objects and scenes over baseline, and the early visual cortex as the contrast of scrambled stimuli over the fixation baseline. We also attempted to localize a face-selective region in the posterior superior temporal sulcus, a face-selective region in anterior inferotemporal cortex and a scene-selective region in retrosplenial cortex, but do not report results for these regions here since they could only be identified in a minority of the participants. All regions of interest were combined into bilateral versions before further analysis since we did not have distinct predictions concerning functional lateralization. Example regions of interest can be viewed in S8 Fig. Typical MNI coordinates for each region are provided in S9 Table.
The sigmoidal ramp model comprises 1000 model units, each of which exhibits a monotonically increasing response in a random direction extending from the origin of the face space (Fig 3). The response y at position x along the preferred direction is described by the sigmoid
y[raw]=1/(1+exp((−x+o)/s));
where the free parameters are o, which specifies the horizontal offset of the response function (zero places the midpoint of the response function at the norm of the space, values greater than zero corresponds to responses shifted away from the norm), and s, which defines response function saturation (4 corresponds to a near-linear response in the domain of the face exemplars used here, while values near zero correspond to a step-like increase in response). The raw output of each model unit is then translated toward the population-mean response
y[ final ]=(y[ raw ]−y[ mean ])*(1−p)+y[ mean ]
where p is a free parameter that defines the strength of measurement-level population averaging (0 corresponds to no averaging, 1 corresponds to each model unit returning the population-mean response).
The exemplar model comprises 1000 model units, each of which prefers a Cartesian coordinate in the face space with response fall-off captured by an isotropic Gaussian. The free parameters are w, which controls the full width at half-maximum tuning width of the Gaussian response function, and d, which controls the width of the Gaussian distribution of tuning centers (0.1 places Z = 2.32 at 10% of the eccentricity of the caricatures while 3 places this tail at 300% of the eccentricity of the caricatures).
We also constructed an inverted-Gaussian variant of this model where the distribution of distances was inverted at Z = 2.32 and negative distances truncated to zero (1% of exemplars). This model was fitted with similar parameters as the original Gaussian exemplar model.
The Gabor filter model is an adaptation of a neuroscientifically-inspired model that has previously been used to successfully predict single-voxel responses in the early visual cortex ([24], github.com/kendrickkay/knkutils/tree/master/imageprocessing). The model is composed of 5 banks of Gabor filters varying in spatial position, phase (2 values) and orientation (8 directions, Fig 4). We measured each filter’s response to the last frame of each animation, and corrected for phase shifts by collapsing the two signed phase value filter outputs into a single non-negative estimate of contrast energy (specifically, the square root of the sum over the two squared phase values). This is a standard processing stage in the Gabor filter model [24]. The resulting rectified response vectors were weighted according to filter bank membership (5 free parameters). We estimated measurement-level population averaging using two pooling stages: a hemifield-specific pool, where filters were pooled according to whether their centers fell left or right of the vertical meridian, followed by a global pool.
We used a fixed control predictor to estimate whether coding based on pixelwise features would produce the same face-space warping we observed in our data (Fig 4). The pixelwise correlation predictor was generated by stacking all the pixels in each of the face animations into vectors and estimating the correlation distance between these intensity values.
We estimated the noise ceiling for Z-transformed Pearson correlation coefficients based on methods described previously [56]. This method estimates the explained variance that is expected for the true model given noise levels in the data. Although the true noise level of the data cannot be estimated, it is possible to approximate its upper and lower bounds in order to produce a range within which the true noise ceiling is expected to reside. The lower bound estimate is obtained by a leave-one-participant-out cross-validation procedure where the mean distance estimates of the training split are correlated against the left-out-participant’s distances, while the upper bound is obtained by performing the same procedure without splitting the data. These estimates were visualized as a shaded region in figures after reversing the Z-transform (Fig 4).
All statistical inference was performed using T tests at the group-average level (N = 10 in all cases except the occipital face area and transverse occipital sulcus, N = 9). Correlation coefficients were Z-transformed prior to statistical testing. Average Z statistics were reverse-transformed before visualization for illustrative purposes.
Fold-wise generalization performance estimates are partially dependent, which can lead to sample variance underestimates [58,59] and greater than intended false positive rates when conventional parametric statistics are used. However, we simulated the effects of this potential bias and found no consistent inflation in false-positive rates for simulations of the parameters used in the current study (S1 Code, S9 and S10 Figs). Thus, the inferential statistics reported in the current study appear to be robust to this slight dependence and maintain their intended frequentist properties.
|
10.1371/journal.pntd.0005443 | Protective immune responses against Schistosoma mansoni infection by immunization with functionally active gut-derived cysteine peptidases alone and in combination with glyceraldehyde 3-phosphate dehydrogenase | Schistosomiasis, a severe disease caused by parasites of the genus Schistosoma, is prevalent in 74 countries, affecting more than 250 million people, particularly children. We have previously shown that the Schistosoma mansoni gut-derived cysteine peptidase, cathepsin B1 (SmCB1), administered without adjuvant, elicits protection (>60%) against challenge infection of S. mansoni or S. haematobium in outbred, CD-1 mice. Here we compare the immunogenicity and protective potential of another gut-derived cysteine peptidase, S. mansoni cathepsin L3 (SmCL3), alone, and in combination with SmCB1. We also examined whether protective responses could be boosted by including a third non-peptidase schistosome secreted molecule, glyceraldehyde 3-phosphate dehydrogenase (SG3PDH), with the two peptidases.
While adjuvant-free SmCB1 and SmCL3 induced type 2 polarized responses in CD-1 outbred mice those elicited by SmCL3 were far weaker than those induced by SmCB1. Nevertheless, both cysteine peptidases evoked highly significant (P < 0.005) reduction in challenge worm burden (54–65%) as well as worm egg counts and viability. A combination of SmCL3 and SmCB1 did not induce significantly stronger immune responses or higher protection than that achieved using each peptidase alone. However, when the two peptidases were combined with SG3PDH the levels of protection against challenge S. mansoni infection reached 70–76% and were accompanied by highly significant (P < 0.005) decreases in worm egg counts and viability. Similarly, high levels of protection were achieved in hamsters immunized with the cysteine peptidase/SG3PDH-based vaccine.
Gut-derived cysteine peptidases are highly protective against schistosome challenge infection when administered subcutaneously without adjuvant to outbred CD-1 mice and hamsters, and can also act to enhance the efficacy of other schistosome antigens, such as SG3PDH. This cysteine peptidase-based vaccine should now be advanced to experiments in non-human primates and, if shown promise, progressed to Phase 1 safety trials in humans.
| Schistosomiasis, a severe disease caused by parasites of the genus Schistosoma, is prevalent in 74 countries in the Middle-east, Africa, South America and South-East Asia, and affects more than 250 million people, particularly children. Praziquantel is the most effective anti-schistosome drug and is the mainstay of mass drug administration (MDA) programs. However, concerns about the emergence of drug-resistant parasites or reduced efficacy have highlighted the urgent need for an efficacious vaccine that could interfere with parasite establishment and transmission. Here we examine the immunogenicity and vaccine potential of functionally active recombinant forms of the gut-derived cysteine peptidases of S. mansoni, cathepsin B1 (SmCB1) and cathepsin L3 (SmCL3), alone, or in combination. We demonstrate their capacity to induce high levels (>60%) of protection without the need of adjuvant against challenge infection with S. mansoni cercariae. A vaccine that combined SmCB1 and SmCL3 with another secreted molecule, S. mansoni glyceraldehyde 3-phosphate dehydrogenase (SG3PDH), induces predominant type 2 immune responses, and consistently evokes superior protection (>70%) against S. mansoni challenge infection in both outbred CD-1 mice and hamsters. This efficacious adjuvant-free cysteine peptidase-based vaccine should be brought forward into trials in non-human primates for assessment as a prospective vaccine against human schistosomiasis.
| Schistosomiasis is caused by infection with helminth parasites of the genus Schistosoma. It is a water-borne debilitating disease that prevails in 74 developing countries of the Middle East, sub-Saharan Africa, and South America, and infects >250 million people. The infective stage, the cercaria, is shed by freshwater snails in large numbers and infects their human host by penetration of the skin. The parasites migrate via the blood to the lungs and then the liver before finally settling as male and female worms in the mesenteric (Schistosoma mansoni, S. japonicum) and bladder venules (S. haematobium). Schistosomiasis infects mainly rural populations, particularly children while they bath and play in freshwater tributaries [1]. Infection proceeds relatively unnoticed until eggs released by fecund females and destined to leave via the intestine or bladder, instead become trapped in the liver, gastrointestinal tract, or urinary bladder tissues. Here they induce potent inflammatory responses that lead to the development of severe granulomatous inflammation and fibrosis in the liver or bladder. In children, symptoms include anemia, abdominal pain and diarrhea, as well as growth and cognitive impairments [1–3].
Praziquantel is the only drug readily available for the treatment of schistosomiasis and because of its low cost, safety and efficacy has been the principal means of intervention to control schistosomiasis through mass drug administration (MDA) programs. Treatment with praziquantel, however, only reaches ~13% of the target population and because of the large pill size and bitterness it is generally not recommended for children below 6 years of age [4]. Furthermore, praziquantel does not prevent reinfection, therefore requiring repeated treatment, and has reduced efficacy in people with heavy infections [3,4]. There are fears surrounding the potential for the emergence of drug-resistant parasites [3] and a recent report from an MDA program in Uganda suggests that high exposure to the drug may reduce its effectiveness [5].
An efficacious and safe vaccine administered to young children would prevent infection and diminish transmission, as well as increase the likelihood of parasite elimination [1–3]. However, to date, only a few antigens have been advanced to Phase 1 trials in humans, including Sm-TSP-2 and Sm-14 for S. mansoni and Sh28GST (Bilhvax) for S. haematobium, while only one other, Smp80, is undergoing trials in non-human primates [3, 6–9]. The dearth of vaccines in the pipeline is worrying and therefore a major effort is required to employ new ways of discovering, shortlisting and delivering vaccine candidates.
Our research has focused on the identification of vaccine molecules in the parasite excretory-secretory products (ESP) because these are involved in host-parasite interaction and induce potent immune responses [10]. Contrary to cytosolic and surface membrane antigens, we reasoned that these were more accessible to attack by antibodies and activated immune effector cells [10]. We recently demonstrated that candidate vaccines in ESP formulated in the presence of the type 2 cytokines, thymic stromal lymphopoietin (TSLP), interleukin (IL)-25 or IL-33, induced considerably high protection in mice, in the range of 50% - 75%, against a challenge infection of S. mansoni cercariae [11]. We also discovered that a similar level of protection could be achieved by delivering antigens with the cysteine peptidase, papain, which also induces type 2 immune responses [11–13]. To avoid employing a plant-derived peptidase, we then showed that the gut-derived papain-like cysteine peptidases of Schistosoma mansoni, cathepsin B1 (SmCB1), and Fasciola hepatica, cathepsin L1 (FhCL1), could induce predominant type 2 immune responses and highly significant (P < 0.005) reduction of between 66% and 75% in challenge S. mansoni and S. haematobium worm burden and worm egg counts of outbred mice and hamsters [12, 14, 15].
The objectives of the present study were to compare and contrast the immunogenicity and protective potential of SmCB1 with another gut-derived schistosome cysteine peptidase, SmCL3 [16], alone and in combination in CD-1 outbred mice. We report that functionally active recombinant SmCB1 and SmCL3-based vaccine administered subcutaneously induces highly significant (P < 0.002) protection of > 60% against S. mansoni challenge infection in both mice and hamsters. A vaccine combining SmCB1 and SmCL3 with another secreted molecule, S. mansoni glyceraldehyde 3-phosphate dehydrogenase (SG3PDH), which is also located at the host-parasite interface [17–19], elicited impressive levels (>70%) of protection to S. mansoni challenge infection suggesting that this efficacious trivalent vaccine should now be brought forward into trials in non-human primates for assessment as a potential vaccine to control human schistosomiasis.
All animal experiments were performed following the recommendations of the current edition of the Guide for the Care and Use of Laboratory Animals, Institute of Laboratory Animal Resources, National Research Council, USA, and were approved by the Institutional Animal Care and Use Committee (IACUC) of the Faculty of Science, Cairo University, permit numbers CUFS F PHY 21 14 and CUFS-F-Imm-5-15.
Female CD-1 mice and female Syrian hamsters (Mesocricetus auratus) were bred at the Schistosome Biological Materials Supply Program, Theodore Bilharz Research Institute (SBSP/TBRI), Giza, Egypt until 6 weeks of age and then maintained throughout experimentation at the animal facility of the Zoology Department, Faculty of Science, Cairo University. Cercariae of an Egyptian strain of S. mansoni were obtained from SBSP/TBRI, and used immediately after shedding from Biomphalaria alexandrina snails. CD1 mice were infected via whole body exposure as previously described [11, 13]. Hamsters were anesthetized, the abdomen shaved and wetted with sterile deionized water, and then exposed to cercariae via the ring method as previously described [12].
Recombinant S. mansoni glyceraldehyde 3-phosphate dehydrogenase (rSG3PDH) expressed in the bacterium Escherichia coli was prepared and purified to homogeneity [17] (S1 Fig). This preparation contained <0.06 Endotoxin Units/ml as judged by the Pyrogen Gel-Clot Limulus Amebocyte Lysate test. Functionally active S. mansoni cathepsin B1 (SmCB1) and cathepsin L3 (SmCL3) were produced in methyltrophic yeast Pichia pastoris GS115 (Invitrogen) and PichiaPink (Thermo Fisher) strain, respectively, using methods previously described in our laboratory [20]. Recombinant cathepsins were purified by Ni-NTA column, desalted by dialysis, and stored in phosphate buffered saline, pH 7.3 at -80°C. Cathepsin peptidase purity was determined by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) and activity using the fluorogenic substrate Z-Phe-Arg-NHMec (Sigma-Aldrich, UK) and PolarStar Omega fluorescence reader (BMG Labtech) [20–22] (S1 Fig). Concentration of the purified peptidases was evaluated using the Protein Assay Kit of BioRad.
Blood samples were obtained from individual naïve, unimmunized and immunized mice (3 or 4 per group) seven days following infection with viable S. mansoni cercariae, unless stated otherwise. Sera were separated and stored at -20°C.
Epidermal cells (EC) were isolated from three mice per group two days after S. mansoni infection, at the time when larvae are still resident in the epidermis, following the protocol described by Jensen et al. [23] with minor modifications. The whole procedure was recently detailed [13]. Lymph node cells (LNC) from inguinal and popliteal lymph nodes (LN) were recovered (3 mice per group) four days after S. mansoni infection at the time when larvae are in the dermis or dermal capillaries. LNC were also obtained from mice 21 days after immunization with vaccine molecules. Spleen cells (SC) were prepared from spleens removed from three mice per group at seven and 14 days after S. mansoni infection [11, 13, 14].
EC, LNC, and SC were re-suspended in RPMI-1640 medium supplemented with 200 U/ml penicillin, 200 μg/ml streptomycin, 25 mM HEPES, 50 ng/ml amphotericin, 5 x 10–5 M 2-mercaptoethanol, and 5% fetal calf serum (culture medium). Twenty μg/ml polymyxin B (Sigma-Aldrich) was added as an inhibitor of any residual lipopolysaccharide contamination of recombinant antigens. EC, LNC and SC were cultured at a concentration of 1 x 106 cells/200 μl culture medium/well in duplicate wells of 96 round-bottomed well plates (Corning Costar, Bedford, MA), stimulated with 0 or 20 μg/ml membrane filter (0.45 μm)-sterilized immunogen, and maintained at 37°C in a humidified atmosphere containing 3.0% CO2. After 72 h of incubation, cultured cells were frozen and thawed for release of intracellular cytokines, the lysate centrifuged (15,000 x g) and supernatants stored at -76°C until assayed for cytokine concentrations by capture enzyme-linked immunosorbent assay (ELISA) [11, 13, 14].
Mouse serum antibody titer and isotype were assessed by indirect ELISA for binding to 250 ng/well immunogen in duplicate wells [11, 13, 14]. Horseradish peroxidase (HRP)-labeled goat anti-mouse IgG (H+L) conjugate (Promega, Madison, WI) was used to detect bound antibodies. Individual sera were serially diluted to determine the antibody titer and select the appropriate dilution for assessing the antibody isotypes. Based on these results, sera were diluted 1:100 or 1:200 to estimate the level of IgM and IgG class antibodies, and 1:25 for IgE and IgA antibodies. Biotin-labeled rat monoclonal antibody to mouse IgG1 (Pharmingen, San Diego, CA), IgA, and IgE (BioLegend, San Diego, CA), was diluted 1:500 in washing buffer (0.01 M phosphate buffered-saline, pH 7.2, 0.05% Tween 20/0.1% bovine serum albumin). Alkaline phosphatase- (AKP) or HRP-labeled streptavidin was diluted 1:1000. Monoclonal antibody to IgM, IgG2a and IgG2b (Pharmingen) labeled with AKP were diluted 1:500, 1:1000 and 1:1000, respectively. Reactivity was estimated spectrophotometrically after adding SureBlue TMB Microwell Peroxidase Substrate (Kirkegaard and Perry Laboratories, Inc., Gaithersburg, MD) or p-nitro phenyl phosphate (PNPP) substrate (Calbiochem, San Diego, CA).
Release of mouse IL-1β, IL-5, IL-12, IL-17, IL-25, IFN-γ, TSLP (ELISA MAX Set, BioLegend, San Diego, CA), and IL-13 (DuoSet ELISA Development System, R&D System Europe) was measured in supernatants of duplicate cell cultures by capture ELISA, following the manufacturer's instructions.
Worm burden as well as liver and intestine worm egg load in individual mice and hamsters (5–10 animals per group) were evaluated six to seven weeks after the challenge infection with S. mansoni cercariae [11, 13, 15]. Percent (%) change was evaluated by the formula: % change = [mean number in unimmunized infected controls–mean number in immunized infected animals / mean number in infected controls] × 100.
Egg developmental stages were evaluated using 3–5 fragments of the distal portion of the ileum. After washing in 0.9% saline solution and slight drying on absorbent paper, each intestinal fragment was placed between 2 slides and analyzed by light microscopy to classify the eggs. For each fragment, up to 400 eggs were counted and classified according to their developmental stage as immature, with miracidium occupying less than two-thirds of the shell, mature containing an already developed miracidium, and non-viable, dead, clearly calcified, opaque [24].
The remaining control and immunized mice were infected with 150 cercariae of S. mansoni 3 weeks after the second immunization. EC and LNC were isolated 2 and 4 days later, respectively from three mice/group, incubated with 0 or 20 μg/ml immunogen, and cell culture supernatants assessed for the levels of released TSLP, IL-1, IL-12, IL-25. On day 7 and 14, when a large proportion of migrating larvae are in the lung capillaries and liver sinusoids, respectively, serum and SC from three mice/group were assessed for humoral and cytokine responses to the immunogen. Parasitological parameters (see above) were evaluated for the remaining 6–8 mice per group at 49 days after the challenge infection (Fig 1A).
(b) Evaluation of the immunogenicity and vaccine potential of SmCL3 alone or in combination with SmCB1 and SG3PDH (Fig 1B). Fifty-two CD-1 mice were randomly divided into 4 groups of 13 mice, and immunized subcutaneously twice, with a 3 week-interval, with 100 μl D-PBS containing 0 (control Group 1), 10 μg SmCL3 (Group 2), 10 μg SmCL3 and 10 μg SmCB1 administered subcutaneously on different sides of the tail (Group 3), 10 μg SmCL3 and 10 μg SmCB1 administered as above and 10 μg SG3PDH intramuscularly, to avoid immediate exposure to the cysteine peptidases (Group 4). Three weeks after the second immunization all mice were challenged with 120 cercariae of S. mansoni. Humoral and SC cytokine responses to the immunogens at seven days after infection were assessed using three mice per group. Parasitological parameters were evaluated for the remaining 10 mice per group at 42 days after the challenge infection (Fig 1B).
To assess the vaccine in a different host, 40 hamsters were allocated into 4 groups of 10 animals, and immunized following a protocol similar to that described in (c) above with the exception that these were administered 15 μg rather than 10 μg of each immunogen for both immunizations. Five hamsters in each group were challenged 4 and 6 weeks after the last immunization with 200 or 150 cercariae (Fig 2B), respectively. Parasitological parameters were examined for all hamsters on day 42 or 43 after challenge infection.
All values were tested for normality. Mann–Whitney 2-tailed test was used to analyze the statistical significance of differences between experimental and control values and considered significant at P < 0.05.
We have previously demonstrated the ability of two immunizations with functionally active SmCB1 to protect mice against a challenge infection of S. mansoni. Here we have compared a second cysteine protease, SmCL3 that is also expressed in the S. mansoni digestive tract [16], to SmCB1 for its capacity to induce immune responses and protection against parasite challenge. In keeping with previous results [14] subcutaneous immunization of mice with SmCB1 induced IgG antibody responses (titres >1:6400) which can be detected at 7 and 21 days after the second immunization (Fig 3, panel A). At 7 and 14 days following the challenge infection with 150 cercariae the antibody responses to SmCB1 were significantly (P < 0.05) boosted as shown by ELISA performed at serum dilutions of 1:200 (Fig 3, panel B). Antibodies binding to SmCB1 were of the IgM, IgG1, and IgG2b isotype to a titre of 1:400 whereas little or no IgE was detected (titre of 1:25). By contrast, we did not detect any antibodies to SmCL3 at 7 or 21 days after two immunizations. Although, antibodies to SmCL3 were detected following parasite challenge, these were of very low titre and markedly lower than those elicited by SmCB1 (Fig 3, panel B).
To examine early immune responses in the skin of immunized and challenged mice, EC were removed two days after the challenge infection, incubated in the presence or absence of SmCB1 and SmCL3 and secretion of the cytokine measured. Epidermal cells from control or SmCB1- or SmCL3-immunized mice did not release detectable levels of IL-1, IL-12, or IL-25 following stimulation in culture for 72 h with 0 or 20 μg/ml immunogen. However, both unstimulated and immunogen-stimulated EC from the test mice did release detectable levels of TSLP with levels produced by EC of SmCB1-immunized mice significantly (P < 0.05) lower compared to infected controls and SmCL3-immunized mice (Fig 4A).
Inguinal and popliteal LNC taken from mice 4 days after infection released low but significant (P < 0.05) levels of IL-4 in response to stimulation with SmCB1 or SmCL3 (Fig 4B) but only SmCB1 induced significant (P < 0.005) levels of IL-17 (Fig 4C) and IFN-γ (Fig 4D). Similar cytokine responses were observed for SC taken at 7 and 14 days after the challenge infection and stimulated with SmCB1 and SmCL3 (S2 Fig).
Both SmCB1 and SmCL3 vaccine antigens induced highly significant (P < 0.005) reduction in total worm burdens (54.2% and 65.4%, respectively) that was reflected in comparable reduction in both male and female worms together with a significant (P < 0.05) decrease in worm egg counts in the liver and small intestine tissues of mice (Table 1). No significant differences were observed between SmCB1 and SmCL3 vaccine efficacy and both immunizations also induced highly significant (P < 0.005) increase in the percentage of dead ova compared to the infected control mice (Fig 5).
Consistent with the above experiments we found that only one out of 3 mice immunized with SmCL3 produced antibodies that bound to SmCL3 in levels significantly (P < 0.05) higher than unimmunized 7 day infected mice. Addition of SmCB1 and SG3PDH to SmCL3 did not influence production of anti-SmCL3 antibodies even though antibodies to SmCB1 (to a titre higher than 1:800) were induced in mice immunized with SmCL3 + SmCB1 (Fig 6A). No SmCL3- or SmCB1-binding IgE antibodies were detected even at 1:25-diluted serum from naïve or immunized mice from all groups, confirming the data revealing that SmCL3 vaccine does not induce IgE specific antibodies, and that SmCB1 is a weak IgE specific antibody inducer, but not when incorporated with SG3PDH [14].
SC taken from non-infected or unimmunized control CD-1 mice that were infected for 7 days with S. mansoni did not release detectable levels of IL-4, IL-5, IL-13, IL-17 or IFN-γ when stimulated with SmCL3 or SmCB1. SC obtained 7 days after infection with S. mansoni from mice that were immunized with SmCL3 alone, SmCL3 + SmCB1 or SmCL3+ SmCB1 + SG3PDH did not release IL-5, IL-13, IL-17 or IFN-γ when stimulated with SmCL3, yet released IL-4 in response to SmCL3 and especially SmCB1, similarly to the observation reported previously with SmCB1 and FhCL1 [14]. The SC response to SmCB1 was strikingly different from SmCL3 as the former induced greater percent of mice responding with detectable levels of IL-17 or IFN-γ (Fig 6B and 6C).
Consistent with the first vaccine experiment reported above, CD-1mice immunized twice subcutaneously with functionally active SmCL3 protease displayed a highly significant (P < 0.0001, Mann-Whitney) reduction in challenge worm burdens (48%) compared to the control mice when challenged with S. mansoni cercariae three weeks after the second immunization. The level of protection was not significantly increased when SmCL3 and SmCB1 were employed as a cocktail vaccine (reduction of 51.1% in worm burdens). However, significantly (P = 0.0035) higher protection of 69% was observed when the vaccine was composed of SmCL3 + SmCB1 + SG3PDH (Fig 7A). Furthermore, immunization of CD-1 mice with SmCL3 + SmCB1 was associated with highly significant decrease in parasite egg load in liver and small intestine (53% and 59%, respectively, P <0.0001) which was further decreased when SG3PDH was added to the vaccine cocktail (73% and 70%, respectively, P <0.0001) (Fig 7B and 7C).
A third series of vaccines trials was performed to support the protective efficacy of SmCB1 and SmCL3 and to compare this data in two laboratory outbred models, CD-1 mice and hamsters. The focus of the previous experiments was to compare the immunogenicity of SmCB1 and SmCL3. The focus in the third experiment was to compare the adjuvant effects of SmCB1 and SmCL3 on responses to SG3PDH. Accordingly, we analyzed the antibody response to SG3PDH in the presence of the peptidase in the vaccine cocktail (Fig 8A and 8B). Antibody reactivity to SG3PDH was observed in challenged mice immunized with SmCB1+ SG3PDH, SmCL3+ SG3PDH or SmCB1+SmCL3+ SG3PDH and was significantly (P < 0.005) higher than non-immunized mice (Fig 8A). The adjuvant effect of SmCB1 and SmCB1 + SmCL3 on mouse humoral reactivity to SG3PDH was characterized by the production of substantial amounts of IgG1 antibodies (Fig 8B) which contrasts with our previous studies showing that murine humoral responses to SG3PDH administered alone is limited to IgG2a and IgG2b [10, 11]. Cytokine production by SC in response to SG3PDH was characterized by the detection of low but significant levels of IL-4, IL-5, IL-13, IL-17 and particularly IFN-γ (Fig 8C).
The combination of SmCB1 or SmCL3 with SG3PDH as a vaccine cocktail induced highly significant (P <0.02 - <0.008) reduction in challenge worm burden in immunized CD-1 mice (66–71%) and hamsters (51–66%) compared to unimmunized animals. A very significant (P = 0.0006) reduction of 76% was achieved in total worm burden (reflected in both male and female challenge worms) in mice given the vaccine formula SmCB1 + SmCL3 + SG3PDH. The cocktail vaccine also elicited highly significant (P < 0.008) decrease (66% - 74%) in challenge worm burden in hamsters, challenged either 4 or 6 weeks after the second immunization (Table 2). The decrease in total egg counts in liver and intestine was less pronounced in hamsters compared to mice. However, SmCB1 or SmCL3 + SG3PDH and SmCB1 + SmCL3 + SG3PDH did induce significant (P < 0.05) reduction in percentage of immature ova and highly significant (around P = 0.0001) increase of >80% in the percentage of dead ova in small intestine of immunized as compared to unimmunized hosts (Table 2).
There is an urgent need to develop vaccines for schistosomiasis given our almost complete dependency on the drug praziquantel and the lack of new compounds entering the development pipeline [4]. Peptidases, particularly those involved in the degradation of blood proteins for nutrient, have for a long time been considered important targets at which we could direct both drugs and vaccines [21, 22]. Cathepsin B, SmCB1, is one of the most prominent and immunogenic peptidase secreted from the gastrodermis of schistosomes [20–22, 25]. Recently we discovered that this enzyme could induce high levels of protection in mice against a challenge infection with cercariae of S. mansoni when delivered subcutaneously as a functionally active recombinant form without adjuvant [14, 15]. The protection afforded by active cathepsin B was suggested to be partly due to its ability to activate components of the innate immune system in a non-specific manner as the cysteine peptidase papain from plants [11, 13] and the papain-like cathepsin L1, FhCL1, of the related trematode Fasciola hepatica [14, 15], also had the capacity to induce protection against S. mansoni. Notwithstanding, we felt it judicious to pursue SmCB1 as a protective vaccine molecule rather than these other peptidases for schistosomiasis because it was derived from the parasite itself. Furthermore, we found that inactive SmCB1 also possesses protective properties, albeit lower than active cathepsin B [14], and recently Ricciardi et al. [26] showed that SmCB1 formulated in Montanide ISA 720 VG induced elicited high-level protection against S. mansoni. In the present study, we pursued another schistosome peptidase that was recently shown to be located in the schistosome digestive tract, S. mansoni cathepsin L3 (SmCL3) [16], and investigated its immunogenicity and protective properties alone and in combination with the cathepsin B peptidases without adjuvant. To do this we produced a functionally active SmCL3 protease in the methylotrophic yeast Pichia pastoris similar to the manner in which we produced SmCB1 [20].
To our surprise, we found that SmCL3 induced weaker humoral immune responses when administered subcutaneously to outbred CD-1 mice compared to SmCB1 despite that SmCL3 displays only 46% identity with mouse cathepsin L [16]. Two immunizations of SmCL3 elicited no detectable IgG antibody responses to the immunogen at 7 and 21 days after secondary immunization, reflecting an inability to induce T helper cells and release of cytokines that promote antibody production. The SmCL3 poor immunogenicity could be ascribed to its uniquely broad proteolytic substrate specificity [16]. In support, the inability of SmCL3 to strongly activate the cellular arm of the host immune system was also observed in vitro whereby ex vivo-stimulated SC released low, albeit detectable, levels of IL-4. By contrast, SmCB1 induced a greater level of cytokine responses from LNC and SC and these were not limited to the type 2 cytokine IL-4 as IL-17 and IFN-γ were also detected. Furthermore, IgM, IgG1, IgG2a and IgG2b antibodies to SmCB1 were produced in vivo at 7 and 21 days after secondary immunization. Although a low level of anti-SmCL3 antibody was detected following a challenge infection of vaccinated mice these were far lower than those observed for SmCB1, suggesting a more prominent infection-associated anamnestic response for the latter peptidase which is supported by it well-known immunogenicity in experimental and natural infection [27]. However, both peptidases stimulated detectable levels of TSLP from EC cells in vitro that suggests a role for this master cytokine of type 2 immune responses [28, 29] in initiating protective responses. Our previous studies have shown that delivery of TSLP to mice induced type-2 related protective immune responses against S. mansoni challenge infection [11], which could have involved the recruitment and activation of eosinophils, basophils, and mast cells [29–36].
Despite the lower immune responses induced by SmCL3 compared to SmCB1, the peptidase elicited a highly significant (P < 0.005) reduction (~65%) in challenge S. mansoni worm burden, and not significantly different from the protection levels induced by SmCB1 (59%). In this trial immunization with either S. mansoni cysteine peptidase was associated with low but significant (P < 0.05) decreases in parasite eggs within the intestine and liver in comparison to the infected controls, and a very significant effect (P < 0.005) on the maturation and viability of eggs in these tissues observed. This is relevant as dead ova are likely unable to elicit inflammatory responses in the host or participate in transmission of the infection [24]. The data may indicate that cysteine peptidase-induced immune responses that may not only act directly on the female worm reproductive system via reducing fecundity but also interfere somehow in the development of the embryo within the egg. Perhaps the latter effect is due a blocking of the ability of female worms to digest blood proteins effectively and, consequently, acquire sufficient amino acids for egg protein synthesis.
In repeat experiments, an analysis of immune responses to SmCL3 delivered in combination with SmCB1 or SG3PDH, confirmed its lack of ability to elicit production of considerable levels of SmCL3-specific antibodies (IgG1 antibodies were observed but only to a titer of 1:25) and to stimulate SC ex vivo barely detectable levels of IL-4. Despite this apparent lack of immunogenicity significant (48%, P < 0.0005) reduction in total worm burden and parasite egg load in liver and small intestine at three weeks after infection was observed. However, the protective potential of SmCL3 was not increased upon inclusion of SmCB1 to the vaccine. Likewise, we previously found that a combination of FhCL1 and SmCB1 peptidases did not elicit the additive protective effect of each peptidase given alone [14]. However, inclusion of SG3PDH with the SmCB1 and SmCL3 cysteine peptidases increased protection to >69% in S. mansoni worm burden and improved on the vaccine’s capacity to reduce egg loads in the liver and small intestine tissues. The enhanced protection could be associated with immune responses directed against lung-stage larvae that express SG3PDH on their surface and also release it as a component of their ESP [18, 19].
In a final series of vaccine trials (shown in Table 2) we set out to support the data obtained in previous tests but also to compare the protective potential in mice and hamsters. Furthermore, in the case of hamsters, the time period between the second immunization and challenge infection with S. mansoni cercariae was compared at four and six weeks. These studies were also motivated by the recently published discussion article by Wilson and colleagues [37] who argued that the mouse model may give spurious vaccine results due to physiological features of the murine pulmonary system (non-specific, inflammation-associated leakiness of the lung vasculature that drive parasites into the alveoli) [37]. Our previous vaccine trials with cysteine peptidases [11, 14, 15] were cited as an example, if not evidence, of a vaccine that could induce these non-specific responses. In these trials in mice, antibody and cytokine responses to SG3PDH were observed and revealed that, (a) IgG responses to SG3PDH was comparable in all three groups combination vaccines, SmCL3 + SG3PDH, SmCB1 + SG3PDH and SmCB1 + SmCL3 + SG3PDH and comprised IgG1 and IgG2 isotypes, and (b) type 1, 2 and 17 cytokine were detected in all groups indicating the induction of mixed Th1/Th2 responses, although the group given SmCB1 + SmCL3 + SG3PDH exhibited a more skewed Th1/Th17 response to SG3PDH (Fig 7). These responses are similar to the profile observed when SG3PDH was co-administered with the cytokines TSLP, IL-25 and IL-33 [11]. The protection reported by Ricciardi et al. [26] using Montanide-formulated SmCB1 also was attributed to a mixed Th1/Th2/Th17 response. Most importantly, the data revealed that combining SmCB1 or SmCL3 with SG3PDH provided an adjuvant-like effect that was accompanied by highly significant (P < 0.05- <0.005) reduction in challenge worm and worm egg burden and the proportion of immature ova (Table 2), Consistent with the observed potent antibody and cellular immune responses to SG3PDH, the protection level achieved with the complete vaccine formula, SmCB1 + SmCL3 + SG3PDH, was remarkably high, ~76% in total worm burden. High levels of protection were also observed in vaccinated hamsters with SmCB1 or SmCL1 and SG3PDH (50–62%), with particularly high reductions of 66% and 74.1% in total worm burden recorded in hamsters challenged four and six weeks after the second immunization with SmCB1 + SmCL3 + SG3PDH. These protective anti-worm data were accompanied with consistent, reproducible, and highly significant increases in the number of dead ova within the small intestine.
It has been several decades since the World Health Organization (WHO) proposed a threshold level of 40% reduction in challenge worm burden in an experimental host for sponsoring a schistosome vaccine for pre-clinical trials and still very few molecular vaccines have passed this relatively-low bar [3, 37]. Here we provide further evidence for the protective capacity of cysteine peptidases when administered subcutaneously in functionally active form. The mechanism by which cysteine peptidase elicit protective responses is still not fully understood, especially considering their low immunogenicity when administered in this fashion (particularly SmCL3, as observed in this study), but we suggest that a contributing factor is the provocation of a Th2 or Th1/Th2/Th17 immune response by activating innate immune responses. Further work is required to explore whether the mechanism activation of immune responses by cysteine proteases and whether this can be enhanced by other protective surface or ESP antigens in order to promote greater anti-parasite activity as we have shown in this study with SG3PDH. Although we have not addressed all the suggestions of Wilson et al. [37] to test their assertions that the mouse model is one that predisposes towards spurious or misleading vaccine results, we have at least demonstrated that protection can be induced in another rodent model hamster, and that protection levels are retained even when he challenge infection is given six weeks after the final immunization. It must be also emphasized that our vaccine formulation is delivered without adjuvant and, therefore, the likelihood of non-specific physiological factors being involved in protection in lessened. We believe that the multiple trial data in the present study justifies the candidacy of the cysteine peptidase-based vaccine for funding to support independent trials in mice and non-human primates and, if shown promise, should be positioned in a development pipeline for human Phase 1 trials.
|
10.1371/journal.pntd.0000727 | Mechanisms of Vascular Damage by Hemorrhagic Snake Venom Metalloproteinases: Tissue Distribution and In Situ Hydrolysis | Envenoming by viper snakes constitutes an important public health problem in Brazil and other developing countries. Local hemorrhage is an important symptom of these accidents and is correlated with the action of snake venom metalloproteinases (SVMPs). The degradation of vascular basement membrane has been proposed as a key event for the capillary vessel disruption. However, SVMPs that present similar catalytic activity towards extracellular matrix proteins differ in their hemorrhagic activity, suggesting that other mechanisms might be contributing to the accumulation of SVMPs at the snakebite area allowing capillary disruption.
In this work, we compared the tissue distribution and degradation of extracellular matrix proteins induced by jararhagin (highly hemorrhagic SVMP) and BnP1 (weakly hemorrhagic SVMP) using the mouse skin as experimental model. Jararhagin induced strong hemorrhage accompanied by hydrolysis of collagen fibers in the hypodermis and a marked degradation of type IV collagen at the vascular basement membrane. In contrast, BnP1 induced only a mild hemorrhage and did not disrupt collagen fibers or type IV collagen. Injection of Alexa488-labeled jararhagin revealed fluorescent staining around capillary vessels and co-localization with basement membrane type IV collagen. The same distribution pattern was detected with jararhagin-C (disintegrin-like/cysteine-rich domains of jararhagin). In opposition, BnP1 did not accumulate in the tissues.
These results show a particular tissue distribution of hemorrhagic toxins accumulating at the basement membrane. This probably occurs through binding to collagens, which are drastically hydrolyzed at the sites of hemorrhagic lesions. Toxin accumulation near blood vessels explains enhanced catalysis of basement membrane components, resulting in the strong hemorrhagic activity of SVMPs. This is a novel mechanism that underlies the difference between hemorrhagic and non-hemorrhagic SVMPs, improving the understanding of snakebite pathology.
| Snakebite accidents by vipers cause a massive disturbance in hemostasis and tissue damage at the snakebite area. The systemic effects are often prevented by antivenom therapy. However, the local symptoms are not neutralized by antivenoms and are related to the temporary or permanent disability observed in many patients. Although the mechanisms involved in coagulation or necrotic disturbances induced by snake venoms are well known, the disruption of capillary vessels by SVMPs leading to hemorrhage and consequent local tissue damage is not fully understood. In our study, we reveal the mechanisms involved in hemorrhage induced by SVMPs by comparing the action of high and low hemorrhagic toxins isolated from Bothrops venoms, in mouse skin. We show remarkable differences in the tissue distribution and hydrolysis of collagen within the hemorrhagic lesions induced by high and low hemorrhagic metalloproteinases. According to our data, tissue accumulation of hemorrhagic toxins near blood vessel walls allowing the hydrolysis of basement membrane components, preferably collagen IV. These observations unveil new mechanistic insights supporting the local administration of metalloproteinases inhibitors as an alternative to improve snakebite treatment besides antivenom therapy.
| Snakebite envenoming is an important neglected disease in many tropical and subtropical developing countries. As recently reviewed, globally, venomous snakebite is estimated to affect more than 421,000 humans per year, with 20,000 of fatalities. However, if we take into account the non-reported accidents, these data may be as high as 1,841,000 envenomings and 94,000 deaths [1]. Antivenom therapy was set at the end of 19th century and is still the only efficient approach to treat snakebites. It cures systemic symptoms of envenoming while the local effects are not covered and usually leads to temporary or permanent disability observed in many patients [2], [3]. In Brazil, the majority of the accidents reported to the Ministry of Health are caused by viper snakes [4]. The victims of viper envenoming frequently present systemic disturbances in hemostasis including spontaneous bleeding and blood incoagulability, and strong local effects characterized by edema, ecchymoses, blisters and extensive hemorrhage [2]. Hemorrhagic toxins play an important role in vascular damage and subsequent generation of ischemic areas that largely contribute to the onset of local tissue necrosis that may result in amputation of affected limbs [5], [6].
The pathogenesis of venom-induced hemorrhage involves direct damage of microvessels by the snake venom metalloproteinases (SVMPs). They are multidomain Zn2+-dependent proteinases that share structural and functional motifs with other metalloproteinases, such as MMPs (Matrix Metalloproteinases) and ADAMs (A Disintegrin And Metalloproteinase) [7], [8]. SVMPs are classified from PI to PIII according to their domains constitution (Reviewed by Fox and Serrano [9]). The mature form of the PI class is composed only of the metalloproteinase domain with the characteristic zinc-binding site present in all classes of SVMPs, MMPs and some ADAMs. P-II and P-III SVMPs exhibit additional non-catalytic domains, such as disintegrin, disintegrin-like and cysteine-rich domains, similar to those found in ADAMs, which are related to adhesive properties [9]. Despite sharing similar catalytic activity, not all SVMPs induce hemorrhage in in vivo models. In general, P-III SVMPs that include disintegrin-like and cysteine-rich domains are potent hemorrhagic toxins while P-I SVMPs show reduced hemorrhagic activity. There are also a number of non-hemorrhagic SVMPs that may be found preferentially in the P-I class and rarely in P-III class, which often function as pro-coagulant enzymes [10], [11], [12].
The mechanism of hemorrhage induced by SVMPs has been investigated in several studies [13], [14], [15], [16], [17]. However, the precise molecular and cellular events associated with microvessel disruption remain unknown. The degradation of vascular basement membrane components has been proposed as a key event for the onset of capillary vessel disruption. The four major components of basement membranes are type IV collagen, laminin, nidogen/entactin and perlecan. Type IV collagen and laminin individually self-assemble into supramolecular structures, and both networks are crucial for basement membrane stability [18]. Nidogen/entactin and perlecan bridge the laminin and type IV collagen networks, increasing their stability, influencing the structural integrity of basement membranes [18], [19]. Thus, the hydrolysis of vascular basement membrane components by SVMPs could profoundly affect the stability of the endothelium, resulting in bleeding. In this regard, in vitro, SVMPs efficiently degrade basement membrane components such as laminin, type IV collagen, nidogen/entactin, presenting minor effects on interstitial collagens [20]. However, catalytic activity is apparently similar in hemorrhagic and non-hemorrhagic SVMPs, indicating that the hydrolysis of basement membrane components is not the only mechanism acting on vascular damage induced by the hemorrhagic toxins.
Endothelial cells have also been investigated as potential targets of hemorrhagic toxins. The survival signals promoted by endothelial cell anchorage can be disrupted by SVMPs using mechanisms dependent or independent of their proteolytic activity. Both P-I and P-III SVMPs interfere with adhesion components involved in focal adhesion between endothelial cells and the extracellular matrix, affecting the organization of actin filaments and stress fibers, which culminates in cell death by apoptosis [21], [22]. However, apoptosis of endothelial cells shows little correlation with the hemorrhage induced by SVMPs. Hemorrhagic and non-hemorrhagic SVMPs induce comparable rates of apoptosis in endothelial cells in culture [23] and the onset of hemorrhage induced by SVMPs occurs much earlier than the induction of apoptosis of endothelial cells in vitro. In addition, no apoptosis of endothelial cells was observed in SVMP-induced hemorrhage in the dermis of mouse ear skin, in vivo [24].
An additional mechanism involved in SVMPs hemorrhagic activity could be related to the accumulation of hemorrhagic SVMPs close to capillary vessels, through the adhesive properties of the non-catalytic domains, allowing the hydrolysis of basement membrane components and disruption of the blood vessels. This mechanism has been suggested previously [14], [16], [25], but up to the moment, there are no experimental evidences that this occurs in vivo. In this study, we evaluated this hypothesis analyzing the tissue distribution of jararhagin, a highly hemorrhagic P-III SVMP, and BnP1, a weakly hemorrhagic toxin from P-I class, and the hydrolysis of basement membrane collagen and laminin within the hemorrhagic lesions, using the mouse skin as experimental model. We clearly showed a correlation between the binding of toxins to basement membrane and their ability to induce hemorrhage. Moreover, we showed the in situ degradation of basement membrane collagen IV, but not laminin, suggesting that collagen is an important target for the tissue accumulation of hemorrhagic SVMPs.
The conducts and procedures involving animal experiments were approved by the Butantan Institute Committee for Ethics in Animal Experiments (License number CEUAIB 191/2004).
BALB/c mice (18–22 g body weight) were used throughout the study.
As a model for class P-I SVMP with low hemorrhagic activity, we used BnP1 (GI:172044591), a 25 kDa metalloproteinase, isolated from Bothrops neuwiedi venom according to Baldo et al., [23], which contains only the catalytic domain. Jararhagin (GI:62468) was used as a model of highly hemorrhagic P-III SVMP, comprised of catalytic, disintegrin-like and cysteine-rich domains. It was isolated from Bothrops jararaca venom, as previously described [26]. Jararhagin-C was isolated from Bothrops jararaca venom as described [27]. It is devoid of catalytic activity and does not induce hemorrhage, as it contains only the non-catalytic domains of jararhagin – disintegrin-like and cysteine-rich. BSA (Bovine serum albumin) was used as control of non-toxic protein in different set of experiments. Toxins were used in the native form or labeled with Alexa Fluor 488® (Molecular Probes, USA), following the instructions of the manufacturer.
Groups of three BALB/c mice were intradermically injected with jararhagin (10 µg) and BnP1 (5 µg or 50 µg), dissolved in 20 µL of PBS (phosphate buffered saline). The control group received only 20 µL of PBS. After 15 minutes, the animals were sacrificed by CO2 inhalation and the dorsal skin corresponding to the site of the injection was carefully dissected out and fixed in methacarn solution (60% methanol, 30% chloroform, 10% glacial acetic acid) for 3 hour at 4°C and thereafter dehydrated in ethanol and embedded in Paraplast (Merck, Germany). Sections of 5 µm were adhered to glass slides using 0.1% poly-L-Lysine (Sigma, UK) and dried at room temperature. Sections were dewaxed in xylene and hydrated in distilled water. Each of the succeeding steps was followed by washing with PBS. Some sections were stained with hematoxylin and eosin for histological analysis. The detection of collagen fibers was performed by staining the sections using the picrossirus method, according to a previously described protocol [28].
The staining of basement membrane components was performed by immunofluorescence assays. After dewaxing and hydrating, the slides were submitted to antigen retrieval using enzymatic treatment of the sections with 4 mg/mL solution of pig pepsin (1,120 units/mg protein) (Sigma, UK) in acid buffer (pH 2.2), for 10 minutes at room temperature, followed by incubation with blocking solution (PBS/BSA 10% and goat serum 1∶1) for 1 hour at room temperature. Afterwards, the sections were incubated with goat anti-rabbit type IV collagen polyclonal antibody (Chemicon, USA), at a 1∶40 dilution, or goat anti-rabbit laminin polyclonal antibody (Chemicon, USA), at a 1∶40 dilution, for 18 hours at 4°C. After washing with PBS, the sections were incubated with Alexa fluor 488 goat anti-rabbit IgG (Molecular Probes, USA), at a 1∶1000 dilution, for 90 minutes at room temperature. The nuclear staining was performed with DAPI (4′, 6′-diamino-2-phenylindole, Sigma, UK), at a 1∶1000 dilution. Negative control consisted of omitting the primary antibody step from the protocol. The sections were examined with a Confocal Microscope LSM 510 Meta (Zeiss, Germany).
The distribution of toxins in the skin vasculature was evaluated after intradermically injection of 10 µg of Alexa Fluor 488-labeled jararhagin (Alexa488-Jar), 5 µg of Alexa Fluor 488-labeled jararhagin-C (Alexa488-Jar-C) or 50 µg of Alexa Fluor 488-labeled BnP1 (Alexa488-BnP1) in BALB/c mice. The control group received 20 µL of Alexa Fluor 488-labeled bovine serum albumin (Alexa488-BSA). After 15 minutes, the animals were sacrificed by CO2 inhalation and the piece of dorsal skin on the site of injection was carefully dissected, frozen in OCT (Optimal cutting temperature) and sectioned 5 µm thick in the cryostat (Leica, CM1510). The sections were then fixed in 3.7% formaldehyde for 10 minutes at room temperature. Nonspecific staining was blocked by incubating the sections for 1 hour at room temperature with PBS containing 1% triton X-100, 5% normal goat serum, 1% BSA, 0.5% glycine and 0.5% fish skin gelatin. Then, the sections were incubated with donkey anti-rat CD-31 polyclonal antibody (BD Bioscience, USA), at a 1∶40 dilution, or goat anti-rabbit type IV collagen polyclonal antibody (Chemicon, USA), at a 1∶40 dilution, for 18 hours at 4°C. The sections were washed with PBS and incubated with TRITC-labeled (Tetramethyl Rhodamine Isothiocyanate) goat anti-rat IgG (Jackson ImmunoResearch, USA), at a 1∶500 dilution, or TRITC goat anti-rabbit IgG (Jackson ImmunoResearch, USA), at a 1∶100 dilution, for 2 hours at room temperature. The distribution of toxins in the tissues was analyzed searching in at least 10 different fields of each section. Alternately, the distribution of jararhagin was analyzed 45 minutes after its injection under the same experimental condition. The sections were examined with a Confocal Microscope LSM 510 Meta (Zeiss, Germany).
In order to investigate the mechanisms involved in the hemorrhage induced by SVMPs, we initially compared the pathological alterations induced by jararhagin, a highly hemorrhagic P-III SVMP and BnP1, a weakly hemorrhagic P-I SVMP, using the mouse skin as experimental model. Tissues were inspected 15 minutes after injection of toxins in order to evaluate the first events involved in the hemorrhagic lesions. At this period, we focused on the direct action of venom toxins, avoiding interference of secondary effects of endogenous components released by the local reaction.
After 15 minutes, macroscopic analysis of the skin injected with doses adjusted at the same molar basis (10 µg jararhagin and 5 µg BnP1), revealed that jararhagin induced intense hemorrhage, whereas only a small hemorrhagic spot at the site of the injection was observed in the samples injected with BnP1 or PBS, used as injection control (Fig. 1A). Only high doses of BnP1 (50 µg) was able to induce hemorrhage, but less intense than jararhagin (Fig. 1A). Morphological analysis under light microscopy showed drastic hemorrhage in the hypodermis and also in the skeletal muscle adjacent to hypodermis in mice injected with jararhagin (Fig. 1B). The equivalent dose of BnP1 induced an enlargement of skin thickness probably due to its edema-forming activity but did not induce hemorrhagic alterations. When a hemorrhagic doses of BnP1 (50 µg) was injected, edema was persistent and only sparse spots of hemorrhage were detected in the hypodermis (Fig. 1B). We also evaluated the action of jararhagin and BnP1 on dermal-epidermal junctions by staining with antibodies anti-laminin and anti-β4 integrin and no alteration of these structures were observed after injection of toxins (data not shown). These results confirm the differences in hemorrhagic activity of jararhagin and BnP1 and show that most of the hemorrhagic incidence occurs in the hypodermis.
Next, we investigated the integrity of extracellular matrix components, mainly collagens and laminin, after injection of hemorrhagic doses of jararhagin (10 µg) or BnP1 (50 µg). After 15 minutes, the control skin injected with PBS, showed a dense network of collagen fibers stained by picrossirus. The bundles of collagen fibers were closely packed characterizing a dense connective tissue (Fig. 2 A, B). In contrast, mice injected with jararhagin showed a clear loosening of the bundles of collagen fibers in the dermis (Fig. 2C). In the hypodermis, where the hemorrhagic lesion occurs, only a few weakly stained fibers were observed, indicating a massive degradation of fibrillar collagen (Fig. 2D). BnP1 induced only a discrete disorganization of collagen fibers throughout the dermis (Fig. 2E) and the hypodermis (Fig. 2F).
The effect of toxins on the distribution of type IV collagen and laminin at the basement membrane was then evaluated by immunofluorescence. In the control skin, type IV collagen (Fig. 3A) and laminin (Fig. 3B) were observed as linear and continuous lines surrounding small blood vessels and in the basement membrane of skeletal muscle cells. In contrast, jararhagin induced a remarkable alteration in the immunostaining of type IV collagen in the basement membrane of blood vessels of the hypodermis, where only traces of type IV collagen deposition were detected. In addition, jararhagin also promoted a notable reduction of type IV collagen immunostaining in skeletal muscle basement membrane (Fig. 3A). After injection of 50 µg of BnP1, only a slight alteration in the immunostaining of type IV collagen was observed. The immunoreaction in the basement membrane of blood vessels and skeletal muscle was more diffuse than the pattern observed in the control, with the presence of some spots. However, the alterations induced by BnP1 are not comparable to the extensive disruption induced by jararhagin on type IV collagen, whose immunoreaction was practically abolished in blood vessels and skeletal muscle basement membrane (Fig. 3A). The effect of jararhagin on laminin distribution was less intense than that observed on type IV collagen. The presence of laminin was detected in tissues injected with jararhagin, but its distribution on basement membrane was not as homogeneous as observed in control tissues suggesting punctual disruptions of basement membrane integrity, probably as a result of collagen degradation. Similar effects were induced by BnP1 (Fig. 3B).
The next step was to analyze the distribution of SVMPs in the skin tissue. After 15 minutes of injection, when the hemorrhagic lesion has already been set, jararhagin was located close to small blood vessels stained by anti-CD31, in the hypodermis region (Fig. 4B). It is interesting to note that co-localization with CD-31 was not observed, suggesting the accumulation of the toxin near the blood vessels. In contrast, after injection of BnP1 (Fig. 4C), only a weak and diffuse fluorescence was observed, slightly higher than in control tissues (Fig. 4A). Similar deposition of the toxins was observed in the skeletal muscle adjacent to the hypodermis. High fluorescence was observed after jararhagin injection (Fig. 4E) in the basement membrane of skeletal muscle and capillaries suggesting its accumulation in these areas. After BnP1 injection (Fig. 4F), only a weak fluorescence was observed. No fluorescence was detected in control tissues (Fig. 4D). Similar pattern of toxin distribution was detected up to 45 minutes after jararhagin injection in areas adjacent to the main focus of the injection. This toxin was detected around hypodermis blood vessels (Fig. 5A), and close to the capillaries in the skeletal muscle (Fig. 5B).
In order to verify the binding of hemorrhagic toxins to the basement membrane, we carried out a double-staining protocol using Alexa488-labeled SVMPs and type IV collagen antibody. According to figure 6, jararhagin showed co-localization with type IV collagen in the basement membrane of venules and capillaries, 15 minutes after injection. Contrarily, no co-localization with type IV collagen was observed in mouse skin injected with BnP1, which showed a similar staining pattern to control samples (Fig. 6). These results confirm the particular binding of hemorrhagic toxins to basement membrane components, explaining their accumulation near blood vessels.
We next addressed the role of non-catalytic domains of SVMPs in the distribution of jararhagin on tissues. For that, mice were injected with Alexa488- labeled jararhagin-C, which consists of jararhagin disintegrin-like and cystein-rich domains, and its distribution was observed in mouse skin. The distribution of jararhagin-C in skin was the same for jararhagin: jararhagin-C was detected close to the CD-31 marker around blood vessels in the hypodermis (Fig. 7A), and close to capillaries in the skeletal muscle (Fig. 7B), and co-localized with basement membrane type IV collagen in the hypodermis venules (Fig. 7C). These results strongly suggest that the non-catalytic domains are determinant to hemorrhagic activity of SVMPs from P-III class, locating the catalytic site specifically to the microvascular wall.
In this work, we unveil an important step for understanding the mechanisms involved in the expressive hemorrhage induced by SVMPs, by comparing the in vivo degradation of extracellular matrix proteins and the tissue distribution of jararhagin, a highly hemorrhagic P-III SVMP, and BnP1, a weakly hemorrhagic P-I SVMP, using the mouse skin as experimental model. This comparison revealed that tissue localization and in vivo degradation of collagens are key events in SVMPs induced hemorrhage.
Jararhagin induced a massive degradation of fibrillar collagen in the hypodermis, where the hemorrhagic lesion was concentrated. In the vascular basement membrane, type IV collagen was the major substrate for jararhagin. In contrast, BnP1 was not able to efficiently degrade these substrates or to induce hemorrhage. Instead, a remarkable edema and dermal alteration were observed, consistent with the dermonecrotic activity already described for BaP1, a SVMP class P-I isolated from B. asper venom [29], . Considering that the macromolecular organization and the biomechanical stability of basement membrane are mainly determined by the type IV collagen network [31], its cleavage would alter the structural stabilization of the other related basement membrane components. Consistent with that, we observed that jararhagin induced only slight alterations in laminin distribution, suggesting that its epitopes are conserved after treatment with the toxin. This characteristic is not restricted to snake venom pathology. Selective degradation of type IV collagen has been associated with other pathologies involving toxic metalloproteinases. The metalloproteinase from Vibrio vulnificus (VVP) is a major determinant for skin lesions of this microorganism, which also causes hypodermic hemorrhage [32].
According to the literature, both P-I and P-III SVMPs are similarly able to hydrolyze extracellular matrix components in vitro, such as matrigel [33] and isolated components, such as type IV collagen, laminin and fibronectin [20], [33]–[37]. Escalante et al. [17] showed that jararhagin and BaP1 had a similar proteolytic activity on matrigel with a slightly different cleavage pattern, since BaP1, exerted a limited proteolysis of both laminin and nidogen, whereas jararhagin predominantly degraded nidogen. However, the hydrolysis of extracellular matrix components in vitro occurs only after long incubation periods, suggesting that distinct mechanisms are involved in the basement membrane digestion in vivo. These authors also analyzed the immunostaining of laminin, nidogen, type IV collagen and the endothelial cell marker VEGFR-2 (vascular endothelial cell growth factor receptor 2) in mouse gastrocnemius muscle injected with hemorrhagic doses of jararhagin and BaP1, observing reduction in the number of capillary vessels and a similar pattern of immunostaining for the basement membrane components laminin, nidogen and type IV collagen in muscular fibers after injection of BaP1 or jararhagin, showing a disorganization of extracellular matrix [17]. Although they showed the first evidence of the catalytic action of SVMPs in vivo and morphological alterations in muscular tissues, the authors failed to detect any difference between weakly and highly hemorrhagic SVMPs.
A parameter not yet explored in the literature consisted of eventual differences in the distribution of toxins in the damaged tissues. In this study, jararhagin, but not BnP1, concentrated in the vicinity of small venules and in capillaries of skeletal muscle. This effect was correlated to the non-catalytic domains of jararhagin. In vitro, SVMPs bind to extracellular matrix proteins, such as type I collagen [15], [38], [39], [40], type IV collagen [40], collagen XII and XIV and the matrilins 1, 3 and 4 [16]. The high affinity for these extracellular matrix proteins could contribute for the accumulation of the toxin in the damaged tissue, enhancing the catalytic action of SVMPs towards basement membrane components.
A fine correlation between collagen binding and hemorrhagic activity has been shown by our group. An anti-SVMP monoclonal antibody neutralizes hemorrhagic activity and collagen binding of jararhagin without interfering with its catalytic activity [25]. Recently, it was shown that jararhagin binds with high affinity to type I collagen and type IV collagen, whereas berythrativase, a non-hemorrhagic P-III SVMP isolated from B. erythromelas venom, failed to bind to these substrates [40]. Molecular modeling of the putative epitopes binding to this monoclonal antibody pinpointed a motif present in the hemorrhagic toxin jararhagin and absent in the pro-coagulant enzyme berythractivase, located at the Da-subdomain of disintegrin-like domain [40]. However, it is important to consider that a collagen-binding motif was also detected in the cysteine-rich domain of Atrolysin-A, a hemorrhagic P-III SVMP [15].
One aspect still unclear is the apparent contradiction between the results of collagen hydrolysis in vitro and in vivo. In vitro, most SVMPs (class P-I or P-III) hydrolyse type IV collagen but not fibrillar collagens [20]. Here, an almost complete disassembly of hypodermal fibrillar collagen was observed in tissues treated with jararhagin, but not BnP1. Escalante and co-workers [41] also observed collagen hydrolysis in vivo. Several fragments of collagens were detected in the exudates of muscle injected with BaP1, a class P-I SVMP. Most of the fragments corresponded to non-fibrillar collagen. However, a fragment corresponding to the fibrillar collagen V was also identified. According to our results, class P-I SVMPs induced minor alterations to hypodermal fibrillar collagen. Thus, it is possible to predict that skin homogenates of P-III SVMP lesions would contain more degradation fragments of fibrillar collagens. A possible explanation for the differences observed in vivo could be the disorganization of fibrils due to the toxin binding allowing the degradation of fibrillar collagen on hypodermis. However, data explaining why this occurs only in vivo is still lacking. Other alternative would be that the injection of SVMPs would induce tissue secretion of MMPs, able to digest fibrillar collagen. However, this hypothesis is not consistent with the fast onset of the reaction. Also, P-I SVMPs are very efficient to induce over-expression of MMPs [29], [42] and they appear to be much less efficient to hydrolyze collagen in vivo.
Our results confirm previous suppositions that the non-catalytic domains play a crucial role in the expression of hemorrhagic activity of P-III SVMPs, implying that non-enzymatic mechanisms are also involved in bleeding. The hypervariable region of the cysteine-rich domain is attributed to the binding of SVMPs to a series of substrates containing von Willebrand factor A domains, allowing the catalysis of a specific substrate region [16]. Here we suggest that SVMPs may present additional functional motifs related to their binding to collagens. The cysteine-rich domain exosite would be essential for the enzymatic selectivity of the SVMPs while a disintegrin-like domain collagen binding motif would be responsible for high affinity binding to collagens and tissue concentration of the toxin.
The data presented herein are particularly important to understand the mechanisms involved in the onset of hemorrhage and could contribute to the rational of alternative treatments for snakebites victims. Since accumulation of SVMPs at the site of the bite allows hemorrhage enhancing the local venom effects, the local administration of metalloproteinases inhibitors could represent an interesting approach in order to improve the neutralization of toxins responsible for the local damage. Indeed, it has already been shown that the local injection of batimastat, a peptidomimetic matrix metalloproteinase inhibitor, totally neutralized the proteolytic, hemorrhagic and dermonecrotic effects induced by Bothrops asper venom [43]. Moreover, it has recently been shown that local administration of tetracycline prevented the dermonecrosis induced by Loxosceles spider venom. In addition to its antimicrobial properties, tetracycline was also able to inhibit MMPs, which are important for the progression of dermonecrotic lesions [44].
In summary, we showed that the strong hemorrhage induced by class P-III hemorrhagic SVMPs is related to their accumulation at basement membrane, reaching enzyme concentrations sufficient for its rapid degradation. This mechanism may serve as a rational for the design of alternatives in which local administration of metalloproteinase inhibitors may complement antivenoms in the neutralization of local tissue damage.
|
10.1371/journal.ppat.1001194 | Potentiation of Epithelial Innate Host Responses by Intercellular Communication | The epithelium efficiently attracts immune cells upon infection despite the low number of pathogenic microbes and moderate levels of secreted chemokines per cell. Here we examined whether horizontal intercellular communication between cells may contribute to a coordinated response of the epithelium. Listeria monocytogenes infection, transfection, and microinjection of individual cells within a polarized intestinal epithelial cell layer were performed and activation was determined at the single cell level by fluorescence microscopy and flow cytometry. Surprisingly, chemokine production after L. monocytogenes infection was primarily observed in non-infected epithelial cells despite invasion-dependent cell activation. Whereas horizontal communication was independent of gap junction formation, cytokine secretion, ion fluxes, or nitric oxide synthesis, NADPH oxidase (Nox) 4-dependent oxygen radical formation was required and sufficient to induce indirect epithelial cell activation. This is the first report to describe epithelial cell-cell communication in response to innate immune activation. Epithelial communication facilitates a coordinated infectious host defence at the very early stage of microbial infection.
| All body surfaces are covered by a single layer of epithelial cells. Epithelial cells form a physical barrier to separate the underlying sterile tissue from the environment. In addition, epithelial cells actively sense bacterial and viral infection. The recognition of pathogenic microorganisms results in cell stimulation and the secretion of soluble mediators that attract professional immune cells to the site of infection. This first line host defence works very efficiently despite the often low number of pathogens and the limited amount of mediators secreted per epithelial cell. We therefore investigated whether infection of one individual epithelial cell would result in activation of other, non-infected cells within a confluent epithelial monolayer resulting in a more substantial host response. Indeed, using the model of the gut pathogen Listeria monocytogenes and monitoring infection and epithelial activation at a single cell level, we can clearly show that the epithelial response is mainly mediated by non-infected cells. Also, we identify oxygen radicals as potential mediators to facilitate horizontal epithelial communication upon immune stimulation. Our results thus provide a novel concept of a coordinated epithelial host response upon microbial infection facilitated by horizontal epithelial communication.
| Intestinal epithelial cells line the enteric mucosal surface and provide a physical barrier to maintain the integrity of this vulnerable body surface and prevent invasive infection by luminal microorganisms. Like professional immune cells, intestinal epithelial cells express receptors of the innate immune system such as Toll-like receptors (TLR) or nuclear oligomerization domain (NOD)-like receptors (NLR) [1], [2]. Recognition of microbial structures leads to epithelial production of antimicrobial effector molecules and proinflammatory chemoattractive mediators. Thus, it facilitates an active role in the initiation of the mucosal host response [3], [4], [5]. The recruitment of professional immune cells to the site of infection occurs within hours and provides a highly efficient dynamic mechanism of the epithelial host defence. It remains unclear, however, how low number of pathogenic microorganisms as well as the limited spectrum and only moderate amount of chemokine secretion per epithelial cell facilitates stimulation of an effective host defence. We therefore hypothesized that a horizontal intercellular communication between intestinal epithelial cells might help to induce a coordinated epithelial response towards infectious challenge and thereby to amplify the epithelial innate host defence.
Listeria monocytogenes is an important human pathogen that causes meningitis, sepsis, and abortion in susceptible individuals. It is acquired with food such as unpasteurized milk and cheese and enters the body following penetration through the intestinal epithelial barrier. The microbial pathogenesis and the bacteria-host cell interaction of this facultative intracellular bacterium has been studied for many years [6]. L. monocytogenes induces its own internalization and subsequently lyses the endosomal membrane of its host cell by the secretion of listeriolysin O (LLO) and phospholipases, thus gaining access to the cytosolic space. Here, Listeria upregulates polar expression of ActA that recruits and polymerizes host actin filaments resulting in propulsive locomotion. Together with LLO and the phospholipases this allows to enter neighbouring cells and to spread within the epithelial cell layer. Importantly, recognition of Listeria by the epithelial innate immune system only occurs after internalization and lysis of the endosomal membrane through cytosolic innate immune receptors [7], [8], [9], [10]. Since infection of individual cells can be traced using reporter gene technology, L. monocytogenes provides an excellent model to study cellular responses in respect to immune recognition at the single cell level.
In the present study, we analyzed innate immune recognition and epithelial responses at the single cell level using the model of Listeria infection of polarized intestinal epithelial cells in addition to transfection and microinjection. We present the surprising finding that non-infected epithelial cells were the main source of chemokine secretion in response to bacterial challenge. We identify oxygen radical species produced by NADPH oxidase (Nox) 4 in response to cytosolic bacteria to facilitate horizontal intercellular communication and chemokine production by non-infected cells. These results provide the first experimental evidence for a yet unknown mechanism of intercellular communication between epithelial cells in response to innate immune stimulation and thus significantly broaden our understanding of mucosal innate host defence.
Infection of a confluent monolayer of intestinal epithelial m-ICcl2 cells with wild-type (wt) L. monocytogenes induced rapid cellular activation illustrated by secretion of the proinflammatory chemokine Cxcl-2 (Fig. 1A). Strong epithelial activation was only observed using wt Listeria able to reach the cytosolic space (Fig. 1B) facilitating recognition by cytoplasmic innate immune receptor molecules (Fig. 1C) [7], [8], [9], [10]. Bacterial mutants unable to lyse the endosomal membrane such as isogenic hly or hly/plcA/plcB triple mutants as well as heat inactivated bacteria exhibited a significantly reduced or even absent epithelial activation (Fig. 1B and D). Of note, lack of hly or hly, plcA, and plcB expression did not affect bacterial invasion or intracellular viability (Fig. S1A). Endosomal lysis-dependent stimulation of L. monocytogenes infected epithelial cells was also observed using flow cytometry. A time-dependent increase of the number of Cxcl-2+ and Cxcl-5+ epithelial cells was detected after infection with wt Listeria (Fig. 1E). In contrast, a strongly reduced number of epithelial cells stained positive for Cxcl-2 after infection with hly mutant Listeria (Fig. 1F). In accordance with the published literature, internalization-dependent activation was observed in epithelial cells, but not in macrophages (Fig. S1B). These results suggested that activation of epithelial cells occurred primarily in directly Listeria-infected cells.
To monitor Listeria infection and cellular activation simultaneously at the single cell level, bacteria transformed with a vector expressing green fluorescence protein (GFP) either under control of the inducible actA promoter [11] or the constitutive sod promoter [12] were used for subsequent experiments (for details see Table 1). Surprisingly, flow cytometry revealed that the vast majority of Cxcl-2+ epithelial cells (95%) were Listeria-negative. In addition, only a minor fraction of GFP-positive, Listeria-infected epithelial cells exhibited MIP-2 synthesis (Fig. 2A). Similar results were obtained using biotinylated Listeria (Fig. S2A). These results were confirmed by immunohistological staining. Cxcl-2 and Cxcl-5 synthesis was not restricted to GFP+ Listeria-infected cells but, in fact, predominantly detected in neighbouring non-infected epithelial cells (Fig. 2B and Fig S2B). Also flow cytometric cell sorting and quantitative RT-PCR analysis strongly supported this unexpected result. A marked upregulation of Cxcl-2 and Cxcl-5 mRNA expression was detected in GFPlow expressing (Listeria-negative, Fig. S2C) epithelial cells despite the absence of detectable Listeria DNA (Fig. 2C). Thus, although epithelial activation requires lysis of the endosomal membrane and contact with cytosolic innate immune receptors, a transcriptional cellular response was mainly observed in non-infected cells.
Several mechanisms might account for the observed activation of Listeria-negative epithelial cells. Activated epithelial cells might only appear to be Listeria-negative due to secondary bacterial escape facilitated by propulsion through ActA-induced actin polymerization in the cytosol and subsequent invasion of the neighbouring cell. Cells primarily infected but secondarily left by lateral spread might thereby appear Listeria-negative but in fact would have been previously in contact with cytosolic bacteria (and thus were, in fact, directly activated). To avoid lateral cell-to-cell spread and restrict intraepithelial bacteria to apically infected cells, a Listeria actA mutant strain was employed. ActA-deficient Listeria exhibited a moderately reduced epithelial invasion (Fig. S3A), a lower percentage of infected epithelial cells, and an enhanced number of bacteria per cell (Fig. 3A and 3B). Nevertheless, high numbers of Cxcl-2 producing epithelial cells (Fig. 3B) and a strong chemokine secretion (Fig. 3C) was observed. Also, the number of activated, Cxcl-2 producing epithelial cells remained significantly higher than the number of Listeria-infected cells reaching approximately 10-fold excess of Cxcl-2+ cells (Fig. 3B and Fig. S3B). Thus, indirect activation of epithelial cells was not due to escape from previously infected cells by ActA-driven secondary lateral spread.
Epithelial cell stimulation could also be induced by bacteria either attached to the plasma membrane or remaining intraendosomal and membrane enclosed. To exclude a significant role of attached or intraendosomal Listeria, Cxcl-2+ and GFP+ epithelial cells were quantified after infection with Listeria expressing GFP either constitutively under control of the superoxide dismutase promoter (Psod-gfp) or inducible under the control of the actA promoter (PactA-gfp) (Table 1). Whereas Psod-gfp carrying Listeria exhibited strong reporter expression after growth in bacterial culture medium, only a moderate fluorescence was detected in PactA-gfp –positive bacteria (Fig. S3C). In contrast, strong GFP expression was noted in PactA-gfp Listeria isolated from infected epithelial cells (Fig. S3D). Flow cytometric detection of infected epithelial cells was observed after wt, but not hly mutant PactA-gfp Listeria illustrating the endosomal lysis-dependent induction of the actA promoter-driven GFP reporter gene expression (Fig. S3E). Infection with Psod-gfp or PactA-gfp carrying wt Listeria resulted in a significant number of Listeria-infected epithelial cells. Importantly, a higher number of Cxcl-2+ activated cells as compared to Listeria-infected cells was observed by flow cytometry after infection with both reporter constructs and epithelial Cxcl-2 synthesis was similarly noted in Listeria-negative epithelial cells (Fig. 3D). These results suggest that indirect epithelial activation was not a result of attached or intraendosomal bacteria [13].
Finally, the activation of Listeria-negative epithelial cells might be due to the stimulatory effect of secreted bacterial molecules, such as the cytolytic listeriolysin O (LLO) [14], [15]. Therefore, the membrane damaging as well as the stimulatory effect of recombinant listeriolysin (rLLO) on red blood cells (RBC) and epithelial m-ICcl2 cells was analysed. High concentrations of rLLO induced significant hemoglobin and detectable lactate dehydrogenase (LDH) release by RBCs and epithelial m-ICcl2 cells, respectively (Fig. S3F and Fig. S3G). Quantitation of epithelial cell activation in response to rLLO, however, revealed an only minor response as compared to epithelial Cxcl-2 secretion after viable wt L. monocytogenes infection (Fig. 3E). Similarly, no significant Cxcl-2 secretion by epithelial cells was noted in response to bacteria-free culture supernatant derived from Listeria cultures with bacterial counts precisely corresponding to the infection model described above (Fig. 3F). Yet, culture supernatants derived from wild-type or ActA-deficient bacteria exhibited significant hemolytic activity, in contrast to supernatant from hly-deficient Listeria (Fig. S3H). No significant membrane damage was noted after infection of intestinal epithelial cells with wt, actA, or hly mutant Listeria (Fig. S3I). Although a supportive effect of released bacterial factors cannot be excluded, these results suggest that bacterial mediators do not play a major role in the observed indirect epithelial activation. Thus, neither basolateral cell-to-cell spread nor membrane attachment, or the secretion of LLO in the cell culture supernatant appear to be responsible for indirect epithelial cell activation after L. monocytogenes infection. This suggests the presence of a previously unrecognized mechanism of epithelial intercellular communication in response to bacterial infection.
To examine whether indirect epithelial stimulation by horizontal cell-to-cell communication might be a general effect of transcriptional activation of intestinal epithelial cells, a bicistronic expression vector encoding the NF-κB subunit RelA/p65 together with GFP under the control of a constitutive cytomegalovirus (CMV) promoter was employed (Fig. S4A). Transient overexpression of RelA/p65 alone or bicistronic expression of RelA/p65 and GFP readily induced epithelial activation as illustrated by NF-κB reporter gene upregulation (Fig. S4B) and enhanced chemokine secretion (Fig. S4C). Although RelA/p65-mediated Cxcl-2 production exhibited a slower kinetic as compared to following Listeria infection, a significant number of Cxcl-2+ cells was detected. Of note, RelA/p65-mediated cellular activation was restricted to GFP+, i.e. directly activated epithelial cells (Fig. S4D). Cxcl-2 production by GFP+ cells increased strongly (0.1 versus 2.4%), whereas the number of Cxcl-2+ cells in the GFP- population remained virtually unchanged (0.8% versus 1.2%). In addition, the number of Cxcl-2+ epithelial cells did not exceed the number of transfected GFP+ cells at any time (Fig. S4E). Thus, epithelial activation per se does not induce indirect cell activation by horizontal intercellular communication. Indirect epithelial activation appears rather to be induced by innate immune signal transduction upstream of transcription factor activation.
Next we investigated the mechanism underlying horizontal cell-to-cell communication and coordinated epithelial chemokine upregulation in response to Listeria infection. Functional gap junctional transport was examined by microinjection of transferable Lucifer Yellow together with non-transferable high molecular weight dextran. Fluorescence imaging visualized transport of Lucifer Yellow from the microinjected cell to the surrounding neighbouring cells. Addition of inhibitors of gap junctional transport, effectively reduced lateral diffusion of Lucifer Yellow after microinjection (Fig. 4A and B). Inhibition of gap junctional intercellular communication, however, did not decrease the number of activated epithelial cells after Listeria infection as illustrated by the unaltered high ratio of activated (Cxcl-2+) to infected (GFP+) epithelial cells measured by flow cytometry (Fig. 4C). Although these results do not completely rule out transfer of very small signaling molecules by gap junctional transport channels, they do not support a major role in the process of horizontal communication.
Similarly, the potential role of a secreted protein messenger was examined. Intestinal epithelial m-ICcl2 cells were exposed to brefeldin A (BFA), an effective inhibitor of the secretion of newly synthesized proteins (Fig. S4F), prior and after infection with actA mutant L. monocytogenes. The number of Listeria-induced Cxcl-2+ cells, however, was not altered irrespective whether BFA was administered 30 min prior or 60 min after infection (Fig. 4D). Second, cell culture medium was obtained 10, 20, 30, 40, or 60 min after Listeria infection, centrifuged to remove bacteria, and immediately transferred to naïve uninfected epithelial cells. Yet no epithelial activation was observed after exposure to conditioned culture supernatant despite significant Cxcl-2 synthesis detected in the Listeria infected cell population (Fig. S4G). Of note, factors released by Listeria-infected cells might be unstable or immediately bound to neighbouring cells preventing their efficient release in the conditioned cell culture supernatant.
Finally, widely used pharmacological inhibitors of prostaglandin synthesis and known intestinal epithelial ion channels were employed. Indomethacin, an inhibitor of cyclooxygenase isoenzymes (COX1, COX2) involved in prostaglandin synthesis, thapsigargin, an inhibitor of the endoplasmatic Ca2+ATPase, CFTR II, a selective apical Cl− ion channel inhibitor, and bumetanide, an inhibitor of a basolateral epithelial Na+K+Cl− cotransporter had no significant influence on the number of activated epithelial cells after Listeria infection illustrated as ratio of activated (Cxcl-2+) to infected (GFP+) cells (Fig. S4H). These results do not identify a significant role of gap junctional transport, secreted protein or prostaglandin mediators, or ion fluxes in the observed indirect activation of epithelial cells after L. monocytogenes infection.
Since unstable and highly reactive host-derived factors were not excluded by the previous experiments, a possible involvement of oxygen or nitrogen radicals in horizontal epithelial cell-cell communication was subsequently evaluated. Expression of members of two enzyme families, NADPH oxidases and nitric oxide synthase (NOS), has been described in epithelial cells [16]. Indeed, addition of the NADPH oxidase inhibitor diphenylene iodonium (DPI) resulted in a significant reduction of Listeria-induced epithelial activation (Fig. 5A). DPI did not reduce Listeria survival in epithelial cells (Fig. S5B) and had no effect on LPS or PMA-induced epithelial activation (Fig. S5A). In contrast to DPI, the NOS inhibitor N (G)-nitro-L- arginine methyl ester (L-NAME) did not influence the number of activated epithelial cells (Fig. 5B). In accordance with an inhibitory effect of DPI, synthesis of reactive oxygen intermediates (ROI) after Listeria infection was observed (Fig. 5C). ROI was detected in focal areas of confluent epithelial cells surrounding Listeria-positive, infected cells in accordance with local production and lateral spread of ROI as early as 10 min after infection (Fig. 5D).
Innate immune receptor stimulation by Listeria infection of epithelial cells resulted in rapid activation of the mitogen-activated protein (MAP) kinase Erk in a ROI-dependent manner (Fig. 5E). Whereas impairment of the MAP kinase Erk had no significant effect on ROI production (Fig. 5F), Listeria-induced Cxcl-2 synthesis by intestinal epithelial cells was completely abrogated by Erk inhibition and partially also dependent on the MAP kinases p38 and JNK (Fig. 5G). Of note, Erk inhibition did not affect bacterial invasion and the viability of intracellular Listeria (Fig. S5B). Finally, exposure of epithelial cells to cumene hydroperoxide, a ROI liberating organic agent within the cell culture medium or by microinjection induced Cxcl-2 synthesis in neighbouring cells similar to L. monocytogenes infection (Fig. 5H). Thus, Listeria-infection induces significant epithelial ROI synthesis, which in turn mediates MAP kinase Erk activation and downstream Cxcl-2 production.
Oxygen radical synthesis is performed by an oligomeric protein complex involving a cell type-specific NADPH-oxidase (Nox) protein. Only significant expression of the Nox4 isoform was detected in primary small intestinal epithelial cells (Fig. 6A). Nox4 synthesis was restricted to intestinal epithelial cells as demonstrated by immunostaining with a paranuclear expression pattern in accordance with a previous report (Fig. 6B) [16]. Importantly, downregulation of Nox4 expression in epithelial cells by siRNA interference significantly reduced Cxcl-2 secretion (Fig. 6C) and ROI production upon Listeria-infection (Fig. 6D). In contrast, downregulation of Nox4 expression did not alter LPS- or PMA-induced chemokine secretion (Fig. S6). Thus, ROI production by Nox4 appears to be both necessary and sufficient to induce horizontal cell-cell communication in intestinal epithelial cells leading to chemokine secretion in neighbouring cells in response to Listeria infection.
Communication between individual cells is a fundamental feature of multicellular organisms. For instance, it mediates a coordinated reaction of muscle cell contraction, and allows neuronal signal transmission or endocrinological regulatory circuits. Cell-cell communication is also characteristic for the complex regulatory networks of the adaptive immune system. Cytokines bridge anatomical distances to coordinate and amplify the host response against pathogens. In the present study we investigated whether cell-cell communication between neighbouring cells might also contribute to innate immune activation within a confluent epithelial cell layer to coordinate the antimicrobial host defence at an early stage of the infection. Although inhibition of the overall epithelial responses by interference with the production of soluble mediators and gap junction integrity had previously been noted, the process of immune stimulation and cellular response upon bacterial infection has not been studied at the single cell level [17], [18], [19], [20], [21]. The present study therefore represents the first report to demonstrate epithelial horizontal cell-cell communication upon bacterial innate immune stimulation.
For three reasons, Listeria infection of confluent intestinal epithelial cells represents an ideal model to study epithelial cell-cell communication downstream of innate immune stimulation. First, similar to other pathogenic bacteria Listeria monocytogenes escapes from the endosomal vacuole and proliferates within the host cell cytosol [6]. Endosomal escape is associated with a dramatic change in bacterial gene expression. Although expressed at low levels also during in vitro culture, a very strong upregulation of the actin polymerizing protein ActA provides an excellent reporter for detection of cytosolic entry [22]. Second, bacteria lacking hly, plcA, or plcB mediating endosomal lysis only induce an only minor activation which might result from intraendosomal recognition or a so far unidentified minor mechanism of endosomal escape. Thus, in contrast to macrophages that recognize Listeria also at the plasma membrane, epithelial cell stimulation is mainly observed when bacteria reach the cytosol, facilitating contact with cytosolic innate immune receptors such as Ipaf, Nalp3, and Nod2 [9], [10]. This finding excludes innate immune recognition and receptor-mediated initiation of signal transduction in non-infected, Listeria-negative cells. Third, one amino acid exchange between the mouse and human E-cadherin causes a strongly reduced infection rate in murine epithelial cells [23], leaving most cells of a confluent cell layer uninfected and accessible to the analysis of indirect cellular activation. Using reporter gene technology, intracellular chemokine staining and flow cytometric analysis, we were able to demonstrate that the chemokine secretion in response to Listeria infection is mainly derived from uninfected, indirectly activated epithelial cells. Of note, the commonly used quantification of cytokine secretion in the cell culture supernatant or immunoblotting of total cell lysate proteins would not have disclosed this surprising finding.
Epithelial stimulation on the transcriptional level by p65/RelA overexpression did not result in detectable indirect cell activation. Several possible mechanisms of horizontal cell-cell communication downstream of innate immune receptor signaling were therefore considered. In response to microbial stimulation, epithelial cells produce chemokines, prostaglandins, and cause local alterations of ion concentrations by regulating transmembrane ion channel activity. Also, gap junctional intercellular communication (GJIC) represents a direct cytosolic connection and might be used to forward the information of innate immune recognition within the epithelial cell layer [17], [21]. Ca2+ fluxes via intercellular gap junctions have been shown to promote lung epithelial chemokine secretion [18] and intact gap junction formation has also been linked to innate immune stimulation and maintenance of the epithelial barrier [19]. On the other hand, connexin-26 hemichannel-mediated Ca2+ signaling has also been proposed to promote bacterial invasion and lateral spread [24]. Yet, neither protein secretion, nor ion channel activity or gap junction formation appeared to be involved in Listeria-induced indirect epithelial cell activation.
Instead, our results indicate an important role of reactive oxygen intermediates (ROI) in horizontal epithelial cell-cell communication. ROI represent reduction products of molecular oxygen such as the radical superoxide (•O2−) and hydroxyl (•OH), and the non-radical hydrogen-peroxide (H2O2). ROI production by professional phagocytes during oxidative burst provides significant bactericidal activity but synthesis is also observed in non-phagocytic cells [25]. ROI at subtoxic doses has been recognized as an important intracellular signal transducing molecule during the recent years [26], [27], [28], [29]. In accordance with our results ROI-induced activation of MAP kinase activity has been reported [30], [31], [32], [33]. In addition, an involvement of ROI in the cellular signaling leading to NF-κB activation [34], apoptosis [31], [35], epidermal growth factor receptor signaling [36], regulation of cellular proliferation [37], and antimicrobial peptide production [38], [39] has been described. ROI was also shown to prime Drosophila melanogaster hematogenic progenitor cells for differentiation [40] and to play an important role in the fruit fly's intestinal immunity [41]. Whereas the half-life of oxygen radical hydroxyl (•OH) is extremely short (10−9s) and the superoxide •O2− is membrane impermeable [40], H2O2 is able to diffuse to neighbouring cells and induce cellular activation. Indeed, a tissue gradient of H2O2 was shown to induce rapid recruitment of leukocytes into the wound margin following endothelial hypoxia [42], [43].
NADPH oxidase activation has previously been linked to innate immune mediated antimicrobial killing [25] as well as receptor signal transduction [44], [45], [46], [47], [48], [49], [50], [51]. Here we for the first time report Nox4 expression by intestinal epithelial cells and demonstrate Nox4-mediated ROI production in response to bacterial infection. Although enhanced Nox4 mRNA expression was shown to result in increased ROI production [52], the initiation of Nox4-dependent ROI production upon Listeria infection was noted as early as 5–10 minutes after bacterial challenge. This excludes a significant role of transcriptional regulation of Nox4 in our model. Whereas the prototypical NADPH oxidase of phagocytes, gp91phox (Nox2), requires cytosolic proteins such as p47phox to form a functional NADPH oxidase complex, Nox4 functions independent of cytosolic accessory proteins. Interestingly, Nox expression has previously been linked to innate immune receptor signaling: Nox4 activation was shown to be involved in TLR4-mediated NF-κB activation in human epithelial kidney cells and monocytes [47], [53]. In contrast, our results revealed activation of intestinal epithelial cells by Listeria infection in a Nod2-, Ipaf-, and Nalp3-dependent fashion which was followed by ROI production and subsequent MAP kinase signaling. Our data are therefore in accordance with previous reports on MAP kinase activation after Nox4-mediated ROI production [54], [55]. A future analysis of the local paracellular concentration of the different species of oxygen radicals might help to improve our understanding of the regulatory role of Nox4-mediated ROI production for epithelial cell-cell communication. Interestingly, reduced chemokine synthesis was noted in directly infected, Listeria-positive cells. These cells were also impaired to respond to secondary innate immune stimulation illustrating the immune evasive behaviour of L. monocytogenes (data not shown). Although the underlying mechanism is currently not resolved, high concentrations of ROI were previously associated with reduced susceptibility to immunostimulatory agents [56]. Yet other bacterial or host factors such as antioxidant enzymes might reduce local ROI concentrations and interfere with cellular activation and chemokine production in infected epithelial cells.
In conclusion, our data for the first time analyzed intestinal epithelial activation in response to bacterial infection at a single cell level. We could detect Nox4 expression by intestinal epithelial cells which facilitated rapid ROI production upon infection and paracrine activation of neighbouring cells (Fig. 7). Our findings thus identify horizontal cell-cell communication to allow a coordinated innate immune activation of the intestinal epithelium. The present work significantly broadens our knowledge on the complex processes that underlie mucosal innate immune stimulation and illustrates the specific role of epithelial cells for an efficient activation of the antimicrobial host defence.
Intracellular Cxcl-2 (MIP-2) and Cxcl-5 was detected using rabbit antibodies from Nordic Biosite (Täby, Sweden). The rabbit polyclonal anti-actin antiserum was from Sigma-Aldrich (Taufenkirchen, Germany). The rabbit-anti-mouse Nox4 antiserum was obtained by immunization with recombinant peptide. The rabbit anti-p-p44/42 (phospho-Erk) and the mouse anti-p44/42 (total-Erk) was from Cell Signaling Technology (Beverly, MA, USA). The rabbit anti-Listeria antibody was from Dunn Labortechnik GmbH (Asbach, Germany). Cy5-, Cy3-, HRPO-conjugated secondary antibodies were from Jackson ImmunoResearch (West Grove, PA, USA) and the Alexa Fluor (AF) 488-, and AF 555-conjugated donkey anti-rabbit IgG (H+L) was from Invitrogen (Molecular Probes). The MFP590- and MFP488-labelled phalloidin were purchased from MoBiTec GmbH (Goettingen, Germany). The Sulfo NHS-LC-Biotin was obtained from Pierce, Thermo Scientific (Rockford, IL, USA). Escherichia coli K12 D31m4 LPS was ordered from List Biological Laboratories (Campbell, CA, USA). Recombinant LLO (rLLO) was expressed and purified exactly as described before [57]. rLLO was applied to cells in a serial dilution with 0.05 µg/mL as highest concentration. Cxcl-2 was quantified using an ELISA from Nordic Biosite or R&D Systems (Quantikine, R&D Systems GmbH, Wiesbaden, Germany). The NF-κB reporter construct pBIIX-luciferase carrying two copies (2× NF-κB) of the κB sequences from the Igκ enhancer was provided by S. Ghosh (Yale University Medical School, New Haven, CT, USA). Luciferase activity was quantified with luciferin substrate (PJK GmbH, Kleinblittersdorf, Germany). The bicistronic RelA/p65 expression plasmid was cloned by removing the nef gene from a pCG-nef-IRES-GFP expression plasmid (provided by J. Muench, Institute of Virology, University Clinic of Ulm, Germany) by digestion with the restriction enzymes XbaI and MluI (Fermentas, St Leon-Rot, Germany) and replacing it in frame with the p65 encoding gene amplified from a p65 expression plasmid (obtained from by U. Pahl, University Clinic, Freiburg, Germany) using the forward: 5′-ACC TCT AGA CCA TGG ACG ATC TGT TTC C-3′ and reverse: 5′-ACG ACG CGT GCA CCT TAG GAG CTG ATC TGA-3′ primers and digested with XbaI/MluI prior to ligation. Plasmid DNA for transfection was prepared using the endotoxin-free plasmid kit from Qiagen (Hilden, Germany). Targeted siRNA probes (Tlr2, Rip2, Nox4, Card12, Cias1, control siRNA,) were from Qiagen (Hilden, Germany), the Card15 siRNA was from Santa Cruz (Heidelberg, Germany). Lipofectamin 2000 (Invitrogen, Carlsbad, CA, USA) and INTERFERin (Polyplus Transfection, New York, NY, USA) were used for plasmid and siRNA transfection, respectively. The pharmacological inhibitors and radical donors oleamide, carbonoxolone, α-glycerrhetinic acid, brefeldin A, thapsigargin, CFTR inhibitor II (CFTR II.), indomethacin, bumetanide, N-(G)-nitro-L- arginine methyl ester (L-NAME), UO-126, hydrogen peroxide (H2O2) and cumene hydroperoxide were purchased from Sigma Aldrich. The p38 inhibitor 3-O-Acetyl-beta-boswellic acid and the L-stereoisomer JNK inhibitor 1 were from Enzo Life Sciences (Lörrach, Germany), and diphenylene iodonium (DPI) from Cayman Chemical (Hamburg, Germany). Defibrinated sheep red blood cells (SRBC) were purchased from Oxoid (Basingstoke, UK). The LDH Cytotoxicity Assay Kit was from Cayman Chemical (Hamburg, Germany). Colorimetric (ELISA, LDH), luminescent (luciferase) and fluorescent (ROI) measurements were carried out using a Victor3 Multilabel Plate Reader (Perkin Elmer, Waltham, MA, USA). Cell culture reagents were purchased from Invitrogen. All other reagents were obtained from Sigma Aldrich (Taufkirchen, Germany) if not stated otherwise.
The m-ICcl2 small intestinal epithelial cell line has previously been described [58]. Cells were cultured in a modified, hormonally defined medium with DMEM and F12 (vol 1∶1) supplemented with 5% FCS, 2% glucose, 20 mM Hepes, 2 mM glutamin, 5 µg/mL insulin, 50 nM dexamethasone, 60 nM sodium selenite, 10 ng/mL epithelial growth factor, 5 µg/mL transferrin, and 1 nM 3,3′,5-triiodo-L-thyronine sodium salt. Cell passages 42–70 were used. Cells were grown at 37°C in a 5% CO2 atmosphere on collagen-coated cell culture plates or chambers to reach a polarized, confluent monolayer. Rat tail collagen was ordered from Institut Jacques Boy (Reims, France). Specific targeted or control siRNA was transfected at a final concentration of 10 nM 36 hours prior to functional analysis. Stimulation with lipopolysaccharide (LPS) was performed at a final concentration of 10 ng/mL.
Listeria monocytogenes EGD wild-type (wt), actA, hly deletion mutant strains and the hly/plcA/plcB triple mutant strain are described in Table 1. Fluorescent bacteria were generated by transformation [59] with GFP expression vectors under the control of the actA or sod promoter (PactA-gfp, Psod-gfp; Table 1) Bacteria were routinely grown in Brain Heart Infusion (BHI) broth, supplemented with antibiotics when required. Overnight cultures were diluted 1∶50, grown to middle logaritmic phase (OD600) with mild agitation at 37°C, washed, and added in cell culture medium at the multiplicity of infection (m.o.i.) of 100∶1 (if not stated otherwise) followed by centrifugation (1500 rpm, 5 min, 4°C). 60 minutes after addition of bacteria, epithelial monolayers were washed three times with PBS, and fresh medium containing 50 µg/mL gentamycin was added to the culture medium to restrict extracellular bacterial growth. Unless indicated otherwise, infections were completed after 4 h post infection. To quantify bacterial invasion, co-culture of 20, 40 or 60 min was followed by 1 h incubation in fresh cell culture medium supplemented with 50 µg/mL gentamycin. For 4 h and 6 h infection, gentamycin was supplemented 60 minutes after addition of bacteria, and incubation was carried out for additional 3 h or 5 h. After washing, cells were lysed in 0.1% Triton/H2O and the number of intracellular bacteria was determined (CFU) by serial dilution and plating. Bacteria free conditioned medium were prepared by centrifugation or filtering of cell culture medium, and immediately applied on naïve, uninfected m-ICcl2 cells. The rabbit anti-Listeria antibody was used for immunolabelling of bacteria (1∶500). For alternative intracellular detection of Listeria, bacteria were biotinylated prior to infection according to the manufacturers protocol. Pharmacological inhibitors were added 30 min prior to infection if not stated otherwise.
For intracellular Cxcl-2 or Cxcl-5 visualization, brefeldin A (0.5 µg/mL) was added to the cell culture medium 1 h after stimulation. Cells were fixed in 3% PFA and incubated with anti-Cxcl-2 or Cxcl-5 antiserum (1∶100). Nox4 was detected in formalin-fixed sections of mouse small intestine by incubation with a rabbit anti-Nox4 antiserum (1∶100) for 1 h at room temperature, followed after washing by a TR-conjugated secondary antibody. Cells were mounted in Vectashield Mounting Medium with Dapi (Vector Laboratories, Eching, Germany) and visualized using a Leica DM IRB Inverted Research Microscope with a TCS SP2 AOBS scan head (Leica Microsystems GmbH, Wetzlar, Germany). For fluorescent detection, immunolabelled Listeria was additionally stained with AF 555-conjugated secondary antibody prior to infection, or biotinylated bacteria were labelled by streptavidin-conjugated Cy3. For flow cytometry cells were trypsinized and fixed in Cytofix (BD Biosciences). Cxcl-2 or Cxcl-5 was stained following permeabilization in 0.5% saponin/1% FCS/PBS buffer. Analysis was performed on a FACS Calibur apparatus (BD Biosciences). The data acquisition on GFP+ (recorded in channel Fl-1) and Cxcl-2+ or Cxcl-5+ cells (Cy5-conjugated, Fl-4) was restricted to a total number of 10.000 events. The data acquisition on GFP+ (recorded in channel Fl-1) bacteria was restricted to a total number of 100.000 events. Flow cytometry cell sorting was performed using a MoFlo (XDP Upgrade, Beckman-Coulter) at the Cell Sorting Facility, Medical School, Hanover.
Cell were lysed in 3∶1 WB/SB vol/vol (WB: 50 mM Tris, pH 7.4, 120 mM NaCl; SB: 250 mM Tris, pH 6.5, 8% SDS, 40% glycerol; supplemented with a proteinase inhibitor cocktail [Complete Mini, Roche Diagnostics]). Samples were sonified and the protein concentration was determined (DC Protein Assay; Bio-Rad Laboratories). Protein was separated on 11% acrylamide gels and blotted on nitrocellulose. Membranes were incubated overnight at 4°C with the primary antibody. Detection was performed using peroxidase-labelled goat anti–rabbit or goat anti-mouse secondary antibodies in combination with the ECL kit (GE Healthcare). Before restaining, membranes were stripped for 45 min at 50°C in 62.5 mM Tris HCl, pH 6.7, 100 mM ß-mercaptoethanol and 2% SDS, followed by three 15-min washing steps.
Cells were divided after cell sorting. DNA extraction was performed following incubation in lysosyme (10 mg/mL), proteinase K (10 mg/mL), and 5% SDS using TRIzol (Invitrogen) according to the manufacturer's instruction. DNA was washed in sodium citrate (0.1 mM) and precipitated in 75% ethanol. Listeria genomic DNA was detected by PCR (Taq DNA polymerase from Invitrogen) using primers specific for the listerial hly gene (forward: 5′-ATG TAA ACT TCG GCG CAA CT-3′, reverse: 5′-TCG TGT GTG TTA AGC GGT TT-3′, annealing 57°C, cycles 35). A fragment encoding eukaryotic hypoxanthine phosphoribosyltransferase (Hprt) was amplified using oligonucleotides 5′-TGC TGA CCT GCT GGA TTA CA-3′ and 5′-GCT TAA CCA GGG AAA GCA AA-3′ (annealing temperature 59°C, cycles 32) as control. Amplification products were analysed on a 2% agarose gel and visualized with SYBR Safe (Invitrogen). Total RNA was extracted using the RNeasy Protect Cell Mini Kit (Qiagen) and first-strand cDNAs was synthesized using oligo-dT primers. Real-time PCR was prepared with absolute QPCR ROX mix (Thermoscientific), sample cDNA, intron-spanning forward and reverse primers, as well as the 6-carboxy-fluorescein-conjugated target probe provided in the commercial TaqMan gene expression assay for murine Hprt1 and Cxcl-2 or Cxcl-5 (Applied Biosystems). Analysis were performed using an ABI PRISM Sequence Detection System 7000 (Applied Biosystems). Samples were normalized to the endogenous control. Results were calculated by use of the Δ2-CT method and are presented as fold induction of target gene transcripts in stimulated relative to unstimulated controls.
The fluorogenic probe 5-(and-6)-carboxy-2′,7′-dichlorofluorescein diacetate (DCF-DA, Invitrogen) was used for reactive oxygen intermediates (ROI) visualization. Prior to stimulation, cells were incubated with DCF-DA (10 µg/mL) for 30 min at 37°C. After stimulation or infection for 20 min cells were rapidly rinsed with PBS, fixed in 3% PFA, washed twice with PBS and analyzed by fluorescence microscopy. For quantitative analysis of oxygen radical production, cells were rinsed with PBS to remove the free probe, and lysed in 200µl of 1% Triton/H2O. The lysate was transferred into microcentrifuge tubes, sedimented at 8,000×g for 5 min at 4°C, and 100 µl aliquots were dispensed in 96-well plates in triplicate. The index of oxidation (DCF) was calculated as the ratio of fluorescence intensity as compared to an untreated control.
Cells were grown in collagen-coated 8-well chamber slides (Nunc, Rochester, NY) continuously bathed in cell culture medium. The 70 kDa high molecular weight gap junction impermeant fluorescent compound Texas Red Dextran (Molecular Probes, 10 mg/mL) was mixed with either the <1 kDa low molecular weight Lucifer Yellow (Molecular Probes, 10 mg/mL) or 0.5 mM cumene hydroperoxide in injection buffer (25 mM HEPES, 125 mM K-acetate, 5 mM Mg-acetate, pH 7.1). Fluorescent mixtures were loaded into individual Femtotips II injection capillars (Eppendorf, Hamburg, Germany). Cells were transferred to a LSM 510 META laser scanning confocal microscope equipped with an inverted Axiovert 200M stand (Carl Zeiss, Germany) and single cell microinjection was performed by using an InjectMan NI2/Femtojet injector system at pi: 180 hPa, ti: 0.2s, pc: 25 hPa. A minimum of 10 microinjected cells were analyzed per experiment. To study gap junctional intercellular communication, cells were analysed by live imaging microscopy after 5 min incubation. For ROI donor cumene hydroperoxide stimulation, cells were incubated 1 h, washed, and incubated in prewarmed fresh cell culture medium for an additional 3 h in the presence of 0.5 µg/mL brefeldin A. Cells were fixed in 3% PFA and further analyzed by intracellular chemokine staining and fluorescence microscopy.
All experiments were performed at least three times and results are given as the mean ± standard deviation (SD) of one representative experiment. Statistical analyses were performed using the Student's t test. A p value<0.05 (*) or <0.01 (**) was considered significant.
|
10.1371/journal.ppat.1001056 | Enterohemorrhagic E. coli Requires N-WASP for Efficient Type III Translocation but Not for EspFU-Mediated Actin Pedestal Formation | Upon infection of mammalian cells, enterohemorrhagic E. coli (EHEC) O157:H7 utilizes a type III secretion system to translocate the effectors Tir and EspFU (aka TccP) that trigger the formation of F-actin-rich ‘pedestals’ beneath bound bacteria. EspFU is localized to the plasma membrane by Tir and binds the nucleation-promoting factor N-WASP, which in turn activates the Arp2/3 actin assembly complex. Although N-WASP has been shown to be required for EHEC pedestal formation, the precise steps in the process that it influences have not been determined. We found that N-WASP and actin assembly promote EHEC-mediated translocation of Tir and EspFU into mammalian host cells. When we utilized the related pathogen enteropathogenic E. coli to enhance type III translocation of EHEC Tir and EspFU, we found surprisingly that actin pedestals were generated on N-WASP-deficient cells. Similar to pedestal formation on wild type cells, Tir and EspFU were the only bacterial effectors required for pedestal formation, and the EspFU sequences required to interact with N-WASP were found to also be essential to stimulate this alternate actin assembly pathway. In the absence of N-WASP, the Arp2/3 complex was both recruited to sites of bacterial attachment and required for actin assembly. Our results indicate that actin assembly facilitates type III translocation, and reveal that EspFU, presumably by recruiting an alternate host factor that can signal to the Arp2/3 complex, exhibits remarkable versatility in its strategies for stimulating actin polymerization.
| The food-borne pathogen enterohemorrhagic E. coli (EHEC) O157:H7 can cause severe diarrhoea and life-threatening systemic illnesses. During infection, EHEC attaches to cells lining the human intestine and injects Tir and EspFU, two bacterial molecules that alter the host cell actin cytoskeleton and stimulate the formation of “pedestals” just beneath bound bacteria. Pedestal formation promotes colonization during the later stages of infection. N-WASP, a host protein known to regulate actin assembly in mammalian cells, was previously shown to be manipulated by Tir and EspFU to stimulate actin assembly, and to be required for EHEC to generate actin pedestals. Surprisingly, we show here that N-WASP promotes the efficient delivery of Tir and EspFU into mammalian cells, and that when we utilized a related E. coli to enhance type III delivery of Tir and EspFU, actin pedestals assembled even in its absence. Thus, EHEC stimulates at least two pathways of actin assembly to generate pedestals, one mediated by N-WASP and one by an unidentified alternate factor. This flexibility likely reflects an important function of pedestal formation by EHEC, and study of the underlying mechanisms may provide new insights into the pathogenesis of infection as well as the regulation of the actin cytoskeleton of mammalian cells.
| Enterohemorrhagic Escherichia coli (EHEC) are an important source of diarrheal illness worldwide and are the leading cause of pediatric renal failure in the United States. O157:H7 is the most common EHEC serotype associated with serious illness and includes many of the most virulent strains [1]. During colonization, EHEC induce striking morphological changes of the intestinal epithelium, resulting in the formation of attaching and effacing (AE) lesions. These structures are characterized by the effacement of microvilli and intimate attachment of EHEC to the epithelial cell surface. The adherent bacteria also reorganize the host cell cytoskeleton into filamentous (F-)actin pedestals. In addition to EHEC, several related pathogens, including enteropathogenic E. coli (EPEC), also generate AE lesions and actin pedestals on intestinal epithelial cells during the course of infection [1]. Importantly, mutations in any of these bacteria that abolish their ability to generate AE lesions prevent their colonization [2], [3], [4], [5]. Moreover, an EHEC mutant that is capable of intimate attachment but selectively defective for actin pedestal formation does not expand its initial infectious niche in experimentally-infected rabbits [6].
The capacity to generate actin pedestals depends on the translocation of bacterial effector proteins into mammalian host cells via a type III secretion system (T3SS) [7], [8]. This macromolecular structure spans the inner and outer bacterial membranes, extends from the bacterial surface, and includes a long filamentous appendage that contacts the mammalian cell surface and functions as a conduit for effector secretion. The tip of this filament includes translocator proteins that form pores in target cell membranes and promote the entry of effectors into the mammalian cell.
The EHEC- and EPEC-encoded type III secretion apparatuses are homologous to the T3SSs found in a wide range of pathogens, many of which also trigger actin rearrangements in the host cell. For example, type III translocated effectors of Shigella, Salmonella, and Yersinia induce cytoskeletal changes that can promote bacterial entry into the host cell. Actin assembly may also affect type III translocation, because several effectors that misregulate signaling pathways that control the actin cytoskeleton have a significant influence on the efficiency of translocation by Shigella and Yersinia [9], [10].
For AE pathogens, the T3SS delivers effectors that activate the WASP and N-WASP actin nucleation-promoting factors to promote pedestal formation [11], [12], [13]. WASP, which is expressed in hematopoietic cells, and its homolog N-WASP, which is ubiquitously expressed, stimulate the Arp2/3 complex, a group of seven proteins that collectively nucleate actin into filaments [14], [15]. The C-terminal WCA (WH2-connector-acidic) domain of N-WASP directly binds and activates the Arp2/3 complex, but this domain is normally sequestered by its intramolecular interaction with an internal regulatory element, the GBD (GTPase-binding l;domain). Binding of the GTPase Cdc42 to the GBD disrupts these autoinhibitory GBD-WCA interactions, and frees the WCA domain to activate Arp2/3-mediated actin assembly. Other factors, including the SH2/SH3 domain-containing adaptor proteins Nck1-2, also activate N-WASP, but bind to a proline-rich domain (PRD) that lies between the GBD and WCA regions [16], [17].
One effector essential for intimate attachment and actin pedestal formation by AE pathogens is the Tir (translocated intimin receptor) protein [18], [19]. Upon type III translocation into the mammalian cell, Tir becomes localized in the plasma membrane with a central extracellular domain that binds the bacterial outer membrane adhesin intimin [20]. N- and C-terminal to the intimin-binding domain are two transmembrane segments and the intracellular domains of Tir. For canonical EPEC strains of serotype O127:H6, Tir is the only effector required for pedestal formation, as simply clustering Tir in the plasma membrane is sufficient to recruit the Nck adaptor proteins and trigger F-actin assembly [21].
In contrast to canonical EPEC strains, EHEC strains of serotype O157:H7 require a second translocated effector, in addition to Tir, to trigger pedestal formation. EHEC Tir recruits this effector, named EspFU (also known as TccP) [22], [23], indirectly, as the host protein intermediates IRTKS and IRSp53 are responsible for linking EspFU to Tir during actin pedestal assembly [24], [25]. EspFU contains a C-terminal region with multiple 47-residue proline-rich repeats that each bind to the GBD of N-WASP and directly displace the WCA domain to allow it to activate the Arp2/3 complex [26], [27]. Whereas a single EspFU repeat is capable of activating N-WASP, tandem repeats synergize during actin polymerization by promoting N-WASP dimerization, which allows it to bind Arp2/3 with much higher affinity than monomeric N-WASP [27], [28], [29].
EHEC is unable to generate pedestals on N-WASP-deficient cells [12], and the fact that EspFU targets N-WASP to promote actin assembly provides a highly plausible explanation for this finding. Nevertheless, the observations that actin assembly influences type III translocation by other pathogens raised the possibility that N-WASP may also contribute to an earlier step in the process of pedestal formation. In fact, we show here that N-WASP and actin assembly are important for the translocation of Tir and EspFU into mammalian cells by EHEC O157:H7. Intriguingly, when delivered into cells by EHEC-independent means, Tir and EspFU are fully capable of stimulating actin pedestal formation in the absence of N-WASP. These results add an additional layer of complexity to our understanding of the interactions between EHEC and its host cells, and highlight the functional versatility of EspFU.
N-WASP deficiency in cultured mammalian cells is known to block actin pedestal formation by EHEC [12]. An obvious rationale for this requirement is that N-WASP promotes actin polymerization in the pedestal, as suggested by the observation that EspFU recruits, binds and activates N-WASP [22], [23]. However, given the evidence that actin polymerization might also facilitate the delivery of effectors into the host cell [9], [10], we examined a role for N-WASP during type III effector translocation by EHEC using genetically modified murine fibroblast-like cells (FLCs) [30]. Consistent with the previous characterization of wild type (NW+/+) and N-WASP knockout (NW−/−) cell lines, immunoblotting demonstrated that N-WASP was expressed only in the wild type cells (Figure 1A, left). We also investigated the expression of the N-WASP homolog WASP, which is also a target of EspFU [26], and found that neither WASP mRNA or protein was detected in NW−/−cells (Figure 1A, right). As reported using an independently derived N-WASP-deficient cell line [12], EHEC generated actin pedestals on wild type, but not knockout cells (Figure 1B).
To assess Tir translocation, we fused the C-terminus of the EHEC Tir molecule to the TEM-1 β-lactamase (Bla). The translocation of this fusion protein into host cells can be detected by β-lactamase-mediated cleavage of a FRET reporter, resulting in a change in fluorescent wavelength emission from green (520 nm) to blue (460 nm), as previously described [31]. Such fusions have been used extensively for assessing Tir translocation [32], and maintain Tir function, as our Tir-Bla fusion complemented a bacterial Tir deletion for pedestal-forming function on NW+/+ cells (Figure S1A). After infection of wild type or N-WASP-deficient FLCs with EHEC expressing the Tir-Bla fusion, the percentage of blue cells was scored visually by fluorescent microscopy (Figure S1B) and expressed as a translocation index. By this measure, the translocation of Tir by EHEC into N-WASP-knockout cells occurred ∼3-fold less efficiently than into wild type cells after a 6h infection (Figure 1C). The requirement for N-WASP for efficient translocation was not restricted to Tir, because the level of translocation of an EspFU-Bla fusion into N-WASP-knockout cells was also diminished relative to wild type cells, albeit not quite as low as translocation of Tir-Bla (Figure 1C). In accordance with these results, we found that treatment of HeLa cells with wiskostatin, an inhibitor of N-WASP, significantly impaired translocation of the EspFU-Bla fusion into host cells (Figure S1C).
Given that the Tir-Bla translocation index relies on binary scoring of (green vs. blue) cells by visual inspection, it may not reflect the true severity of the defect in Tir translocation into NW−/− cells. The deficiency in the translocation of Tir into N-WASP-deficient cells by EHEC is predicted to result in a decrease in the amount of Tir clustered beneath bound bacteria. Therefore, to examine Tir localization, we infected wild type or knockout cells with EHECΔtir harboring pHA-TirEHEC, which encodes an N-terminally HA-tagged Tir that can be detected with an anti-HA antibody and visualized microscopically. Whereas Tir foci were readily observed beneath EHEC bound to wild type cells, they were not detected beneath EHEC on N-WASP-knockout cells (Figure 1D), consistent with a significant defect in Tir translocation.
To test whether the requirement for N-WASP for efficient translocation reflects a role for F-actin assembly in promoting translocation, we examined Tir localization beneath bound bacteria after treatment with cytochalasin D, which binds actin filament ends and prevents polymerization. Microscopic visualization revealed that cytochalasin D treatment resulted in a loss of foci of HA-tagged Tir beneath bacteria bound to HeLa cells (Figure 1E). In addition, cytochalasin D and latrunculin A, which binds actin monomers and triggers depolymerization, each partially inhibited of translocation of Tir-Bla fusion protein into HeLa cells (data not shown). Collectively, these data suggest that N-WASP-mediated actin polymerization facilitates EHEC-mediated effector translocation.
We next tested whether impaired Tir translocation into N-WASP-knockout FLCs results in a measurable effect on the ability of Tir to promote bacterial attachment. We infected wild type or N-WASP-knockout cells with the intimin-deficient EPECΔeae or EHECΔeae mutants to allow for translocation of Tir, and then, after killing these bacteria with gentamicin and removing them by washing, challenged these cells with non-pathogenic GFP-expressing E. coli strains that harbor pIntEPEC or pIntEHEC plasmids to express intimin. Previous studies have shown that E. coli/pInt, but not E. coli/vector, attach to monolayers primed with EPEC or EHEC strains that translocate Tir, but not to unprimed monolayers [33], [34], thus allowing a specific measure of native intimin binding to translocated Tir. A bacterial binding index, defined as the percentage of cells with at least five adherent GFP- and intimin-expressing bacteria, was determined microscopically. Bacterial binding to N-WASP-knockout cells primed with EHECΔeae was approximately 3-fold lower than to primed wild type cells (Figure 2A).
EPEC generates pedestals on cultured cells more efficiently than EHEC [35], so we tested whether EPEC might correspondingly translocate Tir into N-WASP-knockout cells more efficiently. In fact, bacterial binding to N-WASP-deficient cells primed with EPECΔeae was indistinguishable from binding to EPECΔeae-primed wild type cells (Figure 2A). To test whether the difference between EHEC and EPEC in functional Tir translocation was due to allelic differences in their respective Tir proteins, we primed wild type or N-WASP-knockout FLCs with EPECΔtir-eae expressing either HA-TirEPEC or HA-TirEHEC, and then challenged cells with E. coli expressing the corresponding intimin ligand. Alternatively, we primed cells with EHECΔtir-eae expressing either HA-TirEHEC or HA-TirEPEC prior to challenge. We found that EPEC was capable of translocating either Tir variant into N-WASP-knockout cells to promote intimin-mediated attachment at nearly wild type levels. In contrast, priming with EHEC expressing either HA-TirEHEC or HA-TirEPEC gave binding values two- to three-fold lower than wild type (Figure 2B). These observations indicate that Tir translocation by EHEC is more dependent on N-WASP than Tir translocation by EPEC, irrespective of the genetic origin of the Tir molecule.
The observations that EHEC does not efficiently translocate Tir or EspFU into N-WASP-deficient cells, raised the intriguing possibility that the defect in EHEC pedestal formation on these cells was due to inefficient effector translocation into cells rather than a lack of Tir-EspFU signaling within the cell. Since, in the functional assay described above, EPEC translocated Tir into N-WASP knockout cells better than EHEC, we adopted a heterologous expression system using KC12, an EPEC derivative that has been chromosomally engineered to express HA-tagged EHEC Tir [22], [36], for achieving delivery of EHEC Tir and EspFU into N-WASP-knockout cells. Importantly, although translocation of TirEHEC-Bla and EspFU-Bla into N-WASP-deficient cells by KC12 occurred with somewhat delayed kinetics compared to wild type cells (Figure S2), the defect in translocation was mild at 6 h postinfection (Figure 3A). To determine if type III translocation by KC12 was reflected in the localization of Tir beneath bound bacteria, we infected N-WASP-knockout cells with KC12/pEspFU, a strain that expresses a myc-tagged EspFU harboring six C-terminal repeats and generates actin pedestals in manner that is mechanistically indistinguishable from canonical EHEC strains [22]. HA-Tir foci were observed with somewhat delayed kinetics and lower frequency in NW−/− than NW+/+ FLCs, but nearly 50% of KC12/pEspFU bound to N-WASP-knockout cells generated Tir foci by 5 h postinfection (Figure 3B).
Given that KC12/pEspFU was only partially diminished for Tir and EspFU translocation, we sought to determine whether this strain could generate actin pedestals on N-WASP knockout cells. Remarkably, upon infection of NW−/− FLCs, numerous actin pedestals were formed by KC12/pEspFU (Figure 3C, top row), indicating that EHEC Tir and EspFU are capable of signaling to the actin cytoskeleton in the absence of N-WASP. Pedestal formation required EspFU, because KC12 lacking pEspFU failed to generate pedestals in these cells (Figure 3C, bottom row). To quantify the efficiency of actin pedestal formation, we infected wild type and N-WASP-knockout cells with KC12/pEspFU, visually identified sites of HA-Tir localization beneath bound bacteria, and then calculated the percentage of those Tir foci that were associated with actin pedestals. This specific scoring method circumvented the inhibitory effects of N-WASP deficiency on effector entry (Figure 1C; Figure 2) and HA-Tir localization in cells (Figure 3B), and specifically measured intracellular signaling after Tir translocation. KC12/pEspFU and the control strain EPECΔtir/pHA-TirEPEC, which generates pedestals using the Nck-N-WASP-dependent pathway [13], [36], [37], both formed pedestals efficiently on wild type cells: after infection for 3h, >95% of Tir foci were associated with pedestals, while at 5h this level reached >98% (Figure 3D). In NW−/− FLCs, EPECΔtir/pHA-TirEPEC, which utilizes Nck adaptor proteins to activate N-WASP, was totally incapable of generating pedestals (Figure 3D), consistent with results utilizing an independently generated N-WASP knockout cell line [13]. In contrast, 80% of KC12/pEspFU-associated Tir foci triggered actin pedestals at 3h, and this level rose to 95% at 5h postinfection. Thus, the more efficient delivery of EHEC Tir and EspFU by the EPEC-derived strain KC12 results in a surprisingly effective ability to induce pedestal formation in the absence of N-WASP.
IRTKS, along with the closely related protein IRSp53, regulates actin dynamics at the plasma membrane [38], and functions as a linker between EHEC Tir and EspFU during N-WASP-promoted pedestal formation [24], [25]. Given that EspFU localized to sites of bacterial attachment in N-WASP-knockout cells (Figure 3C), we assessed whether IRTKS plays a role in EspFU recruitment in the absence of N-WASP by examining the distribution of IRTKS in N-WASP-knockout cells infected with KC12/pEspFU. Immunofluorescence microcopy indicated that IRTKS localized near the tips of pedestals (Figure 4, top row), where it colocalized with EspFU (middle row), consistent with a role in linking Tir and EspFU during N-WASP-independent signaling. Moreover, when these cells were infected with KC12 lacking EspFU, IRTKS still localized to sites of bacterial attachment, suggesting that even in the absence of EspFU, N-WASP, and actin pedestals, the Tir-binding activity of IRTKS is sufficient to promote IRTKS recruitment (Figure 4, bottom row). Thus, N-WASP does not have any apparent effects on the signaling events that occur between type III effector translocation and EspFU recruitment to Tir.
For pedestal formation by EHEC on wild type cells, recruitment and membrane clustering of a complex of Tir, IRTKS and EspFU is sufficient to trigger pedestal formation [24]. However, we have also shown that membrane clustering of HN-Tir-EspFU-[R1-6], a fusion in which the C-terminal cytoplasmic domain of Tir is replaced by six C-terminal repeats of EspFU, is fully functional for pedestal formation [26], [28], indicating that clustering of EspFU alone is sufficient to stimulate this signaling pathway. To similarly determine the potential requirements of EspFU, Tir and IRTKS during N-WASP-independent actin pedestal formation, we tested whether HN-Tir-EspFU-[R1-6] could trigger actin assembly in N-WASP-knockout cells. After transfection with a plasmid encoding HN-Tir-EspFU-[R1-6], we infected NW−/− FLCs with EPECΔtir and treated cells with an anti-HA antibody to visualize the fusion protein and with phalloidin to stain F-actin. These bacteria readily generated actin pedestals on cells expressing HN-Tir-EspFU-[R1-6], but not cells expressing HN-TirΔC, which lacks a C-terminal signaling domain (Figure 5A), indicating that the EspFU repeats are essential for actin pedestal formation in these cells.
To test whether pedestal formation on N-WASP-knockout cells requires any proteins other than the Tir-EspFU fusion, we next treated HN-Tir-EspFU-[R1-6]-expressing cells with the non-pathogenic E. coli strain that expresses intimin. These bacteria, which are incapable of type III secretion and serve to simply cluster the HN-Tir-EspFU-[R1-6] fusion protein, generated actin pedestals on N-WASP-knockout cells in a manner indistinguishable from those formed on wild type cells (Figure 5B and [26], [28]). In contrast, a control HN-Tir fusion protein lacking the C-terminal repeats of EspFU was unable to elicit pedestals. Thus, we conclude that, as is the case for pedestal formation in N-WASP-proficient cells, the central role of Tir and IRTKS in N-WASP-knockout cells is to promote the clustering of the EspFU repeats beneath the plasma membrane. Moreover, in the absence of N-WASP, EspFU remains the most essential component of the signaling pathway that leads to actin pedestal assembly.
The interaction of EspFU with N-WASP or WASP results in the activation of the Arp2/3 complex and actin nucleation in vitro [22], [26], [27], [39]. To examine the potential role of Arp2/3 in pedestals generated in the absence of N-WASP, we first assessed whether this complex is recruited to sites of pedestal formation in N-WASP-knockout cells. Immunofluorescence microscopy using anti-Arp3 antibodies revealed recruitment of the Arp2/3 complex in pedestals formed by KC12/pEspFU in N-WASP knockout cells as well as wild type cells (Figure 6A), suggesting that Arp2/3 is likely involved in actin pedestal formation under both circumstances.
To test for a functional role of the Arp2/3 complex in pedestal formation, we took advantage of the fact that overexpression of the N-WASP WCA domain results in sequestration and/or ectopic activation of the Arp2/3 complex [14], [15], [40]. Whereas 95% of cells expressing a GFP control protein contained pedestals upon infection with EPECΔtir/pHA-TirEPEC or KC12/pEspFU, <5% of wild type FLCs expressing GFP-WCA harbored pedestals (Figure 6B and data not shown), confirming the importance of proper Arp2/3 activity in actin pedestal assembly. Moreover, this dominant negative GFP-WCA construct also blocked actin pedestal formation by KC12/pEspFU in N-WASP-knockout FLCs (Figure 6B). Finally, genetic depletion of the Arp2/3 subunits Arp3 and ARPC4 abolished pedestal formation on wild type HeLa cells, which are predicted to support both N-WASP-dependent and N-WASP-independent pedestal formation (Figure S3). Consistent with previous reports, we found that EspFU derivatives were unable to directly activate the Arp2/3 complex to promote actin polymerization in vitro (Figure S4; [23], [27]). Collectively, these data suggest that in generating pedestals in N-WASP-deficient cells, EspFU recruits an alternate host factor (or factors) that triggers Arp2/3-mediated actin assembly.
WASP and N-WASP are members of a family of nucleation promoting factors (NPFs) that activate Arp2/3, a family that includes WAVE proteins, WASH, and WHAMM [41]. IRSp53, which has been shown to link Tir and EspFU in some cells [25] can bind and activate WAVE2 [38]. In addition, WAFL is a protein with a predicted Arp2/3-binding acidic peptide that associates with actin filaments and has been implicated in endosomal trafficking [42]. To investigate whether these factors could be involved in N-WASP-independent actin pedestal formation, we determined whether they localized to actin pedestals generated in an N-WASP-independent manner. NW−/− FLCs ectopically expressing GFP fusions to WAVE2, WASH, WAFL, WHAMM, or (as a control) N-WASP were infected with KC12/pEspFU and phalliodin-stained to visualize actin pedestals. Pedestals were efficiently formed in the presence of all NPFs, and as expected, GFP-N-WASP distinctly localized to pedestals (and often to their tips) beneath bound bacteria (Figure 7A, top row, and data not shown). None of the other NPFs localized in a similar fashion (Figure 7A). GFP-WAVE2 faintly and diffusely localized to sites of bacterial attachment (Figure 7A, second row), but this localization was also observed around bacteria that were not associated with actin pedestals (data not shown). Furthermore, WAVE2 was not required for N-WASP-independent pedestal formation, because KC12/pEspFU generated pedestals normally on NW−/− FLCs in which WAVE2 expression was stably knocked down by more than 95% (Figure 7B). Together with the observation that EspFU does not directly activate Arp2/3, these data are consistent with the model that EspFU is capable of utilizing an alternate NPF to activate Arp2/3 in NW−/− FLCs.
Allosterically activated N-WASP is associated with more potent actin assembly when multimerized [29], [43], [44], [45], an observation explained by the ability of dimeric N-WASP to bind the Arp2/3 complex with much higher affinity than monomeric N-WASP [29]. Nevertheless, when Tir-EspFU fusions harboring different numbers of repeats are clustered beneath the plasma membrane using anti-Tir antibody-coated particles, a single EspFU repeat is capable of triggering actin pedestal formation in N-WASP-proficient cells [26], [28]. This prompted us to examine the role of the repeat quantity in N-WASP-independent actin assembly. To directly compare a requirement for different numbers of repeats during EspFU-mediated assembly in the presence or absence of N-WASP, we used S. aureus and anti-Tir antibodies to cluster HN-Tir-EspFU fusions harboring various numbers of repeats in wild type and N-WASP-knockout FLCs. We then measured the fraction of cells that contained actin pedestals. Whereas in wild type FLCs, a single repeat resulted in pedestal formation levels of ∼45% (“R1”, Figure 8A, black bars [28]), this derivative generated no pedestals in N-WASP-knockout cells (Figure 8A, gray bars). (Note that the experiments with wild type and N-WASP-knockout cells were performed in parallel, but those using wild type cells were published previously [28] and are shown in Figure 8A for ease of comparison.) Clustering of Tir-EspFU fusions harboring two to six repeats in N-WASP-knockout FLCs resulted in cellular pedestal formation efficiencies of approximately 50–55%, which is significantly less than the levels of 75–90% that were observed in wild type cells (Figure 8A; [28]).
To further investigate the relationship between number of repeat units and N-WASP-independent actin polymerization we sought to measure pedestal formation when EspFU is present in the cytosol and Tir is independently translocated into the plasma membrane. Under these conditions, ∼90% of wild type cells expressing any GFP-EspFU construct containing at least two repeats generated pedestals in response to infection with KC12 (Figure 8B; [28]). To similarly examine pedestal formation when EspFU must act in concert with Tir in the absence of N-WASP, we infected GFP-EspFU-expressing N-WASP-knockout cells with KC12 (Figure S5). Only ∼50% of cells expressing the four- and six-repeat truncations generated pedestals, and just 15% of cells expressing the three-repeat derivative formed pedestals (Figure 8B). No pedestals were observed in cells expressing fewer than three repeats (Figure 8B). In addition, in cells expressing EspFU derivatives harboring three or more repeats, pedestal formation was less efficient without N-WASP. Thus, N-WASP deficiency is associated with a more stringent requirement for multimeric EspFU variants in order to trigger actin assembly, and even for multi-repeat EspFU derivatives that do trigger assembly in N-WASP-knockout cells, the efficiency of pedestal formation was somewhat reduced.
The EspFU repeat contains an N-terminal region that, upon WASP binding, adopts an α-helical conformation that interacts with a hydrophobic groove in the GBD [26]. Thus, alanine substitution of three conserved hydrophobic residues in the EspFU α-helix abolished N-WASP recruitment and actin assembly in mammalian cells [26]. To test whether this region of the EspFU repeat plays an essential role in N-WASP-independent actin assembly, we constructed a Tir-EspFU fusion comprising two EspFU repeats that each harbored the V4A/L8A/L12A triple alanine substitution (referred to as VLL/AAA; Figure 9A). Tir-EspFU-2RVLL/AAA and the corresponding wild type variant, Tir-EspFU-2RWT, were expressed in NW−/− FLCs and clustered in the membrane using an EPEC strain that expresses intimin but not Tir or EspFU, and the cells stained with an anti-HA antibody to visualize the clustered fusion proteins and with phalloidin to stain F-actin (Figure 9B). Clustering of Tir-EspFU-2RWT but not Tir-EspFU-2RVLL/AAA induced robust pedestal formation under bound bacteria (Figure 9B, top row), indicating that the WASP/N-WASP-binding region of EspFU is critical for N-WASP-independent actin pedestal formation.
N-WASP is required for actin pedestal formation by EHEC [12], and the observation that EspFU directly binds and activates this nucleation-promoting factor provided an obvious explanation for this requirement. However, we now show that N-WASP is also important for an earlier step in actin pedestal formation, type III translocation of Tir and EspFU. We evaluated three properties of Tir that would reflect proper translocation into mammalian host cells. We assessed entry of Tir-Bla reporter proteins into the cytosol, quantified the ability of intimin-expressing bacteria to bind to primed host cells containing translocated Tir, and directly visualized the localization and clustering of Tir in the plasma membrane. These approaches each revealed that Tir translocation was diminished in N-WASP-knockout cells. The translocation defect was not restricted to Tir, because EHEC-mediated delivery of an EspFU-Bla fusion protein was also lower in N-WASP-knockout cells. Given that F-actin assembly promotes type III translocation of effectors by other pathogens [9], [10], it seems likely that the ability of N-WASP to promote actin assembly contributes to translocation by EHEC. Consistent with this possibility, translocation was significantly impaired by cytochalasin D or latrunculin A, which inhibit actin assembly, or by wiskostatin, an inhibitor of N-WASP [46].
These results raise the possibility that one of the functions of Tir- and EspFU-driven actin polymerization is to promote efficient translocation of one or more of the other 20–30 EHEC effectors. Interestingly, multiple pathogens encode type III secreted proteins that modify the actin cytoskeleton and have been shown to influence type III translocation. For example, the Shigella type III translocon protein IpaC stimulates Src recruitment and actin polymerization at sites of bacterial entry, and its inactivation diminishes type III translocation [10]. The Yersinia effectors YopE and YopT induce misregulation of Rho-family GTPases, inhibit signaling from these cytoskeletal regulators, and are postulated to temporally limit the phase of high efficiency type III translocation [9], [47], [48], [49], [50]. For EHEC, low levels of Tir translocation still occurred when actin polymerization was disrupted (Figure 1C), and multiple reports have demonstrated that EHEC mutants defective in pedestal formation are still capable of translocation [22], [39], [51], [52], indicating that actin assembly is not absolutely required for this process. This residual level of translocation may also explain the observation that for N-WASPdel/del cells, an independently derived N-WASP-deficient embryonic fibroblast cell line, Tir is translocated by an EHECΔespFU mutant efficiently enough to recruit ectopically expressed GFP-EspFU beneath sites of bacterial attachment [25].
Although it has now been shown that actin assembly promotes type III translocation by several pathogens, the specific function(s) of assembly is not clear. For EHEC, pedestal formation may increase the area of bacterium-host cell contact and/or the stability of bacterial binding, thereby enhancing effector translocation. Alternatively, type III translocation by several pathogens, including EPEC, is thought to occur at lipid microdomains [53], [54], [55], and it has been postulated that actin assembly may facilitate the recruitment of such domains to bound bacteria [9]. Interestingly, we found that although EPEC-mediated translocation of Tir into mammalian cells was somewhat delayed and diminished by N-WASP-deficiency, this defect was not large enough to have a discernable defect in intimin-mediated bacterial binding. The reasons for the lower N-WASP-dependence of translocation by EPEC are not known, but EPEC exhibits particularly robust type III secretion in vitro and generates pedestals more efficiently on cultured cells than does EHEC [35].
We utilized KC12/pEspFU, an EPEC strain engineered to express TirEHEC and EspFU, to more efficiently deliver EHEC Tir and EspFU into N-WASP knock out cells. Surprisingly, these effectors were capable of generating actin pedestals with ultimately high efficiency: 95% of Tir foci beneath cell-bound KC12/pEspFU were associated with pedestals in N-WASP-knockout cells after 5h infection. Thus, the defect in EHEC pedestal formation on N-WASP-knockout FLCs is apparently not due to an inability of Tir and EspFU to stimulate actin polymerization once delivered to the mammalian cell.
The N-WASP-independent pathway of pedestal formation shares several parallels with pedestal formation in wild type cells. Tir and EspFU are the only bacterial effectors required, since ectopic expression of the two proteins in N-WASP-knockout cells, followed by Tir clustering, was sufficient to induce localized actin assembly (data not shown). IRTKS, which has been shown to link EspFU to Tir in N-WASP-proficient cells [24], [25], was recruited to sites of bacterial attachment on N-WASP-knockout cells. Moreover, the central role of the C-terminal cytoplasmic domain of Tir is to promote IRTKS-mediated recruitment of EspFU, because a Tir fusion protein in which this Tir domain is replaced by EspFu was competent for triggering actin polymerization. Finally, the Arp2/3 complex, the actin nucleator that acts in conjunction with N-WASP, was recruited to sites of pedestal formation in N-WASP-knockout cells. Inactivation of Arp2/3 function blocked pedestal formation on both wild type and N-WASP-knockout cells, indicating that this nucleator is required for all pathways of pedestal formation.
EspFU was unable to directly activate the Arp2/3 complex in vitro, suggesting that, in addition to recruiting and activating N-WASP, EspFU also recruits and activates another regulator of actin assembly that directly or indirectly activates the Arp2/3 complex. Interestingly, a triple amino acid substitution in EspFU that disrupts binding of EspFU to WASP/N-WASP abolished pedestal formation in N-WASP-knockout cells, suggesting that the putative alternate regulator of actin assembly may recognize the same or an overlapping segment of EspFU. This finding is consistent with the hypothesis that one of the WASP-related nucleation promoting factors may participate in this pathway. However, GFP-tagged derivatives of WAVE2, WASH and WHAMM were not efficiently recruited to pedestals in N-WASP-knockout cells (Figure 7); although WAVE2 demonstrated a modest degree of colocalization with bacteria, pedestals were formed efficiently on N-WASP-deficient cells genetically depleted for WAVE2 (Figure S4). Interestingly, KC12/pEspFU did not form pedestals on N-WASPdel/del cells, suggesting that this independently derived N-WASP-deficient cell line [13] may lack the putative alternate actin assembly factor (D.V., J.L., unpub. obs.). One can imagine several scenarios by which N-WASPdel/del cells might aid in the identification of the factor(s) responsible for driving actin polymerization in the absence of N-WASP, a finding that might provide new insights into the normal regulation of actin assembly in mammalian cells.
Notably, pedestal formation by this N-WASP-independent pathway occurred somewhat less efficiently than when both N-WASP and the putative factor are present, because clustering of Tir-EspFU fusion protein generated pedestals 25–30% less efficiently on knockout cells than on wild type cells. In addition, pedestal formation on N-WASP-knockout cells showed a more stringent requirement for multiple EspFU repeats. Whereas clustering of a single EspFU repeat was sufficient to stimulate actin pedestals in the presence of N-WASP [26], [28], pedestal formation was only triggered upon clustering of two or more repeats in the absence of N-WASP. In addition, while two repeats are required to complement an espFU-deficient strain for pedestal formation on wild type FLCs [28], three repeats were required to detect pedestals in NW−/− FLCs, and four or more repeats were required for maximal levels of complementation.
A correlation between the number of EspFU repeats and stimulation of actin assembly, both in vitro and in vivo, has been observed previously [27], [28], [29], [39]. Assuming that at least three repeats are required for N-WASP-independent pedestal formation and that an N-WASP-independent pathway for actin assembly confers a selective advantage in nature, one would predict that the vast majority of EspFU alleles found among E. coli isolates would carry at least three repeats. In fact, of 435 EPEC or EHEC strains in which espFU was detected by PCR, 433 (or >99.5%) of espFU alleles appeared, by length of the PCR product, to encode three or more proline-rich repeats [56]. Therefore, future characterizations of the N-WASP-independent mechanism of actin pedestal formation will enhance our understanding of the role of EspFU in the survival, propagation, or pathogenesis of EHEC.
All EHEC strains used in this study were derivatives of TUV93-0, a Shiga toxin-deficient version of the prototype 0157:H7 strain EDL933 [36]. The parental EPEC strain was the 0127:H6 prototype JPN15/pMAR7. EPEC KC12 [36], EPECΔtir [36], EPECΔtir-eae [36] and EHECΔeae [57], EHECΔtir-eae [57] were described previously. Non-pathogenic laboratory strains of E.coli harboring plasmids encoding EHEC or EPEC intimin (pInt) have also been described [33]. These strains were transformed with a separate plasmid for expression of GFP (a gift from A. Poteete). For beta-lactamase (Bla) translocation assays, plasmids pMB196 (pTirEHEC-Bla) and pMB200 (pEspFU-Bla) were constructed as follows: PCR products encoding Tir and EspFU with their endogenous promoter were amplified with primers flanked with EcoRI and KpnI restriction sites, digested with the appropriate enzymes and cloned into the similarly digested plasmid pMM83 [58]. Plasmids used for expression of GFP-EspFU and HA-Tir-EspFU fusion proteins in mammalian cells were described previously [28]. For dominant negative transfection, the WCA domain of rat N-WASP was amplified by PCR and cloned into the KpnI-EcoRI sites of plasmid pKC425 [59]. All E. coli strains were grown in LB media at 37°C for routine passage. Before infection of mammalian cells, EHEC and EPEC were cultured in DMEM containing 100 mM HEPES pH 7.4, in 5% CO2 to enhance type III secretion. HeLa cells and FLCs were cultured in DMEM containing 10% FBS, 2mM glutamine and 50 µg/ml penicillin/streptomycin. Transfection of plasmid DNA was performed as previously described [52].
Total RNA from N-WASP-knockout cells was isolated using TRIzol reagent (Invitrogen). A first strand cDNA was synthesized using the M-MLV (Moloney Murine Leukemia Virus) Reverse Transcriptase (RT) (Invitrogen). Primers WASP_F (5′-GTGCAGGAGAAGATACAAAAAAGG-3′) and WASP_R (5′-GATCCCAGCCCACGTGGCTGACATG-3′) were used in a 40 cycle PCR reaction to detect WASP cDNA. cDNA from activated B cells, which express abundant WASP, was used as a positive control.
Murine WAVE2 sequence (5′-GAGAAAGCATAGGAAAGAA-3′) was cloned into the Hpa1 and Xho1 sites of plasmid Lentilox 3.7 (pLL 3.7) to generate WAVE2 RNAi stem loops. The virus was packaged into 293T cells using a four-plasmid system and was collected 48 hours after transfection. To knockdown WAVE2, N-WASP-deficient cells were transformed with the lentivirus containing the WAVE2 RNAi stem loop. Knockdown efficiency was evaluated by western blot of the transformed N-WASP KO cells using anti-WAVE2 antibody (Santa Cruz Biotechnology). N-WASP-deficient cells transformed with the empty lentiviral vector were used as control.
Infections of HeLa cells and FLCs with EHEC and EPEC strains were performed as described in earlier work [36]. To evaluate intimin-mediated bacterial attachment in priming-and-challenge assays, FLCs were infected (“primed”) for 3h with EPECΔeae, or EPECΔtir-eae mutant harboring plasmids encoding HA-TirEPEC or HA-TirEHEC, or for 5h with EHECΔeae, or EHECΔtir-eae mutant harboring plasmids encoding HA-TirEPEC or HA-TirEHEC. These strains translocate Tir but do not form pedestals, and were removed from the cell monolayers after gentamicin treatment and washing. The primed cells were then infected (“challenged”) for 1h with non-pathogenic laboratory strains of E.coli harboring plasmids encoding either EHEC or EPEC intimin (pInt) and a plasmid that expresses GFP. A bacterial binding index, defined as the percentage of cells with at least five bound GFP-expressing bacteria, was determined microscopically (Figure 2).
To determine the translocation index of Bla fusions into NW+/+ or NW−/− FLCs, cells were infected for 6 hours with EPEC or EHEC strains harboring Tir-Bla or EspFU-Bla fusions. Infected monolayers were washed with PBS and incubated for 1–2 hours at room temperature after addition of CCF2-AM (Invitrogen) supplemented medium. CCF2-AM treated cells were fixed and analyzed microscopically using a 20× objective. The percentage of blue cells, reflecting effector translocation, was estimated for 10–20 fields per experiment. For studies involving chemical inhibitors of actin assembly, DMSO, wiskostation or cytochalasin D (Sigma) was added 1 hour before infection. To determine the effect of wiskostatin on effector translocation, HeLa cells were infected with EHEC/pEspFU-Bla at a density of 2×107 bacteria/well in DMEM containing either DMSO or 6 µM wiskostatin. Plates were spun at 200 RCF for 5 minutes and then incubated at 37°C in 5% CO2 for 90 minutes. Cells were washed twice with PBS, overlaid with 100 µl of CCF2/AM loading solution in PBS, and then incubated for two hours at room temperature. Plates were transferred to a Synergy 2 microplate reader (BioTek) and excited at 400 nm (10-nm band-pass) and the emission signal was read at 460 nm (40-nm band-pass) and 528 nm (20-nm band-pass). After subtracting out background, the 460/528 nm ratio was calculated to determine the level of effector translocation. Upon treatment with cytochalasin D, HeLa cells were infected with EHEC/pHA-Tir and samples processed for detection of Tir foci and F-actin pedestals by immunofluorescence microscopy, as described below.
Infected cells were fixed in 2.5% paraformaldehyde for 35 minutes and permeabilized with 0.1% Triton-X-100 in PBS as described previously [36]. Bacteria were visualized using DAPI (1 µg/ml; Sigma), and F-actin was detected using 4 U/ml Alexa568-phalloidin (Invitrogen). To visualize HA-Tir derivatives, EspFU-myc, IRTKS, or IRSp53, cells were treated with mouse anti-HA tag mAb HA.11 (1∶500; Covance), mouse anti-myc 9E10 mAb (1∶250; Santa Cruz Biotechnology), or mouse anti-IRTKS mAb (1∶100; Novus Biologicals) prior to treatment with Alexa488-conjugated goat anti-mouse antibody (1∶150; Invitrogen). To visualize the Arp2/3 complex, cells were treated with rabbit anti-Arp3 antibodies (1∶150; gift from R. Isberg, Tufts University) prior to treatment with Alexa488-conjugated goat anti-rabbit antibody (1∶150; Invitrogen). To determine the pedestal formation efficiency of EPEC variants expressing HA-tagged Tir (Figure 3D), the percentage of sites of translocated Tir (HA-Tir foci) that were associated with intense F-actin staining in FLCs were counted. To determine the pedestal efficiency in mammalian cells expressing HA-Tir-EspFU fusions or GFP-EspFU derivatives, which were identified by anti-HA or GFP fluorescence, the percentage of cells harboring at least 5 adherent S. aureus particles (Figure 8A) or KC12 bacteria (Figure 8B) that were associated with actin pedestals was quantified. At least 50 cells were examined per sample. Cells expressing extremely high fluorescence levels of EspFU were refractory to pedestal formation and were not included in these analyses.
In vitro actin polymerization assays were performed using 500 nM EspFU derivative, 2.0 µM actin (7% pyrene-labeled) and 20 nM recombinant Arp2/3 complex, in the presence of 20 nM N-WASP/WIP complex or not, and polymerization was measured as described previously [28].
Cells were collected in PBS plus 2mM EDTA, washed with PBS, and lysed in lysis buffer [50mMTris-HCl, pH 8.0, 150mM NaCl, 1% Triton X-100, 1mM Na3VO4, 1mM PMSF, and 10µg/mL each of aprotinin, leupeptin, and pepstatin (Sigma)] before mixing with sample buffer. Samples were boiled for 10 min, separated by 10% SDS/PAGE, and transferred to PVDF membranes. Membranes were blocked in PBS containing 5% milk before treatment with anti N-WASP, anti-WASP (Santa Cruz Biotechnology), anti-Nck1 (Upstate), anti-Arp3, or anti-tubulin DM1A (Thermo Scientific) antibodies. Following washes, membranes were treated with secondary antibodies and developed [36].
|
10.1371/journal.pgen.1006539 | Genomic Footprints of Selective Sweeps from Metabolic Resistance to Pyrethroids in African Malaria Vectors Are Driven by Scale up of Insecticide-Based Vector Control | Insecticide resistance in mosquito populations threatens recent successes in malaria prevention. Elucidating patterns of genetic structure in malaria vectors to predict the speed and direction of the spread of resistance is essential to get ahead of the ‘resistance curve’ and to avert a public health catastrophe. Here, applying a combination of microsatellite analysis, whole genome sequencing and targeted sequencing of a resistance locus, we elucidated the continent-wide population structure of a major African malaria vector, Anopheles funestus. We identified a major selective sweep in a genomic region controlling cytochrome P450-based metabolic resistance conferring high resistance to pyrethroids. This selective sweep occurred since 2002, likely as a direct consequence of scaled up vector control as revealed by whole genome and fine-scale sequencing of pre- and post-intervention populations. Fine-scaled analysis of the pyrethroid resistance locus revealed that a resistance-associated allele of the cytochrome P450 monooxygenase CYP6P9a has swept through southern Africa to near fixation, in contrast to high polymorphism levels before interventions, conferring high levels of pyrethroid resistance linked to control failure. Population structure analysis revealed a barrier to gene flow between southern Africa and other areas, which may prevent or slow the spread of the southern mechanism of pyrethroid resistance to other regions. By identifying a genetic signature of pyrethroid-based interventions, we have demonstrated the intense selective pressure that control interventions exert on mosquito populations. If this level of selection and spread of resistance continues unabated, our ability to control malaria with current interventions will be compromised.
| Malaria control currently relies heavily on insecticide-based vector control interventions. Unfortunately, resistance to insecticides threatens the continued effectiveness of these measures. Metabolic resistance, caused by increased detoxification of insecticides, presents the greatest threat to vector control, yet it remains unclear how these mechanisms are linked to underlying genetic changes driven by the massive selection pressure from these interventions, such as the widespread use of Long Lasting Insecticide Nets (LLINs) across Africa. Therefore, understanding the direction and speed at which this operationally important form of resistance spreads through mosquito populations is essential if we are to get ahead of the ‘resistance curve’ and avert a public health catastrophe. Here, using microsatellite markers, whole genome sequencing and fine-scale sequencing at a major resistance locus, we elucidated the Africa-wide population structure of Anopheles funestus, a major African malaria vector, and detected a strong selective sweep occurring in a genomic region controlling cytochrome P450-based metabolic pyrethroid resistance in this species. Furthermore, we demonstrated that this selective sweep is driven by the scale-up of insecticide-based malaria control in Africa, highlighting the risk that if this level of selection and spread of resistance continues unabated, our ability to control malaria with current interventions will be compromised.
| Scaling up of malaria prevention and treatment has averted over 660 million cases of malaria since 2000 [1]. The vast majority of this reduction has come from mosquito control with pyrethroid insecticide-based interventions, primarily the use of long lasting insecticide-treated bednets (LLINs) and, to a lesser extent, indoor residual spraying (IRS). Resistance to insecticides in major malaria vectors such as Anopheles funestus threatens the continued success of these interventions. Unless resistance is managed, the massive reduction of malaria transmission from scaling up these interventions could be reversed [2]. A key prerequisite for resistance management is to understand the evolution of insecticide resistance to predict the speed and direction of spread of resistance and provide vital information to implement successful control strategies [3]. There are multiple mechanisms of insecticide resistance, these include changes to insecticide target molecules that render the insecticide unable to bind, behavioural changes leading to the avoidance of insecticide contact, thickening of the insect’s cuticle to prevent the insecticide reaching its target and detoxification of the insecticide before it reaches its target (metabolic resistance). Of these, metabolic resistance has the greatest operational significance [3, 4], yet it remains unclear how mosquito populations exhibiting these mechanisms respond to insecticide-based interventions including LLINs. Selective sweeps associated with target-site resistance have been assessed in mosquito species [5, 6] but no such assessment has been made for metabolic resistance. In the major malaria vector An. funestus, pyrethroid resistance is mainly conferred by metabolic resistance associated with a major quantitative trait locus (QTL) at which two duplicated cytochrome P450 monooxygenases (CYP6P9a and CYP6P9b) are the main resistance genes [7] [8]. The predominance of metabolic resistance in An. funestus makes this species very suitable to assess metabolic resistance-based evolutionary responses of mosquitoes to the massive scale up of pyrethroid-based vector control interventions across Africa.
In the context of increasing reports of insecticide resistance in malaria vectors such as An. funestus across Africa [9–16], it is important to determine the relative contributions to this of gene flow and of the autochthonous appearance of insecticide resistance. Previous studies suggest significant genetic structure among An. funestus populations across Africa [17]. Whether such differences in genetic structure explain the contrasting insecticide resistance patterns seen in this species [9–16] and could help predict the speed and direction of spread of resistance remains to be determined.
Here, using microsatellite analysis, whole genome sequencing and fine-scale sequencing at a resistance locus, we elucidated the Africa-wide population structure of An. funestus and detected a strong selective sweep occurring at a major cytochrome P450-based pyrethroid resistance locus. Moreover, we demonstrated that this selective sweep is driven by the scale-up of insecticide-based malaria control in Africa, highlighting the risk that if this level of selection and spread of resistance continues unabated, our ability to control malaria with current interventions could be compromised.
Analysis of genetic diversity at 11 microsatellites from across the An. funestus genome sampled in six populations from West (Ghana and Benin), Central (Cameroon), East (Uganda) and southern Africa (Malawi and Mozambique) revealed reduced diversity in two microsatellites at the telomeric end of chromosome arm 2R (Fig 1A). These microsatellites; AFUB6, located in a genomic region spanning a pyrethroid resistance QTL (rp1) and FunR, located in the same QTL in the 5' un-translated region of the pyrethroid resistance gene CYP6P9a (Fig 1B), showed few alleles and low heterozygosity (S1 Table).
We hypothesised that this low diversity was due to a selective sweep in this QTL region driven by pyrethroid-based vector control interventions. To further localise the putative sweep, we genotyped 5 additional microsatellites upstream and downstream of FunR and AFUB6 on chromosome 2R. This revealed that the reduced diversity was restricted to the two markers within rp1 (S2 Table). Consistent with our hypothesis, the lowest gene diversity at FunR was seen in the two southern African populations from Malawi and Mozambique (Fig 1A), which also show the highest levels of pyrethroid resistance [4, 18].
Due to the stronger signature of selection observed in the highly pyrethroid resistant populations of southern Africa, we aimed to establish if insecticide-based interventions were driving this selection. Mosquitoes collected in Malawi and Mozambique in 2002, predating the scaling up of malaria control interventions in southern Malawi and southern Mozambique were compared to mosquitoes collected in the same areas in 2009–2010 to determine whether the selective sweep coincided with the scale-up of pyrethroid-based vector control interventions that occurred over this period.
Pooled template sequencing was carried out on two pools of mosquito genomic DNA: one from mosquitoes collected in 2014 and one from mosquitoes collected in 2002 in Malawi (S7 Table). Sequences were aligned to the FUMOZ reference genome (S8 Table) and stringent filtering performed to remove SNPs at the extremes of coverage depth (S9 Table). Analysis of a total of 979,808 variant sites genotyped in both samples detected 3,078 variant sites (on 368 genomic scaffolds) with significantly different allele frequencies between 2002 and 2014 collections. A significant correlation was observed between the number of significant sites plotted against total scaffold length (p<<0.01 for both Pearson’s and Spearman’s tests) (Fig 4A). However, a number of outlier scaffolds appear to be enriched for significant sites. The most extreme of these is scaffold KB669169, with more than 400 significant variants. This scaffold contains the rp1 locus and most of the significant sites are clustered around this region (Fig 4B) with a striking loss of diversity between 2002 and 2014 evident across the locus (Fig 4C and 4D). This valley of reduced variability around rp1 is the typical signature of selective sweeps as previously observed in other pool-seq studies [21, 22]. In no other scaffold was the signature so striking (S7 Fig), which validated the results of our microsatellite and targeted sequencing analyses. The rp1 region of scaffold KB669169 contains a number of sequencing gaps. To ensure this did not affect our conclusions, sequence data were additionally aligned to the sequenced 120kb BAC clone containing rp1 [7]. This confirmed the results observed in the whole genome analysis (Fig 4E; S8 Fig). The 2014 post-intervention sample exhibits a pattern similar to FUMOZ (the pyrethroid resistant laboratory colony) at rp1 with reduced diversity spanning the cytochrome P450 cluster including the duplicated CYP6P9a and CYP6P9b genes, whereas the 2002 pre-intervention sample is highly diverse across the locus.
In order to help maintain the continued success of current insecticide-based malaria control interventions, this study has established the Africa-wide population structure and the full genomic signature of pyrethroid-based interventions in a major malaria-transmitting mosquito providing key evidence of the evolutionary response of mosquito populations to the massive scale up of insecticide-based interventions in Africa.
This study revealed that southern African populations of An. funestus are more genetically differentiated to other populations as they always form a unique cluster compared to other African regions based on both Bayesian analyses and Fst estimates. The population from Uganda appears to be intermediate between southern and West/Central Africa. This result is similar to patterns of genetic structure previously reported for this species [17]. Patterns of genetic structure observed in this study support the contrast in resistance patterns between populations of An. funestus and suggest the presence of barriers of gene flow between populations of this species. The causes of these barriers remain unknown although it could be associated with the absence of An. funestus around the Equatorial belt, or the presence of the Rift Valley which affects the population genetic structure of An. gambiae [23]. Such hypothesis will need to be validated by assessing more populations, e.g. from both sides of the Rift Valley. The patterns of gene flow described here for An. funestus give an indication on the risk and speed of spread of insecticide resistance alleles between these populations. This is further supported by the observation that the 119F resistance allele of the GSTe2 gene conferring DDT resistance probably arose in West or Central Africa, where it is common, but remains absent from southern Africa, despite selection pressure from DDT use [19].
This study identified a major selective sweep on the 2R chromosomal arm, its location coinciding with that of the main pyrethroid-resistance QTL explaining 85% of genetic variance to resistance, and containing key cytochrome P450 genes conferring pyrethroid resistance [7, 24]. Overall, multiple analyses provide evidence for a selective sweep on the 2R chromosome driven by metabolic cytochrome P450-based pyrethroid resistance; (i) Reduced diversity of microsatellites flanking pyrethroid resistance genes on 2R in Malawi and Mozambique. (ii) Reduced genetic diversity of genomic sequences flanking pyrethroid resistance genes on 2R in Malawi and Mozambique. (iii) Reduced genetic diversity of the CYP6P9a gene in Malawi and Mozambique. (iv) No signature of a selective sweep prior to the widespread use of LLINs and IRS in Malawi and Mozambique.
Altogether, this study provides conclusive evidence of the extensive selective sweep acting on cytochrome P450-based metabolic resistance to insecticides in mosquitoes. We conclude that positive selection on the region spanning the rp1 pyrethroid resistance locus in Anopheles funestus has occurred in southern Africa between 2002 and 2009 in response to the increased use of pyrethroid-treated bednets. No major change in agricultural use of pyrethroids has occurred in southern Africa over this period, but LLIN use and, to a lesser extent pyrethroid-based IRS, has been scaled up massively in this period. This highlights the risk of relying on a single insecticide class for vector control and emphasizes the need for novel insecticides and vector control tools to tackle the spread of resistant vector populations.
Anopheles funestus mosquitoes were collected between 2009 and 2010 from the following locations: Chokwe, Mozambique (24° 33’S, 33° 01’E); Chikwawa, Malawi (16° 3’S, 34° 50’E); Tororo, Uganda (0° 45’N, 34° 5’E); Lagdo, Cameroon (9° 05’N, 13° 40’E); Pahou, Benin (6° 23'N, 2° 13'E); and Obuasi, Ghana (6° 12’N, 1° 40’W). Additional collections in southern Africa were made in 2011 in Salima (13° 57’S, 34° 30’E) and Nkhotakota (12° 56’S, 34° 17’E) in Malawi, plus collections from Chikwawa and Mozambique in 2002. Genomic DNA was extracted from whole mosquitoes using the method of Livak [40] or the DNeasy DNA Extraction Kit (Qiagen Inc., Valencia, CA, USA Mosquitoes were identified morphologically [41] and were species-typed using An. funestus sensu stricto-specific PCR primers [42]. Only confirmed An. funestus s.s. mosquitoes were used in this study.
Contrasting resistance profiles have been described for various populations of An. funestus across Africa. For example, the resistance pattern observed in North Cameroon in 2007 (DDT and dieldrin resistance)[16] is different to that of southern Africa (high pyrethroid, DDT and carbamate resistance)[4, 15], East Africa (pyrethroid and DDT resistance but full susceptibility to carbamates)[12] and Ghana (West Africa) (DDT resistance and pyrethroid resistance)[14].
Microsatellite markers were chosen to span the entire genome [17, 43]. Mosquitoes were collected from six countries (Obuasi, Ghana: N = 45, Pahou, Benin: N = 48, Lagdo, Cameroon: N = 48, Tororo, Uganda: N = 48) with two collection time points in Chikwawa, Malawi (2010: N = 48, 2002: N = 48) and Chokwe (2009: N = 48) and Morrumbene (2002: N = 45) Mozambique. 17 microsatellites (both di- and tri-nucleotide repeats) were amplified from genomic DNA using 1.5 μl of reaction Buffer, 0.2 μl of dNTP mix (25 mmol), 0.325 μl of both the forward (included a 19bp tag) and reverse primers, 0.2 μl of Hot Start Taq (Qiagen Inc.), 1 μl of MgCl2 and 1μl of genomic DNA (15ng/ul). Forward and reverse primers are listed in S1 Table. PCR thermocycler conditions were: 5min at 95°C followed by 35 cycles of denaturing at 94°C for 30s, annealing at 58°C for 30s and extension at 72°C for 30s, finishing with an extension step at 72°C for 10min. Fragment sizing was carried out using a Beckman Coulter CEQ8000. Fragment sizes were visualized and recorded using the fragment analysis software on the Beckman Coulter CEQ8000. Micro-Checker version 2.2.3 [44] was used to check for null allele and scoring errors.
Microsatellite data analysis was mainly carried out using Genepop version 4.0.10 [45]. Tests for deviation from Hardy Weinberg Equilibrium (HWE) were carried out for each locus using Genepop option 1.3. The null hypothesis was HWE and the alternative hypothesis a deficit of heterozygotes, and Bonferroni correction for multiple testing used to adjust the 0.05 and 0.01 critical values. The inbreeding coefficient FIS, linkage disequilibrium (LD), log likelihood ratio statistics (G-test) and tables created using Markov chain algorithm of Raymond & Rousset [45] were all preformed. Gene diversity at each microsatellite locus, estimated by 1-Q(inter) where Q(inter) is the homozygosity among individuals, and among populations. Genetic differentiation (FST) were estimated using Genepop [45]. Pairwise Fst values were used to generate neighbour joining trees using the software MEGA 5.2 [46].
Bayesian analysis of population structure was implemented using STRUCTURE version 2.3.4 [47]. Individually based admixture models were used to estimate the ancestral allele source observed in each individual, where the ancestral source population is unknown. A total of 285 individuals from six African countries, genotyped at 16 loci, were analysed for cluster number K = 1–12 (10 replicate runs for each) using a set of 9 putatively neutral loci and comparing 8 2R loci to the 8 non-2R markers. A burn-in period of 50,000 generations and Markov Chain Monte Carlo (MCMC) simulations of 100,000 iterations were used. The admixture model was used as it allows individuals to have mixed ancestry where a fraction (qk) of the genome of an individual comes from an ancestral cluster (where tkqk = 1)[47]. Structure Harvester [48] was used to infer the most likely number of clusters (K) using Evanno’s method [49]. CLUMPP [50] was used to collate the data from all 10 runs for each given K value, for plotting.
Five DNA fragments evenly spaced to span the 120kb BAC, originally isolated using a laboratory colony, FUMOZ [7], upstream and downstream of CYP6P9a were sequenced in order to assess the presence of a selective sweep around this key resistance gene across Africa (Cameroon, Malawi and Mozambique). Primers used to amplify the fragments are listed in S1 Table.
A subset of 10 mosquitoes used for the microsatellite analysis were randomly selected from each collection site for analysis. For BAC fragments, 770-1000bp were sequenced using: 3 μl of 10X KAPA Taq buffer A (KAPA Biosystems), 0.24 μl of 5 U/μl KAPA Taq, 0.24 μl of 25 μM dNTP mix, 1.5 μl of 25μM MgCl2, 1.02 μl each of antisense and sense primers, 2 μl of gDNA (10ng), and 22.48 μl of dH2O. The 30 μl solution underwent a denaturing step at 95°C for 5min, followed by 35 cycles of 94°C for 30s, 57°C for 30s and 72°C for 1min and 30s, followed by a final extension step of 72°C for 10min.
The CYP6P9a gene was amplified from the same gDNA samples used for the rp1 BAC analysis. CYP6P9a was amplified, covering the 5’ UTR, the gene’s two exons and one intron with a total sequence of approximately 2kb using previously published primers and parameters [8]. PCR products were purified using the QIAquick PCR Purification Kit and directly sequenced using Sanger sequencing. Sequences were first analysed for quality then manually assessed for polymorphisms using BioEdit [51]. Sequences were aligned using ClustalW [52].
Heterozygous sites in sequence data were phased using the PHASER algorithm implemented in DnaSP version 5.10 [53]. For all sequences, DnaSP was used to calculate the number of segregating sites (S), the number of haplotypes (h), the nucleotide diversity (π) and the haplotype diversity (Hd). Two tests of neutrality, Tajima’s D and Fu and Li’s D*, were also carried out. For sequences encoding proteins, 1524bp of the CYP6P9a gene, the numbers of synonymous and nonsynonymous polymorphisms and nonsynonymous (KA) and synonymous (KS) polymorphisms per site were calculated. The KST statistic in dnaSP 5.1 [54] was used to estimate the levels of pair-wise genetic differentiation between populations. The statistical significance of the KST* estimates was assessed by permutation of subpopulations identities and re-calculating KST* 10,000 times as implemented in dnaSP5.1.
Hudson, Kreitman and Aguade’s (HKA) and McDonald and Kreitman’s (MK) tests of neutrality were performed using the An. gambiae orthologue of CYP6P9a, CYP6P3 (AGAP002865-PA) as an out-group. For the MK test on the BAC25 fragment, which spans the CYP6AA2 gene in An. funestus, the An. gambiae CYP6AA2 gene (AGAP002862-PA) was used as an out-group.
Maximum likelihood phylogenetic trees where generated for the BAC25 sequences (Tamura’s Model) and for CYP6P9a (Kimura’s 2-parameter model) using MEGA 5.2, with 500 bootstrap replicates [46]. Haplotype networks were determined, based on a 95% connection limit with gaps treated as a fifth state, using TCS [55]. Individual haplotypes were labelled by colour and shape (circle denoting haplotypes unique to only one sequence, squares denote haplotypes containing multiple sequences).
A whole genome scan was performed comparatively between pre- and post-intervention mosquitoes in order to detect all selective sweep signatures associated with the scale-up of insecticide-based interventions. Pooled template whole genome sequencing libraries were generated as follows. Genomic DNA was purified from individual female mosquitoes collected from Chikwawa in Malawi in 2014 and 2002 using the DNeasy Blood and Tissue Kit (QIAgen, Hilden, Germany), following the manufacturer’s instructions and including an RNase treatment step to remove RNA. The gDNA from each mosquito was quantified using the Quant-iT PicoGreen dsDNA Assay Kit (Thermo-Fisher, Waltham, USA) on a FLUOstar Omega Microplate Reader (BMG Labtech, Aylesbury, UK). Equal quantities of gDNA from 40 individuals for each collection were pooled in equal amounts and the pools used to generate Illumina TruSeq Nano DNA fragment libraries (Illumina, San Diego, USA) with an insert size of 350bp. Libraries were sequenced (2x125bp paired-end sequencing) on different lanes of an Illumina HiSeq2500 (Illumina) at the Center for Genomics Research (University of Liverpool, UK), each multiplexed with three other libraries (not used in this study), to produce approximately 50,000,000 read pairs per library.
Raw sequence reads were trimmed of adapter sequence and low quality bases, using cutadapt [56] and sickle [57], and filtered to remove trimmed reads shorter than 10bp. Trimmed reads were aligned to the Anopheles funestus (FUMOZ) reference genome sequence (version Afun1.3) downloaded from VectorBase [58], using bowtie2 [59]. Aligned reads were filtered to remove duplicate reads and those not properly paired (mapped in forward and reverse orientation within 500bp of each other) or with mapping quality <10. Retained reads were used to detect single nucleotide polymorphisms (SNP).
SNP calling was carried out using SNVer [60]. SNPs were filtered to remove those called in regions of in the top or bottom 25% of read coverage depth, to remove artefacts caused by misaligned paralogous sequences. Variant sites with coverage data within the allowed range for both libraries were analysed to identify those with significantly different allele frequencies in each pool. A chi-squared test for a significant difference in allele frequency was applied to each variant site. P-values were corrected for multiple testing using the method of Benjamini and Hochberg [61] and sites with an adjusted p-value less than 0.05 were considered significant. In addition, a rolling mean non-reference allele frequency was calculated and plotted for sets of 101 adjacent sites incremented by one site per step across each scaffold.
The DNA sequences reported in this paper have been deposited in the GenBank database accession numbers: KU168962-199123. The whole genome sequence read data reported in this study were submitted to the European Nucleotide Archive (ENA) under the study accession PRJEB13485 (http://www.ebi.ac.uk/ena/data/view/PRJEB13485) and the sample accessions ERS1115465 and ERS1115466.
|
10.1371/journal.pntd.0003694 | Development of Ss-NIE-1 Recombinant Antigen Based Assays for Immunodiagnosis of Strongyloidiasis | Strongyloides stercoralis is a widely distributed parasite that infects 30 to 100 million people worldwide. In the United States strongyloidiasis is recognized as an important infection in immigrants and refugees. Public health and commercial reference laboratories need a simple and reliable method for diagnosis of strongyloidiasis to identify and treat cases and to prevent transmission. The recognized laboratory test of choice for diagnosis of strongyloidiasis is detection of disease specific antibodies, most commonly using a crude parasite extract for detection of IgG antibodies. Recently, a luciferase tagged recombinant protein of S. stercoralis, Ss-NIE-1, has been used in a luciferase immunoprecipitation system (LIPS) to detect IgG and IgG4 specific antibodies. To promote wider adoption of immunoassays for strongyloidiasis, we used the Ss-NIE-1 recombinant antigen without the luciferase tag and developed ELISA and fluorescent bead (Luminex) assays to detect S. stercoralis specific IgG4. We evaluated the assays using well-characterized sera from persons with or without presumed strongyloidiasis. The sensitivity and specificity of Ss-NIE-1 IgG4 ELISA were 95% and 93%, respectively. For the IgG4 Luminex assay, the sensitivity and specificity were 93% and 95%, respectively. Specific IgG4 antibody decreased after treatment in a manner that was similar to the decrease of specific IgG measured in the crude IgG ELISA. The sensitivities of the Ss-NIE-1 IgG4 ELISA and Luminex assays were comparable to the crude IgG ELISA but with improved specificities. However, the Ss-NIE-1 based assays are not dependent on native parasite materials and can be performed using widely available laboratory equipment. In conclusion, these newly developed Ss-NIE-1 based immunoassays can be readily adopted by public health and commercial reference laboratories for routine screening and clinical diagnosis of S. stercoralis infection in refugees and immigrants in the United States.
| Strongyloidiasis is a neglected tropical disease that affects millions worldwide and needs more attention and better diagnostic methods. Strongyloides stercoralis can undergo an autoinfection cycle and can cause hyperinfection involving the pulmonary and gastrointestinal systems and disseminated infection in other organs. Although endemic areas are mostly developing countries in tropical and subtropical regions with only sporadic transmission in temperate areas, the disease is a threat to developed world populations through immigrants, refugees, travelers, and military personnel. The disease can have catastrophic effects when a patient is immunocompromised or when an infected organ is transplanted into a vulnerable recipient. Due to the threat to public health, the intricate life cycle of S. stercoralis, the need to perform multiple follow-up diagnostics to ensure treatment success, and the necessity to rule out multiple co-endemic parasitic infections, it is imperative to develop new diagnostic assays that are simple and efficient while retaining maximal sensitivity and specificity. In this study, we use a well-known recombinant protein, Ss-NIE-1, to optimize assays using both an ELISA format and a multiplex platform to meet these needs.
| Strongyloides stercoralis, an intestinal nematode that migrates through the skin and lung, is a widely distributed disease that infects 30 to 100 million people worldwide [1]. Unlike other helminthic parasites, S. stercoralis can complete its entire life cycle within a single human host through autoinfection and can cause an asymptomatic chronic infection that may go undetected for decades in immunocompetent hosts [2, 3]. In the United States, S. stercoralis causes more deaths than any other soil-transmitted helminth, with mortality rates as high as 87% in cases of hyper-infection in immunocompromised hosts [3].
The standard diagnosis of strongyloidiasis relies on the detection of larvae in the stool [4], but a single stool sample analysis will identify no more than 70% of positive cases [5]. Due to the low sensitivity of the stool assay, immunodiagnosis using a crude antigen-based enzyme-linked immunosorbent assay (ELISA) has been developed as the laboratory test of choice for clinical diagnosis of strongyloidiasis. The Immunoglobulin G (IgG) ELISA utilizes crude extract prepared from L3 S. stercoralis larvae obtained from infected dogs. Reliance on native parasite materials and the canine infection model are major disadvantages of this test. As a result, a number of recombinant antigen-based ELISAs have recently been developed. Recombinant antigens can be purified easily and can be reproducibly generated in large amounts [6–8]. Antibody detection assays utilizing recombinant protein Ss-NIE-1, a 31-kDa antigen derived from S. stercoralis L3 parasites [8], have reported sensitivities and specificities of 84–98% and 95–100%, respectively, and are comparable in performance to the crude antigen-based ELISA [6–13].
We have incorporated Ss- NIE-1 into a standard ELISA format assay and into a fluorescent bead format assay (Luminex) to detect S. stercoralis-specific Immunoglobulin subtype G4 (IgG4). We have previously used the Luminex system for the simultaneous determination of IgG antibody responses to multiple infections in a single assay run [14–17] and we hope to add the new multiplex bead antibody test to our Neglected Tropical Disease assay panel. We compared the performance of the Ss-NIE-1 recombinant antigen-based ELISA and Luminex bead assays to the published assay performance parameters for the Ss-NIE-1 luciferase immunoprecipitation system (LIPS) based assay [6, 10] and the crude antigen-based IgG ELISA [10]. Because previous research has documented that not all cases of strongyloidiasis are successfully treated with a single course of therapy [18], we also used the fluorescent bead assay to determine if a decrease in antibody was measureable after treatment using a select set of sera.
Although some samples were exhausted during the initial ELISA development and some new samples were added during Luminex assay development, the same sets of sera were used for testing the Ss-NIE-1 ELISA and Ss-NIE-1 Luminex assays and many samples were assayed using both techniques. The sets of human sera used were: (1) samples proven positive for S. stercoralis based on the presence of larvae in the stool or sputum (ELISA N = 258, Luminex N = 175); (2) presumed negative samples from U.S. residents with no history of foreign travel (ELISA N = 182, Luminex N = 207); (3) a convenience panel of samples from patients with various diseases other than S. stercoralis focusing mainly on worm infections and including 63 sera from proven cases of lymphatic filariasis from Haiti (ELISA N = 143, Luminex N = 159) [19]; (4) and sera from patients with S. stercoralis infections, before and after treatment (ELISA N = 48, Luminex N = 25) [18]. All sera were anonymous and were used in accordance with approved human subjects’ protocols.
The development of an IgG4 standard reference curve for the Ss-NIE-1 ELISA was performed as described by Scheel et al [22]. Immunoglobulin G4, human myeloma plasma was purchased and stored frozen in 20 mM phosphate, pH 7.4, with 150 mM NaCl and 0.05% Sodium azide (NaN3) (Athens Research & Technology, Athens, GA). IgG4 was diluted into antigen sensitizing buffer (ASB) (0.05 M Tris/HCl, pH 8.0 + 1 M KCl + 2 mM EDTA) to create standard curve points. IgG4 concentrations were chosen based on previous experiments in standard curve development and adjusted to produce the highest OD value, ~ 2.0.
Checkerboard titrations for antigen concentration, serum dilution, conjugate dilution, and substrate 3, 3’, 5, 5’-Tetramethylbenzidine (TMB) time were carried out on Immulon 2HB Microwell plate (Thermo Scientific, Cat. Number 6506). For optimization of the Ss-NIE-1 ELISA, optimal conditions were chosen based on the signal to noise ratio between defined strong S. stercoralis positive and normal human serum samples. The optimized Ss-NIE-1 ELISA steps are as follows: the micro-well plate was sensitized with 100 μL/well of Ss-NIE-1 antigen at a concentration of 0.3 μg/mL in antigen sensitizing buffer (0.05 M Tris/HCl, pH 8.0 + 1 M KCl + 2 mM EDTA) for 2 hours at room temperature on a plate shaker. Following antigen sensitization, the plate was washed 4 times with PBS/0.3% Tween. The plate was blocked for 30 minutes with 100 μL/well of 10 mM Nickel Chloride (Aldrich, Cat. Number 339350) in PBS/0.3% Tween/5% Instant Nonfat Dry Milk (Nestle, Glendale, CA), and then washed as before. StabilCoat Immunoassay Stabilizer (SurModics, East Prairie, MN) was then added 100 μL to each well and incubated for 30 minutes on a plate shaker at room temperature. After discarding the blocking solution, the plate was dried for 4 hours at 30°C in a vacuum oven chamber. The sensitized and blocked plate was stored at 4°C in sealed aluminum foil with desiccator.
Human serum samples were tested in 100 μL/well at 1:50 dilution in PBS/0.3% Tween/5% Instant Nonfat Dry Milk. Following 30 minutes incubation at room temperature on a plate shaker (speed ~ 800 rpm), the plate was washed 4 times with PBS/0.3% Tween. Proper conjugate concentration of mouse anti-human IgG4 (clone HP6025), affinity purified, horseradish peroxidase labeled (Southern Biotech, Birmingham, AL; Cat. Number 9200–05) was added to each well at 100 μL/well at 1:1,000 dilution in PBS/0.3% Tween and incubated 30 minutes at room temperature on a plate shaker with the plate being washed 4 times following incubation with PBS/0.3% Tween. The substrate used was SureBlue, 3, 3’, 5, 5’-Tetramethylbenzidine (TMB) Microwell Peroxidase Substrate (KPL, Gaithersburg, Maryland). We used 100 μL/well of TMB to develop the plate for 5 minutes, and the reaction was stopped by adding 100 μL 1 N H2SO4 Analyzed ACS Reagent (J.T. Baker, Phillipsburg, NJ). The signal was read at A450nm using a VersaMax Kinetic ELISA Microplate Reader with SoftMax Pro v5.4 Software (Molecular Devices Corporation, Palo Alto, CA).
Data were tabulated and analyzed using Microsoft Excel. Determination of the cut-off value and assay performance was calculated using SAS version 9.0. The concentration of Ss-NIE-1 ELISA was measured in in ng/mL; Luminex results are reported as mean fluorescence intensity minus background blank (MFI). The J-index, a single measurement of assay performance, was calculated as described previously ([25, 26].
The recombinant Ss-NIE-1-His protein expressed at Genscript was successfully used to develop an ELISA (Table 1). Using a cutoff value of 0.80 ng IgG4/mL, the assay correctly identified 245 of 258 parasitologically confirmed strongyloidiasis cases for a sensitivity of 95% (Table 1). The overall specificity of the ELISA was determined to be 93% using a panel of non-endemic US controls and sera from patients with other (mainly parasitic worm) diseases (Tables 1 and 2). Among the 142 donors with defined diseases or parasitic infections shown in Table 2, the specificity was only 86%, but only the two trichuriasis patients were 100% cross-reactive. Although the numbers of samples were also quite small, cross reactivities of ≥ 33% were observed among sera from echinococcosis, gnathostomiasis, and hookworm patients. Table 2 shows the cross-reactivity of the various presumed negative sera. Given the likelihood of polyparasitism among many of these serum donors (i.e., 63 lymphatic filariasis patients from S. stercoralis-endemic Haiti), the 99% specificity observed among US negative controls may be a reasonable upper bound to the value. The J-index, a single measure of assay performance, was 0.88.
Positive sera with low and medium level reactivity were used to measure inter-assay variation as previously described in the Materials and Methods. The inter-assay coefficient of variation was determined to be 22% for the low positive control serum and 10% for the medium positive control serum.
The His tagged Ss-NIE-1 recombinant protein could not be coupled to magnetic beads. Thus, we proceeded using a GST-tagged Ss-NIE-1 protein and successfully coupled the Ss-NIE-1 protein to the MagPlex microspheres. The intra- and inter-assay coefficients of variation were determined to be 4.2% and 13.9%, respectively, for average values of 641 MFI and 585. Using a cutoff value of 8 MFI, the sensitivity and specificity of the IgG4 Luminex assay were 93% and 95% (Table 1), respectively. As with the ELISA described above, the specificity among US negative controls was much higher (99%) than among donors with defined diseases or parasitic infections (91%) (Table 2). High reactivities (≥33%) were only observed among sera from the two amebiasis and three hymenolepiasis donors. The J-index was identical to that of the ELISA at 0.88.
Antibody longevity in subjects infected with strongyloidiasis following treatment with thiabendazole [18] can be seen in Table 3. Peak and median antibody responses decreased over time using both assay formats, but the antibody levels remained above the cut-off point for most of the subjects even 18 months post treatment. Using the Kobayashi criteria of cure, which considers a patient to be cured if the ratio of serological results post-treatment compared to pre-treatment is less than 0.6, 70% of the subjects would be reported as cured 3–6 months following treatment [27].
Strongyloidiasis is an increasingly important health problem in the US among immigrants and refugees. Patients with occult strongyloidiasis are at risk of disease if they become immunosuppressed, and organ donors with unrecognized S. stercoralis infection pose a risk to recipients if their infected organs are transplanted [2]. Identification of the parasite in stool specimens is insensitive, and, because parasitological examination requires collection of multiple stool specimens over 3 days, serological testing would be preferred if available. We elected to develop novel immunodiagnostic assays to meet this need. We employed a well described recombinant protein with proven performance as an immunodiagnostic antigen, the Ss-NIE-1 protein [6–8]. Based on the potential importance of IgG4 antibody responses in filarial infections, we decided to develop methods that detect S. stercoralis-specific IgG4 antibodies. The Ss-NIE-1 IgG4 Luminex bead assay achieved a sensitivity and a specificity comparable to those reported for other strongyloidiasis assays such as the crude antigen ELISA, and 26-kDA ELISA, and Ss-NIE-1 ELISAs and the Ss-NIE-1 LIPS [6, 7, 9, 10, 12, 18]. Compared to the CDC S. stercoralis crude antigen ELISA, the Ss-NIE-1 IgG4 ELISA and the Ss-NIE-1 IgG4 Luminex assay achieved similar sensitivity without compromising specificity. The possible factors contributing to improved specificity could be the use of recombinant antigen, assay optimization, or detection of IgG4 versus IgG. During Ss-NIE-1 ELISA optimization, we found that non-specific antibody binding in the normal human sera could be decreased by adding a pre-blocking step with 10 mM nickel chloride in PBS/0.3% Tween/5% milk. The decrease in background noise allows the assay to have a higher specificity (93%).
ELISA based tests can only be used to detect antibody responses against one antigen of interest at a time. Because the differential diagnosis of S. stercoralis often includes multiple helminths, a submitted serum sample must frequently be tested using several parasite-specific ELISAs to determine the possible cause(s) of infection. xMAP Luminex technology offers an assay platform that can simultaneously detect antibodies to multiple diseases/infections, and multiplex bead-based antibody assays are generally as sensitive as conventional ELISA, have a wide dynamic range, and are highly reproducible from assay to assay [28]. For these reasons, we elected to transfer the Ss-NIE-1 assay to the Luminex platform. In our hands, the Ss-NIE-1 IgG4 Luminex bead assay was slightly less sensitive and slightly more specific than the ELISA, but had comparable overall performance as measured by the J-index.
With the exception of toxocariasis and lymphatic filariasis, our cross-reactivity data must be interpreted with some caution due to the small numbers of samples available for testing. When sera from echinococcosis, hookworm, and trichuriasis patients were tested with Ss-NIE-1 by ELISA, 33% or more of the sera reacted, some quite strongly. A portion of this reactivity likely resulted from ELISA-specific background as many of these same sera did not react with the Ss-NIE-1 antigen in our newly developed Luminex bead assay. Bisoffi et al. [9] found no ELISA or LIPS assay cross-reactions to Ss-NIE-1 among their Echinococcus- or hookworm-infected donors. Although true cross-reactivity between these parasites may exist, polyparasitism cannot be ruled out in the remaining Luminex-positive subjects. An analysis of Ss-NIE-1 LIPS results from a hookworm/ Ascaris lumbricoides/ H. nana/ S. stercoralis co-endemic region of Argentina failed to demonstrate an association between infection with other parasites and an antibody response to the Ss-NIE-1 antigen [10].
The Ss-NIE-1 ELISA and Luminex had 10% cross-reactivity against the subjects with lymphatic filariasis. Again, this problem could be a true cross-reactivity or could also be explained by polyparasitism with soil-transmitted helminthes in Haiti. Unfortunately, we have no data about the presence of other parasitic infections in the individuals from whom these serum samples were obtained [29]. Norsyahida [11] reported that although the use of IgG4 conjugate did decrease the cross-reactivity to filariasis compared to the total IgG responses, some cross-reactivity was found with the IgG4 based assay. However, the Norsyahida group did not use a recombinant antigen [8], and the observed cross-reactivity could be due to the use of crude extract antigen in their assay. Of note, the group mentioned the importance of testing for filariasis in subjects with strongyloidiasis. Such testing could most easily be accomplished on a multiplexing platform such as Luminex that can detect antibodies against filariasis and strongyloidiasis simultaneously.
Decreases in antibody titers post-treatment were observed (as suggested by Satoh [30]) and the percentage of patients who met the serologic definition of cure (≥40% decline in antibody response compared to pre-treatment) were consistent with the previous reports [18]. The only significant difference was that the CDC S. stercoralis crude ELISA results showed that 92%of subjects at >9–18 months had been cured of S. stercoralis infection, but only 71% could be considered as cured based on the SS-NIE-1 ELISA and Luminex assay. We have no definitive explanation for these observed differences, except the fact that the crude ELISA uses a complex antigen versus a single antigen which was used in our studies.
Overall, excellent assays for detecting S. stercoralis specific antibodies have been developed. Because these assays use recombinant proteins, negating the need for native parasite materials these assays can be adopted for use in public health laboratories for refugee screening or in commercial laboratories for diagnosis of clinical strongyloidiasis and for screening possible transplant donors with occult disease. As both ELISA and Luminex based assays performed similarly, studies in low infrastructure, endemic setting could use the ELISA format. Although polyparasitism is a potential problem, with a strong specificity of 93%, cross-reactivity should not be an issue. For a country-wide study to determine the prevalence of strongyloidiasis, a multiplexing capability of Luminex will be more cost efficient.
|
10.1371/journal.pntd.0000363 | Cost-Effectiveness of Chagas Disease Vector Control Strategies in Northwestern Argentina | Control and prevention of Chagas disease rely mostly on residual spraying of insecticides. In Argentina, vector control shifted from a vertical to a fully horizontal strategy based on community participation between 1992 and 2004. The effects of such strategy on Triatoma infestans, the main domestic vector, and on disease transmission have not been assessed.
Based on retrospective (1993–2004) records from the Argentinean Ministry of Health for the Moreno Department, Northwestern Argentina, we performed a cost-effectiveness (CE) analysis and compared the observed CE of the fully horizontal vector control strategy with the expected CE for a vertical or a mixed (i.e., vertical attack phase followed by horizontal surveillance) strategy. Total direct costs (in 2004 US$) of the horizontal and mixed strategies were, respectively, 3.3 and 1.7 times lower than the costs of the vertical strategy, due to reductions in personnel costs. The estimated CE ratios for the vertical, mixed and horizontal strategies were US$132, US$82 and US$45 per averted human case, respectively. When per diems were excluded from the costs (i.e., simulating the decentralization of control activities), the CE of vertical, mixed and horizontal strategies was reduced to US$60, US$42 and US$32 per averted case, respectively.
The mixed strategy would have averted between 1.6 and 4.0 times more human cases than the fully horizontal strategy, and would have been the most cost-effective option to interrupt parasite transmission in the Department. In rural and dispersed areas where waning vertical vector programs cannot accomplish full insecticide coverage, alternative strategies need to be developed. If properly implemented, community participation represents not only the most appealing but also the most cost-effective alternative to accomplish such objectives.
| Despite decreasing rates of prevalence and incidence, Chagas disease remains a serious problem in Latin America, especially for the rural poor. Without vaccines, control and prevention rely mostly on residual spraying of insecticides. Under the aegis of the Southern Cone Initiative, and in agreement with global trends in decentralization of the health systems, in 1992 the Argentinean vector control launched a new vector control program based on community participation. The present study represents the first thorough evaluation of the overall performance of such vector control program and the first comparative assessment of the cost-effectiveness of different vector control strategies in a highly endemic rural area of northwestern Argentina. Supported by results of independent studies, the present work shows that in rural, poor and dispersed areas of the Gran Chaco region, the implementation of a mixed (i.e., vertical attack phase followed by horizontal surveillance) strategy constantly supervised and supported by national or local vector control programs would be the most cost-effective option to interrupt vector-borne transmission of Chagas disease.
| Over the past 15 years, the burden of Chagas disease has been significantly reduced (from ∼30 million human cases in 1990 to ∼9–11 million in 2006) as a consequence of the direct actions promoted by several multinational regional initiatives [1],[2]. The key for such success was the long term implementation of residual insecticide applications to kill triatomine bugs, the screening of blood donors for the presence of Trypanosoma cruzi, and the treatment of infected infants born to infected mothers [3]. In the Southern Cone, disease transmission by the main vector, Triatoma infestans, was interrupted in Uruguay, Chile and Brazil and in southern Argentina [1],[3]. However, limited success was obtained in the Gran Chaco region of northern Argentina, Bolivia and Paraguay (the core of T. infestans distribution) where Chagas disease is still highly prevalent.
Within its 1.3 million km2, the Gran Chaco provides favorable conditions for the development of Chagas and other neglected diseases, including high levels of poverty and social exclusion, low population density, population mostly rural, subsistence economy, and a weak health system [4],[5]. Recent estimations of Chagas disease prevalence in rural populations of this region show values ranging from 25% to 45% in Argentina, 17% to 49% in Bolivia and 14% to 56% in Paraguay [5],[6],[7], much higher than the overall 1.7% estimated for the Southern Cone countries [2]. Furthermore, the lack of effectiveness of pyrethroid insecticides in peridomestic habitats [8],[9] coupled with the presence of sylvatic populations in Bolivia and Argentina [10] [L.A. Ceballos, unpublished results] and the emergence of insecticide resistance in Argentina and Bolivia [11],[12] renders the elimination of T. infestans from the Gran Chaco an elusive challenge.
In Argentina, Chagas disease vector control began in 1962 with the creation of the National Chagas Service (NCS) [13],[14]. Inspired by the old malaria programs, NCS established a vertical and centralized structure based on the application of insecticides (mostly HCH and organophosphates) by qualified personnel. Overall, the program strongly reduced T. infestans infestation and T. cruzi seroprevalence [14],[15], but failed to achieve full coverage of insecticide applications (as late as 1990, many districts in the Gran Chaco have not yet been sprayed) and to interrupt disease transmission. As a consequence of decentralization and reduced health budgets, by the end of 1980's NCS did not have enough resources to maintain a vertical structure nor to warrant the continuity and contiguity of vector control actions.
Aware of these limitations, NCS started researching on alternative vector control strategies [16],[17]. Based on promising field results [16], and under the aegis of the Southern Cone Initiative, in 1992 NCS launched a new vector control program (“Plan Ramón Carrillo”) based on community participation and on the incorporation of appropriate technology [14],[17],[18]. This new strategy was embedded in the Primary Health Care (PHC) system of Argentina, and included the transference of knowledge and practices of control and surveillance of T. infestans to PHC agents, community leaders and rural villagers, who became the first line of T. infestans control [14],[17],[19]. During 1993–2001, 15,500 community leaders sprayed with residual insecticides all of the 961,500 houses in the endemic area during the attack phase; 85% of such houses were under community-based vector surveillance [14]. As a consequence of the vector control activities, five provinces, all outside the Gran Chaco, were certified free of vector-borne transmission by 2001 [19]. However, a different scenario was observed in the Argentinean Gran Chaco, with 5 of its 9 provinces reporting vector-borne transmission of Chagas disease by the year 2000 [14]. An evaluation of the effects of the horizontal strategy at the district-wide level in this region is lacking.
In its conception, the horizontal strategy involved the participation of rural communities only in the surveillance phase [16]. However, budget and personnel constraints forced NCS to implement a fully horizontal strategy (i.e., community participation in both the attack and surveillance phases), with the consequent loss of quality of insecticide applications targeting the prevailing high bug infestation levels. Although the horizontal strategy was originally thought to increase the coverage and frequency of insecticide applications while saving the costs of salaries due to the incorporation of unpaid personnel [16],[20],[21], no direct comparative cost-effectiveness (CE) analysis between the horizontal and the preceding vertical strategy was performed to date.
As a part of a larger project on the eco-epidemiology of Chagas disease in northern Argentina, the objectives of the present study were to assess the effects of the horizontal vector control strategy on the prevalence of infestation by T. infestans and on the occurrence of human acute cases over a 12-year period (1993–2004) in the Moreno department; and to perform a comparative cost-effectiveness analysis between different vector control strategies (fully horizontal, vertical and mixed) in a highly endemic district of the Argentine Chaco.
We analyzed longitudinal data from the NCS for the Moreno Department (centroid at 62° 26′ W, 27° 15′ S), located in the Province of Santiago del Estero, northwestern Argentina (Figure S1). This district was chosen because: a) it is located in the Gran Chaco region; b) historically it presented the highest rates of disease incidence and T. infestans infestation; c) all previous control programs failed to reach full coverage of spraying activities; d) an ongoing long-term longitudinal study [22] developed in five rural communities of the Department allowed us to derive key parameters for the present study.
In 2001, Moreno had approximately 25,000 habitants and 5,439 houses, 54% of which were rural houses belonging to 275 communities [23]; most of the rural communities (75%) consisted of 1–10 houses (Figure S1). Health infrastructure in Moreno is composed of three hospitals located in the three major cities, and approximately 22 PHC centers scattered among rural communities. Rural houses usually have adobe walls and thatched roofs, one or two bedrooms, and a 5–10 m wide veranda in the front. The peridomestic environment includes structures that do not share a roof with the bedrooms, such as storerooms, chicken coops and corrals. Exploitation of forest resources (hardwood for charcoal and logs, hunting), raising goats (and cattle) and subsistence agriculture are the main sources of income of rural villagers.
Under the horizontal strategy launched in 1992, NCS activities focused on: a) training of local villagers in spraying with pyrethroid insecticides and in bug detection activities; b) spraying of rural communities when a human acute case was detected; c) evaluating domiciles and peridomiciles for the presence of T. infestans bugs, and d) the delivery of insecticides, manual compression sprayers, and other supplies to all community leaders. Training workshops for villagers took place at each local school. Workshops provided basic information on Chagas disease epidemiology, and training in insecticide spraying methods and detection of domestic infestation using sensor boxes [18]. At least one resident or PHC agent from every community was selected as a “leader”, and was in charge of storing and distributing the insecticides and sprayers to the villagers who requested them. Each leader was provided with a 5-liter manual compression sprayer, pyrethroid insecticides, and forms to report the spraying activities to NCS personnel on a regular basis [16]. No salary was paid to leaders for their duties.
Insecticide was distributed in small bottles (doses) with the amount of insecticide necessary to fill a 5-liter manual compression sprayer. Villagers were in charge of spraying all domestic and peridomestic structures in their house. After spraying, villagers had to return the manual compression sprayers to the leader, indicating the number of insecticide doses applied, and whether they found T. infestans bugs before, during, or immediately after spraying. The monthly number of sprayed houses, the amount of insecticide and domestic boxes used, and the number of house compounds infested by T. infestans in domiciles and peridomestic habitats were then reported by leaders to NCS. The program scheme included an attack phase with two spraying rounds of every rural house separated by six months. After the first or second spraying rounds a community was considered under surveillance phase. Suspension concentrate (SC) deltamethrin applied at 25 mg/m2 or 20% SC cypermethrin at 125 mg/m2 were the insecticides and doses most commonly used (Table S1).
We performed a generalized CE analysis [24] and compared the observed CE of the fully horizontal vector control strategy with the expected CE of a vertical or a mixed strategy (i.e., vertical attack phase followed by a horizontal surveillance phase). Generalized CE analysis is based on the evaluation of a suite of interventions against the counterfactual of “doing nothing”, thereby providing a unique framework for evaluating and comparing health interventions, and a gateway for identifying opportunities to improve them.
Direct and indirect costs were estimated separately for the attack and surveillance phases. Direct costs included staff (salaries and per-diems), supplies (consumables used for insecticide spraying and vector surveillance) and mobility (fuel and minor vehicle fixes during fieldwork) (see Text S1 for more details). Straight line depreciation was used to reflect the cost of the use of vehicles (10 years) and manual spraying compressors (5 years). Indirect costs included the maintenance of vehicles and the payment of salaries during the time in which personnel was not assigned to field activities. Costs in Argentine pesos were inflated to 2004 US dollars. Costs were only estimated for activities performed in rural communities. Observed costs for the implementation of the fully horizontal strategy were obtained from NCS records, whereas for the vertical and mixed strategies costs were estimated based on the number of houses of Moreno and the personnel and supplies needed for each strategy (see Text S1 for more details).
The number of Chagas disease human cases (symptomatic and asymptomatic) averted by each strategy was chosen as a measure of their effectiveness. Averted cases were estimated as the difference between the number of human cases observed (horizontal) or expected (vertical and mixed) for each strategy and the number of cases expected in the absence of vector control actions. The number of human cases (I) was estimated by applying the following discrete model: I(t) = λt*St ; where λt represents the instantaneous incidence rate and St the number of susceptible individuals in year t. Estimation of averted cases was based on the following assumptions: (1) the acquisition of infection is independent of age and sex.; (2) infection is irreversible; (3) mortality, immigration and emigration are negligible; (4) on average, each year there were 631 live births [23]; (5) congenital transmission is negligible; (6) the susceptible population at year 0 is equivalent to 67.7% of total rural population [22]; (7) in the absence of control actions, the instantaneous incidence rate (λ) is constant in time and space and equivalent to the observed value in rural communities of the Moreno Department in 1992 in the absence of control interventions (4.3 per 100 person-years) [25]; (8) reported symptomatic cases are only 7% of total cases [26].
Cost-effectiveness was estimated as the ratio of direct or indirect costs to the number of averted cases, and expressed as 2004 US dollars per averted case. The strategies evaluated were: 1) fully horizontal; 2) vertical (assuming interruption of disease transmission after the attack phase); 3) mixed with vector-borne transmission (i.e., a scenario with persistent transmission throughout the surveillance phase), and 4) mixed without vector-borne transmission (i.e., the attack phase effectively interrupted vector-borne transmission). A sensitivity analysis was performed to evaluate the relative effects of individual key parameters on the absolute value of the CE ratio.
To assess the long term effects of each strategy, CE ratios were projected over a 25-year period considering a 12.8% inflation rate (the 2002–2006 average for Argentina) [27] and a 3% discounting rate per year. For the vertical strategy, an optimistic scenario in which T. infestans could be eliminated from Moreno after 10 years of sustained vector control actions was considered. For the 25-year projection, total costs (and hence, CE ratios) accrued in the vertical strategy after year 10 were considered zero due to the interruption of NCS visits after the elimination of T. infestans.
Because vector-borne transmission of T. cruzi occurs mostly in rural or peri-urban areas, we excluded the three main cities of Moreno (totaling 2,528 houses) from all the analyses. To compare the prevalence of domestic infestation according to the number of times a community was sprayed, we applied Kruskal-Wallis tests with Dunn contrasts [28]. Multiple lineal regression analysis was applied to test whether spraying coverage (i.e., the percentage of houses in the community that were sprayed in the most recent round) and the number of times the community was sprayed from 1993 to 2000 were significantly associated with the prevalence of domestic infestation in year 2000 (the year with more simultaneous records of community infestation). Statistical analyses were performed using SPSS 14.0 (SPSS Inc., Chicago, IL) and STATA 9.1 (Stata Corp, College Station, TX).
Of the 275 rural communities found in Moreno in 1993, 242 (88%) were sprayed with insecticides at least once during 1993–2004 and were thus considered under vector surveillance. The remaining 33 communities were only visited by NCS personnel for community training, insecticide delivery or vector evaluations, but were never registered as sprayed. Most (79%) of these rural communities had an average of 1 to 4 houses. Only 55 (23%) of the rural communities declared under vector surveillance had two insecticide spray cycles under the attack phase.
Villagers performed 79% of the 5,759 insecticide sprays registered in Moreno (Table S1). The total average number of insecticide doses per household was 7.0 during the attack phase and 5.3 during the surveillance phase (Table S1). A total of 1,793 insecticide fumigant canisters were delivered during 1993–2004, at a rate of 146 and 152 canisters per year in the attack and surveillance phases, respectively. Moreover, a total of 12,982 domestic biosensor boxes were delivered to the communities for vector surveillance (Table S1).
The prevalence of infestation by T. infestans in rural domiciles was 77% in 1993 and decreased to 4% by 1996 (Figure 1A), coinciding with the attack phase. After 1996 domestic infestation fluctuated between 10% and 28%. Peridomestic infestation followed the same trend as domestic infestation, with a decline from 78% to 9% by 1996, and ranging from 22% to 38% during 1997–2004 (Figure 1A). The prevalence of infestation in 25 communities during 1999–2001 was positively correlated with the infestation prevalence assessed by timed manual collections performed by NCS staff in the same communities in 2002 in domiciles (r = 0.45, P<0.02), but not in peridomiciles (r = −0.14, P>0.4). Because leaders' reports likely underestimated peridomestic infestation (with which they have less contact), this information will not be analyzed hereafter.
The initial attack phase apparently produced a downward trend in the reported number of human acute cases, from an average of 10 per year during 1988–1993 to 0 in 1997 (Figure 1B). From 1998 to 2004 the annual number of cases fluctuated between 0 and 3 with no clear trend. All reported cases were symptomatic, and referred for standard treatment at Hospital Independencia in Santiago del Estero's capital.
Domestic infestation prevalence varied significantly with the number of times each community was reported as sprayed with insecticides by rural villagers (Kruskal-Wallis χ2 = 17.9; g.l. = 4; P = 0.003) (Figure 2). In communities reported as never sprayed since 1993, the median domestic infestation prevalence was 100% (Figure 2). In communities registered as sprayed once during 1993–2000, a significant reduction in the median domestic infestation prevalence was observed (Dunn contrast, Q = 3.02; g.l. = 1; P<0.05). The increase of insecticide spraying frequency from one to three was not followed by a significant reduction in domestic infestation prevalence (Q<2.93; P>0.05). However, when communities were reported as sprayed four or more times during 1993–2000, a significant reduction in the median domestic infestation prevalence was observed (Q = 4.08; g.l. = 1; P<0.05).
The prevalence of domestic infestation by T. infestans in 2000 was significantly and positively associated with the time since last insecticide spray (multiple linear regression coefficient, beta = 0.39, t = 3.34, P<0.001) and negatively associated with the coverage of the last insecticide spray (i.e., beta = −0.38, t = −3.28, P≤0.001). On average, communities with a domestic infestation prevalence ≥50% in 2000 were sprayed 5.0 (Standard Deviation, SD, 1.8) years earlier, whereas communities with domestic infestation prevalences ≤50% were sprayed 3.0 (SD, 2.0) years earlier. The mean coverage of the last insecticide spray was 79% (SD, 28%) for communities with domestic infestation prevalence ≥50% in 2000, and 84% (SD, 24%) for communities with domestic infestation prevalence ≤50%.
Total cost (direct and indirect) of the fully horizontal strategy implemented in Moreno during 1993–2004 was $309,426, of which 47% corresponded to indirect costs (Table S2). Indirect costs represented 38% of the total $849,625 estimated for the vertical strategy (Table S3) and 42% of the $582,885 estimated for the mixed strategy (Table S4). Annual direct costs of the horizontal strategy were between 3.4 (attack) and 3.2 (surveillance) times lower than the annual direct costs of the vertical strategy (Figure 3). The cost in personnel (salaries and perdiems) was the cause of the marked difference between strategies. Personnel costs for the vertical strategy were 8.6 (attack) and 5.6 (surveillance) times higher than personnel costs for the horizontal strategy (Figure 3). The total direct cost of spraying a single house during the attack phase was US$ 15 for the horizontal strategy and US$ 38 for the mixed and vertical strategies, whereas the cost of surveying a single house was US$ 17 for the horizontal strategy, US$ 20 for the mixed strategy and US$ 22 for the vertical strategy.
The CE ratio of each strategy (expressed in 2004 US$ per averted case) is presented in Table 1. The lower the coefficient the more cost-effective a strategy (i.e., less money would be needed to avert a single case). Although the fully horizontal strategy showed direct CE ratios 1.9–3.3 times lower than the other strategies, the estimated numbers of human cases were 1.6 to 4.0 times higher than for the remaining strategies (Table 1). When those strategies that may accomplish the interruption of disease transmission (i.e., vertical and mixed WoT) are compared, it can be seen that the strategy Mixed WoT would be the most cost-effective (Table 1).
Figure 4 shows the results of the sensitivity analysis of CE to various parameters. Changes in the incidence rate (lambda) exerted the highest variation of the direct CE ratio for all strategies. At low incidence rates (0.01 cases per year), horizontal and mixed WoT strategies presented similar and lower CE ratios than the vertical strategy (ΔCE = 201), whereas at high incidence rates (0.08 cases per year) the difference in CE ratios between strategies was less marked (ΔCE = 58 between horizontal and vertical). Variations in the acute infection rate or the baseline human infection prevalence did not affect CE values greatly (range of ΔCE between strategies, 66–109), with the horizontal strategy presenting always lower CE ratios than the mixed and vertical strategies (Figure 4). The last panel of Figure 4 shows the variation of CE ratios due to changes in perdiems (from a 50% reduction to the complete elimination), a scenario compatible with the decentralization of vector control activities. The elimination of perdiem expenses reduced the CE ratio of the vertical and mixed strategies to values closer to the CE ratio of the horizontal strategy (ΔCE between horizontal and vertical strategies = 28).
The long-term effectiveness of each strategy was evaluated by projecting the annual direct CE ratios over a 25-yr period (Figure 5). For the vertical strategy, a scenario that assumed the elimination of T. infestans (and the suppression of the costs associated with vector control) after 10 years was evaluated. Figure 5 shows that the mixed and fully horizontal strategies would be more cost-effective than the vertical strategy for up to 16–19 years of interventions, and that the CE of the horizontal and mixed strategies would converge after 21 years of interventions.
As with other vector-borne diseases, the incorporation of community participation in Chagas' disease control and prevention evolved in response to the failure of some vertical programs to achieve their main objectives (originally, vector elimination and interruption of disease transmission) borne, in part, by the acute limitations in personnel and financial support of the health system [18],[20],[21]. The present study represents the first thorough evaluation of the overall performance of a horizontal Chagas' disease vector control program and the first comparative assessment of the CE of different vector control strategies in a highly endemic rural area of Argentina. The results derived from our work may help NCS and other vector control agencies to better plan and design cost-effective control interventions against Chagas' disease vectors.
To achieve significant levels of vector control in endemic areas, Chagas disease control actions need to be sustained over time [22],[29],[30]. In many Latin American countries, the current scenario of partial decentralization of health services, increased poverty, lack of political interest and declining funding for vector control activities represent a serious challenge for the persistence of vertical control strategies [29]. Furthermore, in rural and dispersed areas where waning vertical vector programs cannot accomplish full coverage, alternative strategies need to be developed. The incorporation of participatory approaches against vector borne diseases not only has proven to be cost-effective but also important for the sustainability of control programs [20],[21],[22],[30],[31]. Our analysis shows that the implementation of a mixed strategy would have averted between 1.6 and 4.0 times more human cases than the fully horizontal strategy and, given the realities observed in the ground, would have been the most cost-effective option to interrupt parasite transmission. If properly implemented, community participation represents not only the most appealing but also the most cost-effective alternative to control Chagas disease vectors in resource-constrained settings.
When CE ratio projections were compared, it was clear that the main difference between strategies arises with the potential elimination of T. infestans, since vector elimination from a defined region is associated with a significant reduction or even suppression of operational budgets [32]. Although initially it was assumed that 10 years of vector control actions would be enough to accomplish the regional elimination of T. infestans from the Southern Cone [33], the eco-epidemiologic reality observed in the Gran Chaco region challenges the feasibility of such assertion [5],[22],[29]. The impossibility of accomplishing the regional elimination of T. infestans would have a significant effect in the long-term costs of vector control actions since it would be necessary to maintain a sustained and indefinite surveillance phase to prevent domestic reinfestation by T. infestans and interrupt vector-borne transmission of T. cruzi.
The success of participatory approaches against tropical diseases is strongly dependent on sustained and continuous collaboration and articulation between external agencies, governments, and communities [30]. In Moreno, such coordination occurred during the attack phase (evidenced by the significant decrease in bug infestation and disease transmission) but not during the surveillance phase. In the latter period, the nearly absence of insecticide sprays and the gradual increase in the prevalence of T. infestans infestation were determined by the shortage of insecticide purchases at the central level and by the shift of personnel from the Chagas control program to the recently established dengue control program. Shortage of insecticides, spare parts for compression sprayers and absence of NCS personnel in the field were probably the main obstacles for villagers to continue control activities during the surveillance phase. It is not surprising that new human acute cases of Chagas disease were reported starting in 1998. However, when a mixed control strategy coordinated and supervised by NCS is implemented, T. infestans infestation can be significantly reduced and vector-borne transmission of T. cruzi successfully interrupted [16],[22],[34],[35],[36].
One of the direct benefits of the inclusion of rural communities in vector control activities is the offset of the high personnel costs associated with vertical, centralized strategies [20],[21]. In Moreno, the implementation of fully horizontal or mixed strategies represented a 1.6–3.5-fold reduction of total direct costs in comparison to a vertical strategy. Such reduction in personnel costs in horizontal strategies, however, came associated with an increase in opportunity costs, because villagers and PHC agents had to divert their available time to control and prevention activities. Given the difficulty to estimate the time villagers devoted to control and surveillance activities, opportunity costs were not included in our cost estimates. Indirect costs represented a significant component of the total cost of each strategy (range, 38–47%), with personnel cost being the most important component. Such high costs were the consequence of NCS centralized structure, since field technicians remained stationed at their central base after long distance travel to the field, devoting their time to activities other than vector control. As shown by the sensitivity analysis, decentralization of NCS structure would be a viable alternative to reduce vector control costs, since perdiem expenses would be sharply reduced and personnel time devoted to vector control, community education and supervision increased.
The measure of effectiveness chosen for the present study allowed the estimation of the cost of averting a single vector-borne Chagas disease human case. However, other measures like the reductions of bug infestation levels, disability adjusted life-years (DALYs) or the quality-adjusted life-years (QALYs) have been proposed for evaluating the effectiveness of control programs [37]. Although vector control actions would have a direct effect on domestic infestation levels by T. infestans, such measure of effectiveness was not used in the present study because vector-borne transmission of T. cruzi seems to occur at even low bug densities [25],[38]. In addition, DALYs and QALYs were not used because of their known underestimation of the disability weight for chronic parasitic diseases [39],[40], and lack of relevant data (e.g., age-adjusted infection prevalence and mortality rates for each infection phase) to parameterize them.
As most (79%) of the insecticide sprays in Moreno were conducted by villagers, the data herein presented show how effective such sprays were on bug infestation and disease transmission. The prevalence of domestic infestation in communities with an active surveillance and 4 spraying rounds or more during 1993–2000 was 10%, indicating that control actions performed by villagers were sufficient to maintain low infestations but not to eliminate T. infestans from domiciles. This may be expected from the observed lower effectiveness of insecticide sprays performed by villagers rather than by NCS technicians, and the resulting higher reinfestation rates. As part of an insecticide trial in 400 houses of Moreno department during 2002–2005, we surveyed heavily infested communities that had been last visited by NCS 4–5 years earlier and found that many villagers did not spray their houses correctly; did not take all the furniture and other items out of the domicile before spraying; changed the dilution of the insecticide to make it last more; and used sprayers in inadequate conditions. Because training workshops had occurred almost 10 years before, most of the young people did not know how to properly spray a house. This demonstrates that community participation cannot be assumed, but has to be systematically supported and promoted [31].
Unit costs of house spraying with pyrethroid insecticides in Moreno were within the cost range estimated for other areas in the Americas, with insecticide costs being a variable (albeit important) budget component of the vector control programs (Table 2). Such costs were much lower and therefore more affordable than the 200–2,000 US$ range estimated for housing improvements [41]. The dependence on residual insecticides for the suppression of disease transmission represents an additional burden for Chagas disease vector control programs because insecticide purchases are negotiated at international market prices. The integration of disease programs (i.e., Chagas and malaria where both diseases overlap) and, particularly, the international call for a significant reduction in insecticide prices allocated for vector-borne disease prevention in developing countries represent, in our opinion, some of the integrated, inter-programmatic, inter-sectoral actions [42] needed for reducing the burden of Chagas disease in the Americas.
|
10.1371/journal.pgen.1007929 | A systematic genetic screen identifies essential factors involved in nuclear size control | Nuclear size correlates with cell size, but the mechanism by which this scaling is achieved is not known. Here we screen fission yeast gene deletion mutants to identify essential factors involved in this process. Our screen has identified 25 essential factors that alter nuclear size, and our analysis has implicated RNA processing and LINC complexes in nuclear size control. This study has revealed lower and more extreme higher nuclear size phenotypes and has identified global cellular processes and specific structural nuclear components important for nuclear size control.
| As cells grow and divide, the size of the nucleus is generally maintained as a fixed proportion of cell size. The mechanism by which this nuclear/cytoplasmic ratio is maintained is unclear. Previous studies have suggested that essential gene products may be important for nuclear size control. Therefore, we have exploited the genetic tractability of fission yeast to carry out a systematic genetic screen of deleted essential genes to identify those with aberrant nuclear size phenotypes. Our study has revealed 25 novel genes that influence nuclear size and our bioinformatic analyses have implicated both RNA processing and protein complexes connecting nuclear chromatin to the cytoskeleton in nuclear size control. Our work sheds light on the biological processes that contribute to nuclear size control in fission yeast contributing to our mechanistic understanding of nuclear scaling, a biological phenomenon that is conserved through evolution.
| Study of sea urchin embryos led Hertwig more than a century ago to the Kern-Plasma-relation theory, which proposes that the ratio between nuclear size and cell size is constant [1]. A constant ratio between nuclear and cell size has since been reported in many cell types from unicellular yeasts and Tetrahymena to cells of multicellular animals and plants [2–11]. In multicellular organisms, the nucleocytoplasmic ratio varies between cell types, but is generally restricted to a narrow range for cells of each type [10].
There are several lines of evidence that nuclear volume is determined by cell volume, or a factor related to it, and not directly by DNA content. Nuclear volume correlates with cell volume across a wide range of cell sizes in both budding [6] and fission [7] yeasts. In fission yeast, genetic mutations and nutritional states were used to generate cells displaying a 35-fold range of cell volumes; the nuclear volume to cell volume (N/C) ratio was observed to be constant across this range [7]. In small scale pilot experiments, nuclear volume increased steadily with cell volume during the cell cycle, maintaining a constant N/C ratio throughout interphase; no sudden increase in nuclear size accompanied DNA replication in S phase [6, 7]. Gradual nuclear growth during interphase was also observed in HeLa cells [12]. These data indicate that nuclear volume is not directly determined by DNA content. Even a 16-fold increase in DNA content was not sufficient to alter the N/C ratio of fission yeast cells [7]. Furthermore, in situations where DNA content remained constant, nuclear volume responded to changes in cell volume; transfer of the nucleus of a hen erythrocyte into the cytoplasm of a larger HeLa cell resulted in an increase in nuclear volume [13], as did transfer of HeLa cell nuclei into the cytoplasm of a larger Xenopus oocyte [14]. Nuclear volume was observed to increase concomitantly with cell volume following treatment of murine hepatocytes with c-Myc [15], and to scale with cell volume during the reductive divisions of post-16 cell stage C. elegans embryogenesis [9]. However, the mechanism by which cell volume influences nuclear volume has not been established. Studies of centrifuged sea snail embryos and multinucleate yeast cells have demonstrated that diffusible cytoplasmic factors, rather than structural constraints of cell volume, influence nuclear size [5, 7]. Nuclear import of lamins has been implicated in nuclear size control in Xenopus and human cells [8, 16], but yeasts, which lack lamins, also display nuclear scaling [6, 7], so there must be other factors central to nuclear size control.
The fission yeast Schizosaccharomyces pombe is genetically tractable, simply shaped and undergoes a closed mitosis, making it a useful system in which to probe the mechanism of nuclear size control in vivo. A screen of non-essential S. pombe gene deletion mutants implicated both bulk nucleocytoplasmic transport and regulation of lipid biosynthesis in nuclear size control [17]. However, no mutants with a low N/C ratio phenotype were identified, leading us to hypothesise that such a phenotype may be sufficiently deleterious to be lethal, and therefore key nuclear size control regulators may be encoded by essential genes [17]. Therefore, to identify novel candidate factors required for the control of nuclear size during interphase, we have carried out a genetic screen of S. pombe essential gene deletion mutants for those exhibiting aberrant N/C ratios. This is the first reported systematic study of the role of products of essential genes in nuclear size control and has uncovered both low N/C ratio and severely aberrant N/C ratio deletion phenotypes.
We screened essential gene deletion mutants from a near genome-wide heterozygous gene deletion collection that represents 99% of S. pombe open reading frames [18, 19]. Following sporulation of the heterozygous diploids, haploid deletion mutants of 427 of the S. pombe essential genes divide a limited number of times following germination to form microcolonies of 4–20 cells (Fig 1A and 1B) allowing their N/C ratio phenotype to be assessed. In three stages, we visually screened these haploid microcolony-forming essential gene deletion mutants for those that exhibit aberrant N/C ratios (Fig 1C, Materials and Methods). Nuclear size was monitored, initially using a lipophilic dye and subsequently the nuclear envelope (NE) marker Ish1-yEGFP (Fig 1D, 1E and 1F). In the third stage of the screen cell size and nuclear size were measured, using brightfield and Ish1-yEGFP fluorescence images respectively, for 50 cells of each strain, and the N/C ratio was calculated (Materials and Methods). This yielded reliable estimates of the N/C ratio of each strain population; for example the wild type population mean was 0.05 ± 0.0026 (95% confidence interval) (Fig 2A). Note that the introduction of an additional copy of Ish1 into cells for tertiary screening influences the N/C ratio (see Materials and Methods). This methodology provides an approach to systematically probe essential gene deletion phenotypes. However, it should be noted that the cells will eventually die as the microcolonies do not proceed to form colonies, and so cells may not be in a steady state equilibrium. The eventual loss of viability in microcolony cells could be due to the reduced concentration of maternal gene product diluted by sequential cell divisions, or to an accumulating stochastic loss of viability due to the gene deletion.
Of the 60 mutants screened with Ish1-yEGFP, 25 were identified as aberrant N/C ratio candidates (Fig 2A, S1 Table). Our screen revealed both low N/C ratio phenotypes, and high N/C ratio phenotypes more extreme than those previously identified. Seventeen of the candidates identified displayed an N/C ratio higher, and 8 lower, than wild type cells. Representative images of the most extreme high and low N/C ratio candidates and wild type cells are shown in Fig 2B, and of all N/C ratio mutant candidates in S1 Appendix. Cell length, cell width, cell volume and nucleus volume measurements for all 25 candidates, and for the wild type control, are shown in Table 1, and cell volume, nucleus volume and N/C ratio measurements are displayed in Fig 3. The most extreme low N/C ratio observed was 42% smaller than the wild type value (pcm1Δ, lacking the gene encoding the P-TEFb-cap methyltransferase) and the most extreme high N/C ratio was 62% greater than wild type (nup107Δ, lacking the gene encoding nucleoporin Nup107). In the pcm1Δ mutant, average cell volume was larger than that of wild type cells, but both the enlarged cells and the smallest cells in the population (which had approximately wild type cell volumes) displayed diminished N/C ratios. Two mutants with altered N/C ratio were observed to undergo asymmetric nuclear division (Fig 4A). These were kms2Δ (lacking the gene encoding linker of nucleoskeleton and cytoskeleton (LINC) complex component Kms2), for which mitotic defects have previously been described [20], and nup107Δ. Mitotic defects have been reported for mutants of other components of the Nup107-120 complex [21]. In addition to N/C ratio changes, we also identified six deletion mutants with nuclear shape defects; for example nuclear tethers thought to be indicative of defective microtubule-spindle pole body attachment [22] were observed in spc24Δ and mis17Δ cells which carry deletions of kinetochore components (Fig 4B).
We assessed the relationships between cellular and nuclear volume and N/C ratio in our screen candidates (S1A and S1B Fig). Mean nuclear volumes ranged from 2.22 μm3 to 14.9 μm3, compared to the wild type of 4.83 μm3. Mean cellular volumes were mostly between 90 μm3 and 200 μm3, with extremes at 40.2 μm3 and 275 μm3, compared to 98.6 μm3 in wild type. All 8 low N/C ratio mutants displayed mean cellular volumes higher than that of the wild type strain, ranging from 121 μm3 to 275 μm3. The 17 high N/C ratio mutants displayed mean cellular volumes both greater and smaller than that of wild type, ranging from 40.2 μm3 to 224 μm3. We analysed the cellular volume and N/C ratio measurements for individual cells within the populations. For all except two of the high N/C ratio and low N/C ratio strains, there was a weak negative correlation between cellular volume and N/C ratio, as is the case in wild type cells, however the gradient of the regression line of this did not correlate with mean N/C ratio of the population (S1C Fig). When only cells with cellular volumes within two standard deviations of the wild type mean were considered, despite lower n values, 21 of the 25 N/C ratio mutants identified by our screen still displayed significantly aberrant N/C ratios (S2 Table). Therefore, the N/C ratio mutants we identified are not a consequence of aberrantly sized cells with wild type sized nuclei.
If vacuolar size was altered in N/C ratio mutants then cytoplasmic volume may be less directly related to cell volume than in wild type cells, but no changes in the vacuoles of the mutants were observed. The vacuoles in wild type cells and the most extreme N/C ratio mutants, visualised using the vacuole staining dye FM 4–64, are shown in S2 Fig. No differences were observed in the N/C ratio mutants compared with wild type cells.
To determine whether specific categories of genes are enriched in the genes deleted in nuclear size mutants identified in S. pombe, we carried out gene ontology (GO) enrichment analysis (Table 2) on both those identified in this study (Table 1) and also our previous non-essential genes screen [17]. Relative to all genes screened, the 33 nuclear size genes are enriched for genes in the GO biological process (GOBP) categories mRNA metabolic process, mRNA processing, RNA processing and gene expression. The greatest enrichment (8.64-fold) was observed for mRNA processing genes. The GO cellular component (GOCC) categories LINC complex, Nem1-Spo7 phosphatase complex, intracellular ribonucleoprotein complex, ribonucleoprotein complex, nuclear part, macromolecular complex and protein complex categories were also enriched.
Gene interactions were identified by network analysis [25] (Fig 5). Two components of the Dss1-Mlo3 mRNA export complex and two components of the Nem1-Spo7 phosphatase involved in lipid biogenesis, were previously identified [17]. Also identified were another interactor of Nem1, protein kinase Prp4, which is reported to negatively genetically interact with Nem1, and Nup107, a nucleoporin reported to physically interact with Dss1.
Two components of the LINC complex, Kms2 and Sad1, Tfb4, of the TFIIH complex, and Pcm1, of the P-TEFb-cap methyltransferase complex, were identified. These latter proteins both have roles in transcription, acting as part of the DNA-directed RNA polymerase II holoenzyme. Deletion mutants of the nuclear proteasome tether Cut8 and anaphase promoting complex (APC) coactivator Slp1 both have aberrant nuclear size and interact genetically, displaying synthetic lethality [26]. Four spliceosomal proteins Smn1, Uaf2, Smg1, Prp38 and Msl1 were also identified.
The most extreme low N/C ratio mutant was a mutant with a deletion of the gene encoding Pcm1, displaying an N/C ratio 58% of the wild type value. The methyltransferase Pcm1 forms the P-TEFb-methyltransferase complex, with cyclin dependent kinase (CDK) Cdk9 and cyclin Pch1, required for 7-methylguanosine capping of mRNA [27]. A C-terminal truncation of Cdk9 (Cdk9ΔC) retains kinase activity but loses the ability to bind to Pcm1; lack of this interaction reduces recruitment of both Cdk9 and Pcm1 to chromatin [28]. We hypothesised that if Pcm1’s role in the P-TEFb-methyltransferase complex is required for appropriate nuclear size control, then Cdk9ΔC cells should, like pcm1Δ cells, display aberrant nuclear size. We observed an N/C ratio significantly lower than that of wild type cells in Cdk9ΔC cells (Fig 6A and 6B). A Cdk9 T loop mutant, Cdk9T212A, in which Cdk9 activation by CDK-activating kinase (CAK) is perturbed [28] also displayed an N/C ratio lower than wild type cells (Fig 6A and 6B), suggesting that the recruitment of Cdk9 to chromatin by Pcm1 and Cdk9 activity are important for nuclear size control.
Here, we describe a systematic genetic screen of S. pombe essential genes for those with deletion mutants that display interphase nuclear size defects and so encode factors that may have roles in nuclear size control. This screen identified 25 potential nuclear size mutants, including 8 with low N/C ratios. This is the first report of mutants displaying low interphase N/C ratios. A previously reported genetic screen of S. pombe non-essential genes identified no low N/C ratios [17]. This essential genes screen also identified mutants with high N/C ratio phenotypes, and mutants with altered nuclear shape.
Nuclear size mutants were enriched for genes encoding products with roles in RNA processing, and gene expression more generally. As processing of mRNAs into mRNPs is required for their nuclear export [29], it is possible that perturbed RNA processing could lead to the nuclear accumulation of defective mRNA transcripts, which in turn could lead to N/C ratio alteration by bulk effects analogous to those observed in the previously reported mRNA export defective mutant rae1-167 that displays an enlarged N/C ratio [17]. Alternatively, accumulation of the proteins themselves may influence the N/C ratio; proteins with these functions are often found as part of large complexes comprised of many protein and RNA components. Perturbing the stoichiometry or localisation of a large complex of this kind could influence nucleocytoplasmic transport by affecting the distribution of transport factors between the nucleus and cytoplasm and thus preventing the nucleocytoplasmic transport of other proteins. A further possibility is that altering gene expression or mRNA processing influences expression level of specific RNAs and proteins important for N/C ratio control. In mutants with low N/C ratio, the level of expression of genes encoding proteins required for lipid synthesis may be reduced, limiting nuclear size.
Nuclear size mutant gene products were enriched for proteins localised to the nucleus and ribonucleoprotein complexes. The largest group of interactors was a group of spliceosomal proteins suggesting that this ribonucleoprotein complex, or the appropriate regulation of its synthesis or nucleocytoplasmic transport, is important for nuclear size control. Perturbation of mRNA synthesis or mRNA processing will influence the number of ribosomes which could be related to the monitoring of cell size and thus alter the N/C ratio. The strain identified with the lowest N/C ratio was pcm1Δ. Pcm1 forms part of the P-TEFb-methyltransferase complex, another example of a ribonucleoprotein, which is required for 5’ capping of mRNA. Mutants of Cdk9, another component of this complex, also displayed a low N/C ratio suggesting that activity of the P-TEFb-methyltransferase complex is also important for nuclear size control.
Both GO enrichment and network analysis implicated the LINC complex and the Nem1-Spo7 phosphatase complex in nuclear size control. The role of the Nem1-Spo7 complex in the regulation of lipid synthesis for nuclear size control has been described previously [17]. LINC complexes are conserved protein complexes that bridge the nuclear envelope, connecting nuclear chromatin to the cytoskeleton [30]. In S. pombe, the KASH domain containing integral outer nuclear membrane protein Kms2 and the SUN domain containing integral inner nuclear membrane protein Sad1 form these bridging complexes [20]. Both kms2Δ and sad1Δ were identified as N/C ratio mutant candidates by genetic screening. It has been suggested that LINC complexes buffer forces on the nuclear envelope preserving nuclear morphology [20], so they may constrain nuclear expansion. This possibility is supported by data from mammalian cells which contain multiple KASH domain containing proteins. Four of them are nesprins consisting of a N-terminal actin binding domain (ABD) separated from a C-terminal transmembrane KASH domain by an extended domain of spectrin repeats, and these are thought to form a filamentous network on the cytoplasmic face of the nuclear envelope [31]. In HaCaT cells, disruption of these interchain interactions by overexpression of the ABD of Nesprin-2 increased nuclear size and expression of reduced length Nesprin-2 decreased nuclear size, demonstrating that the interactions between KASH domain proteins are important for nuclear size control in these cells [31].
Though not the main focus of our study, the screening process also identified deletion mutants with previously unreported nuclear shape defects and asymmetric nuclear division phenotypes. Six mutants with nuclear shape defects were identified; among these were three kinetochore proteins, Ndc80, Spc24 and Mis17, and the LINC complex protein Kms2, supporting a proposed model where LINC complexes buffer inward forces from microtubules on the nuclear envelope [20]. Two mutants were observed to display asymmetric nuclear division, kms2Δ, in which mitotic defects were previously described [20], and a strain carrying a deletion of the gene encoding the nucleoporin Nup107. Mitotic defects have previously been reported in mutants of other components of the Nup107-120 complex, as have defects in Ran nuclear transport [21]; either of these perturbations could lead to asymmetric nuclear division.
This genetic screen of fission yeast has uncovered low and more extreme high N/C ratio phenotypes, and revealed novel roles for essential factors in nuclear size control. Bioinformatic analyses have implicated RNA processing and LINC complexes in the control. This suggests that both the regulation of global cellular processes, balancing levels of nucleoplasmic transcription and cytoplasmic translation, and regulation of the levels of specific structural components of the nucleus are important for nuclear size control.
S. pombe media and methods used were as described previously [32]; cells were grown in YE4S unless otherwise indicated. S. pombe strains used in this study are listed in S3 Table. Gene tagging was performed using PCR-based methods described previously [33] and verified by colony PCR. The following alleles were previously reported: cdk9ΔC [28] and cdk9T212A [28].
427 strains from a near genome-wide heterozygous gene deletion collection carrying deletions of essential genes and reported to form haploid microcolonies [18, 19] were screened in a primary screen for aberrant N/C ratio mutants.
Heterozygous diploids were patched from glycerol stocks onto YE agar + 250 mg/L uracil + 250 mg/L leucine + 100 μg/ml G418 and incubated at 32°C for 3 days then grown into stationary phase in 200 μl YE + 250 mg/L uracil + 250 mg/L leucine + 250 mg/L adenine liquid media in 96-well plates. Cells were then transformed with pON177 sporulation plasmid (mat1-M mating cassette marked with ura4) (gift from Olaf Nielsen). Cells were harvested by centrifugation (2,500 rpm, 3 mins) and resuspended in 20 μl water. 20 μg boiled salmon sperm DNA, 1.5 μg pON177 DNA and 200 μl PEG-LiAC-TE (40% (w/v) PEG4000, 0.1 M LiAc, 1 mM EDTA, 10 mM TRIS at pH 7.5) were added then cells were incubated at 25°C for 3 days. Cells were harvested by centrifugation (2,500 rpm, 3 mins), washed and then transformants selected on EMM agar + 250 mg/L leucine. To induce sporulation, transformants were inoculated into EMM (without NH4Cl) + 1 mg/ml glutamic acid + 250 mg/L adenine + 250 mg/L leucine and incubated at 25°C for 3 days. Spores were harvested by treatment with 1:50 β-glucuronidase from Helix pomatia and stored in water at 4°C.
For visual screening, spores were resuspended in YE4S and incubated at 32°C for 22–24 hours. 5 μg/ml DiOC6 was added to 50 μl of cells then incubated at room temperature for 1 minute before visualisation using a Zeiss Axioskop 40 microscope with a 63X, 1.4 NA objective. 152 potential aberrant N/C ratio mutants were identified and subject to a secondary round of screening as described for the primary screen. The previously described 5300 control strain in which the KanR deletion cassette is inserted in pseudogene SPAC212.05c was used as a control for this and subsequent screening stages [34]. 60 potential N/C aberrant ratio mutants, identified in both primary and secondary screens, were carried forward for tertiary screening.
For tertiary screening, heterozygous diploid gene deletions were first transformed with a NE marker construct designed to integrate into the KanMX4 deletion cassette. NE protein Ish1 was tagged with yEGFP (marked with HygR selection marker with pAgTEF promoter and tCYC1 terminator) at its native locus in wild type cells using the pYM25 plasmid and the PCR-based transformation method previously described [35]. gDNA was isolated from transformants using the MasterPure Yeast DNA Purification Kit (Epicentre) and manufacturer’s instructions and a DNA fragment (Fragment A) with the ish1 open reading frame (ORF) and 1 kb upstream sequence, yEGFP and HygR marker was amplified by PCR. Two synthetic DNA fragments, each with 400 bp homology to KanMX4 (Integrated DNA technologies (gBlocks)), with Fragment A between them, were cloned into a linearised pUC19 vector backbone using a Gibson Assembly Cloning Kit (NEB). The plasmid was digested with KpnI-HF (NEB) and SphI-HF (NEB) to yield the NE marker construct which was transformed into heterozygous deletion mutants as described for pON177 transformation above. Integration of the NE marker construct into the KanMX4 deletion cassette was confirmed by colony PCR and by microscopy; heterozygous diploids were sporulated and asci confirmed to contain two spores containing Ish1-yEGFP and two wild type spores. We confirmed that the N/C ratio of cells with Ish1-yEGFP integrated at the native ish1 locus is not significantly different from that of cells containing the alternative NE marker Cut11-GFP integrated at its native locus (Fig 1D and 1E).
Transformation with pON177, sporulation and harvesting of spores was carried out as described for the primary screen. As in the primary screen, spores were resuspended in YE4S and incubated at 32°C for 22–24 hours then imaged. Cells were imaged using a DeltaVision Elite microscope (Applied Precision). N/C ratio of each strain was measured.
23 mutants with a mean N/C ratio more than one standard deviation greater than or smaller than the wild type population mean and significantly different from the wild type population mean with a p value ≤0.002, and two further mutants with a mean N/C ratio within one standard deviation of the wild type mean but significantly different from the wild type population mean with a p value <0.0001, were identified as aberrant N/C ratio candidates. The gene deletions in these 25 N/C ratio mutants were confirmed by colony PCR using CP3 and CP5 gene specific primers [18] with primers internal to the NE marker construct.
The haploid wild type control used for this screen was derived from the previously described 5300 heterozygous diploid control strain in which the KanR deletion cassette is inserted in pseudogene SPAC212.05c [34]. It is of note that this wild type control strain, in which the Ish1-yEGFP construct is integrated into the deletion cassette deleting the 5300 pseudogene, displayed an N/C ratio of 0.05 in this screen, with other strains distributed around this value (Fig 2A). When Ish1-yEGFP integrated at the native ish1 locus was used to mark the nuclear envelope of wild type haploid cells, an N/C ratio of 0.08 was observed (Fig 1D and 1E). As the native ish1 locus is undisrupted in the screened strains, the introduction of an additional copy of ish1, which encodes a nuclear envelope protein, may cause the observed wild type N/C ratio reduction from 0.08 to 0.05.
Fluorescence imaging was carried out using a DeltaVision Elite microscope (Applied Precision) comprised of an Olympus IX71 wide-field inverted fluorescence microscope, an Olympus Plan APO 60X, 1.4 NA oil objective and a Photometrics CoolSNAP HQ2 camera (Roper Scientific) in an IMSOL ‘imcubator’ Environment Control System unless otherwise stated. Imaging was carried out in liquid media on glass slides at 25°C unless otherwise stated. Images were acquired in 0.2 μm or 0.4 μm z-sections over 4.4 μm, with a brightfield reference image in the middle of the sample, and deconvolved using SoftWorx (Applied Precision) unless otherwise stated. Representative images shown are maximum intensity projections of deconvolved images unless otherwise stated. For FM 4–64 staining experiments, fluorescence imaging was carried out using a wide-field inverted microscope comprised of a Nikon Eclipse Ti2 base, a Nikon Plan Apo 100X, 1.45 NA oil objective and a Photometrics Prime camera and images were deconvolved using Huygens. Colonies growing on solid agar plates were imaged using a Zeiss Axioskop 40 microscope with 10X Nikon objective, 2.5X Zeiss Optovar and Sony Alpha NEX-5 camera.
For vacuole staining, cells were grown in the presence of 50 μM FM 4–64 (Molecular Probes) for 30 minutes at 32°C, then washed in YE4S and incubated in YE4S at 32°C for 35 minutes before imaging.
To determine nuclear volume, cell volume and N/C ratio, cells and nuclei were manually measured, in brightfield reference images and fluorescence images of a NE marker respectively, using image J (NIH) as described previously [7]. n ≥ 50 cells per strain or condition.
Unless otherwise indicated, two-tailed unpaired student’s t-tests were used to determine significance of difference between two populations of N/C ratio measurements. Welch’s correction was used where indicated when an F test indicated that the variances of the two populations were significantly different.
Gene list enrichment analysis was carried out using GO::TermFinder [23] with PomBase annotations [24] and default parameters. Network analysis was carried out using esyN software [25] with PomBase curated interactions [24] and BioGRID curated interactions [36]. Default parameters, with “High and Low” setting for BioGRID interactions selected, were used.
|
10.1371/journal.pntd.0003287 | Health-Related Quality of Life among School Children with Parasitic Infections: Findings from a National Cross-Sectional Survey in Côte d'Ivoire | Parasitic infections are still of considerable public health relevance, notably among children in low- and middle-income countries. Measures to assess the magnitude of ill-health in infected individuals, however, are debated and patient-based proxies through generic health-related quality of life (HrQoL) instruments are among the proposed strategies. Disability estimates based on HrQoL are still scarce and conflicting, and hence, there is a need to strengthen the current evidence-base.
Between November 2011 and February 2012, a national school-based cross-sectional survey was conducted in Côte d'Ivoire. Children underwent parasitological and clinical examination to assess infection status with Plasmodium and helminth species and clinical parameters, and responded to a questionnaire interview incorporating sociodemographic characteristics, self-reported morbidity, and HrQoL. Validity analysis of the HrQoL instrument was performed, assessing floor and ceiling effects, internal consistency, and correlation with morbidity scores. Multivariate regression models were applied to identify significant associations between HrQoL and children's parasitic infection and clinical status.
Parasitological examination of 4,848 children aged 5–16 years revealed Plasmodium spp., hookworm, Schistosoma haematobium, Schistosoma mansoni, Ascaris lumbricoides, and Trichuris trichiura prevalences of 75.0%, 17.2%, 5.7%, 3.7%, 1.8%, and 1.3%, respectively. Anemic children showed a significant 1-point reduction in self-rated HrQoL on a scale from 0 to 100, whereas no significant negative association between HrQoL and parasite infection was observed. The 12-item HrQoL questionnaire proofed useful, as floor and ceiling effects were negligible, internally consistent (Cronbach's alpha = 0.71), and valid, as revealed by significant negative correlations and associations with children's self-reported and clinically assessed morbidity.
Our results suggest that HrQoL tools are not sufficiently sensitive to assess subtle morbidities due to parasitic infection in Ivorian school-aged children. However, more advanced morbid sequelae (e.g., anemia), were measurable by the instrument's health construct. Further investigations on health impacts of parasitic infection among school-aged children and refinement of generic HrQoL questionnaires are warranted.
| Infectious diseases like malaria and parasitic worms affect hundreds of millions of people, and impact physical and cognitive development of children in Africa, Asia, and the Americas. Over the past 20 years, it was debated how the magnitude of ill-health due to these conditions should be assessed. One proposed strategy was to include patient-based ratings of wellbeing by administration of health-related quality of life (HrQoL) questionnaires. In order to provide new evidence on disability from parasitic infections, we conducted HrQoL interviews with children aged 5–16 years from 92 schools across Côte d'Ivoire. Children were examined for parasitic infections and clinical signs like anemia, malnutrition, and organ enlargement. We compared the self-rated HrQoL of infected and non-infected children and also considered their sociodemographic background. We could not identify lowered HrQoL in infected children, but we found that children with anemia reported a 1-point lower score on a 100-point HrQoL scale in comparison with their non-anemic counterparts. We consider our HrQoL questionnaire as useful and valid, but would recommend its further testing and development in few purposefully selected settings. Further investigation of disability induced by malaria and parasitic worm infections is warranted.
| Malaria and the neglected tropical diseases (NTDs) are still of considerable public health relevance in the tropics and subtropics and their successful control is a key issue toward progress of the millennium development goals (MDGs) and the post-2015 agenda of sustainable development [1]–[4]. Preschool-aged children are considered at highest risk of malaria, whereas school-aged children are the most affected by parasitic worm infections (helminthiases) [5]–[7]. The assessment of the precise burden attributable to parasitic infections, however, is a difficult issue and there is ongoing discussion and debate [8], [9]. Over the past 20 years, the magnitude of health loss due to diseases, injuries, and risk factors has been increasingly expressed in disability-adjusted life years (DALYs). This metric is a combined measure of premature death and years of life lived with disability. For measuring the burden of helminthiases and other NTDs, specific disability weights (DWs) of morbid sequelae are considered and, by convention, scaled on an axis from 0 (no health loss) to 1 (health loss equivalent to death) [10]. Former estimates were often criticized for underestimating the true burden of infectious diseases, due to separating out morbidity (e.g., anemia), although such morbidity is partially associated with infection (e.g., hookworm and Plasmodium). Additionally, cultural and socioeconomic contexts are insufficiently taken into account, and DWs were usually based on expert opinion; thus, ignoring community- or patient-based appraisal [11]–[13]. Meanwhile, the Global Burden of Disease (GBD) consortium presented estimates for the year 2010 by incorporating different sequelae to capture direct consequences of infections and judgments on health losses from the general public in culturally and socioeconomic diverse settings [14], [15]. Nonetheless, the use of generic health status measurement instruments to expand the GBD approach has been discussed by the lead authors of the GBD 2010 study [15], thus partially addressing concerns that have been articulated a decade ago [16], [17].
The discussed generic health status measurement instruments evaluate health burden in a comprehensive way based on health-related quality of life (HrQoL) and typically include domains on physical, mental, and social wellbeing, and a visual analogue scale (VAS) for subjective health rating [18]–[20]. Thus far only few studies have assessed HrQoL and derived DWs in individuals with parasitic diseases, indicating the early stage of this approach in the field of parasitology. This issue is further underscored by conflicting results; while negative associations between HrQoL measures and Trichuris trichiura, Schistosoma mansoni, Schistosoma haematobium, and advanced Schistosoma japonicum infections were observed [21]–[23], other studies failed to show significant differences in HrQoL and DWs between infected children and their non-infected counterparts [24]–[26]. A weaker explanatory power in previous studies may partly be explained by a lack of cross-cultural validity of the questionnaires. HrQoL instruments have been developed and broadly validated in Europe and the United States of America and were originally designed for adult respondents. Child-friendly versions meanwhile exist [27], [28], but application in different cultural settings imply careful adaptations in language and scoring, thorough pre-testing, and validity analysis.
Considering the scarcity of empirical data on HrQoL assessments in school-aged children with single and multiple species infections, the aim of the present study is to strengthen the current evidence-base of disability due to parasitic diseases among pupils in Côte d'Ivoire. Hence, a cross-sectional school-based survey was carried out using standardized, quality-controlled parasitological and questionnaire tools. Furthermore, we discuss the utility and validity of a HrQoL questionnaire tailored to a given setting with basic elements from readily available tools.
The study protocol was approved by the institutional research commissions of the Swiss Tropical and Public Health Institute (Basel, Switzerland) and the Centre Suisse de Recherches Scientifiques en Côte d'Ivoire (Abidjan, Côte d'Ivoire). Ethical approval was obtained from the ethics committees in Basel (EKBB; reference no. 30/11) and Côte d'Ivoire (CNER; reference no. 09-2011/MSHP/CNER-P). Additionally, permission to carry out the study was sought from the Ministry of National Education in Côte d'Ivoire. Directors and teachers of the selected schools, district and local health and education authorities were informed about the purpose and procedures of the study. Written informed consent was obtained from parents and legal guardians of children, whilst children assented orally. Participation was voluntary, and hence, children could withdraw from the study at any time without further obligations. All collected data were coded and kept confidential. Participating children benefited from free of charge deworming with albendazole (single oral dose of 400 mg). Children identified to harbor Schistosoma spp. were given praziquantel (single oral dose of 40 mg/kg). In schools where the prevalence of Schistosoma infection was above 25%, the entire study sample was treated with praziquantel. Symptomatic malaria cases, defined as having a positive rapid diagnostic test (RDT) and fever, were offered artemisinin-based combination therapy (ACT; using artesunate-amodiaquine) and paracetamol against fever.
Between November 2011 and February 2012 (i.e., dry season) we conducted a national cross-sectional, school-based study, including parasitological and clinical examinations, and administered a questionnaire. Our aim was to select approximately 100 schools across Côte d'Ivoire, which we considered as a maximum number of locations that we would be able to visit within a 3-month period and our financial and human resources would allow to cover. A lattice plus close pairs design [29], [30] was applied for the sampling of the schools. In brief, a grid indicating latitude and longitude at a unit of 0.5° was overlaid on a map of Côte d'Ivoire that divides the country into two major ecological zones [31]. The southern ecozone is characterized by abundant rainfall (>1,000 mm per annum) and dense forest vegetation cover, whereas the northern ecozone corresponds to a savannah-type profile with markedly less precipitation. In order to achieve a representative sample of the country, 58 and 42 possible survey locations were retained after randomly drawing from each or every second grid cell of ecozones 1 and 2, respectively, taking into account population density from the last available census in 1998. About 27% of the population was estimated to live in the major urban centers in 2007 [32]. We aimed at including at least one fifth of all schools from urban areas. In total, 94 schools were selected and we double-checked that the schools comprised a minimum of 60 children attending grades 3 and 4, using a recent school inventory from a national UNICEF education program (UNICEF 2010; personal communication). Children attending grades 3 and 4 were considered as capable to express themselves and give reliable answers to questionnaire items on household assets, experienced symptoms and diseases, and HrQoL and may be retrievable in case of followed-up studies. The sample size per school was delimited to 60 children due to financial and operational constraints, considering the high number of schools to be surveyed and the maximum number of children that a survey team could sample in a single day, including questionnaire interviews and detailed laboratory work-up of blood, stool, and urine specimens. This sample size exceeds the minimum of 50 children to be surveyed in a school, as recommended by the World Health Organization (WHO) for collection of baseline information on helminth prevalence and intensity in the school-aged population within large-scale surveys [7].
Two schools were omitted in the final analysis. One school refused to participate, while another school was subjected to recent deworming. The latter would have biased the results, since signs and symptoms due to chronic helminth infections and HrQoL are likely to change after anthelmintic treatment. The remaining 92 schools are mapped by ecozone, and stratified by rural and urban setting characteristics (Figure 1).
In advance of the study conduct, directors and teachers of the selected schools were contacted and they were invited to inform parents or legal guardians of 60 children attending grades 3 and 4. Whenever necessary, children from grade 5 were invited to complement sampling to reach the targeted number of 60 children. Children whose parents/guardians had provided written informed consent were invited for participation. The objectives and procedures of the study were explained on the day of the visit. Children were then asked to provide fresh urine and stool samples in plastic containers distributed upon arrival at school. Additionally, a finger-prick blood sample was taken for preparation of an RDT of malaria (ICT ML01 malaria Pf kit; ICT Diagnostics, Cape Town, South Africa) and thick and thin blood films on microscope slides for subsequent analysis of Plasmodium infection. All biological samples were transferred to nearby laboratories and processed the same day. In brief, urine reagent strips (Hemastix; Siemens Healthcare Diagnostics GmbH, Eschborn, Germany) were used to assess microhematuria in urine samples, as a proxy for S. haematobium infection [33]. Of note, reagent strips show a high specificity for indirect diagnosis of S. haematobium among school-aged children in endemic areas [34]. Duplicate Kato-Katz thick smears [35], using 41.7 mg templates, were prepared from each stool sample. Kato-Katz thick smears were allowed to clear for 30–45 min prior to microscopic examination by experienced laboratory technicians. The number of helminth eggs was counted and recorded for each species separately (i.e., S. mansoni, A. lumbricoides, T. trichiura, hookworm, and other helminths). Blood films were stained with a 10% Giemsa solution and examined under a microscope for Plasmodium species identification and quantification of parasitemia (parasites/µl of blood) [36]. For quality control, 10% of the Kato-Katz thick smears and stained blood film slides were re-examined by a senior microscopist. In case of discrepancies (e.g., positive versus negative results or counts of parasitic elements differing by more than 10%), slides were read by a third technician and findings discussed until agreement was achieved.
All participating children underwent a clinical examination, conducted by experienced medical staff, which included hemoglobin (Hb) measurement using a HemoCue analyser (Hemocue Hb 301 system; Angelholm, Sweden) to assess anemia, palpation for liver and spleen enlargement, and measurement of body temperature using an ear thermometer (Braun ThermoScan IRT 4520; Kronberg, Germany) for identification of fever cases (≥38.0°C). Two anthropometric measurements were taken (i.e., height in cm and body weight in kg, precision 0.5 kg) for subsequent calculation of children's nutritional status.
A questionnaire assessing the socioeconomic status, self-reported symptoms and diseases, and HrQoL was administered to all children. Questions on household asset ownership, diseases, and disease-related symptoms were adapted from an instrument previously used in school-based surveys conducted in Côte d'Ivoire [37]. Children were asked for 11 different symptoms (i.e., abdominal pain, blood in stool, blood in urine, diarrhea, dysentery, fatigue, fever, headache, loss of appetite, respiratory problems, and vomiting/nausea) and eight diseases (i.e., cold, cough, eye disease, malaria, malnutrition, schistosomiasis, skin disease, and worms) using a recall period of 2 weeks. To evaluate self-rated HrQoL, the French version of the WHOQOL-BREF tool [18] served as template. Specific questions were dropped and some questions were slightly rephrased to be more specific for the current context, interviewing school-aged children in Côte d'Ivoire. In addition to specific questions focusing on HrQoL, children were asked to rate their general health status using an adapted VAS [38]. This single-item measure basically consists of a thermometer-like scale, in which the anchors are ‘best imaginable health’ and ‘worst imaginable health’, in our case defined as a maximum and minimum value of 10 and 0, respectively. The complete questionnaire instrument was further refined in several rounds of pre-testing in a primary school that was not otherwise involved in the current study. In this pre-testing, children attending grades 2–5 with different cultural backgrounds were included. We determined interview duration using a stopwatch and comprehensibility and appropriateness of the HrQoL part, which was not yet validated from earlier studies, with the goal to achieve a compact, understandable, and locally valid instrument. Questionnaire interviews in the field were conducted by members of the study team and teachers from the selected schools, who were trained beforehand.
Data were double-entered and cross-checked using EpiInfo version 3.5.3 (Centers for Disease Control and Prevention; Atlanta, United States of America) and analyzed in Stata version 10.1 (Stata Corp.; College Station, United States of America). Only data from children with written informed consent, completed questionnaire, valid parasitological results, and clinical assessments were considered for further analysis.
Socioeconomic data were utilized to calculate a wealth index following an asset-based approach as adopted and explained elsewhere [37], [39]. According to their index score, children were stratified into five economic groups according to wealth quintiles (i.e., most poor, very poor, poor, less poor, and least poor). Data on helminth infections were classified into light, moderate, and heavy, following WHO guidelines [7]. Anemia was defined as having a Hb level below 115 g/l in children aged 5–11 years and 120 g/l in children aged12–15 years [40]. The presence of organ enlargement was defined as having a palpable liver or spleen; the latter of grade 1 or higher using a Hackett's scale [41]. Indicators for malnutrition were calculated according to WHO child growth standards for children aged 5–19 years [42]. They included stunting (height-for-age), wasting (body mass index (BMI)-for-age), and underweight (weight-for-age). The latter is considered a valid measure for nutritional status in children up to 10 years only and was incorporated in a summary measure for malnutrition, defined as Z-score <−2 for any of the three nutritional indicators.
HrQoL questionnaire answers were coded as 1, 2, or 3 (in question 1 up to five codes; Appendix S1) with higher scores indicating fewer problems for a certain issue or activity. HrQoL questionnaire scores were summarized into three main domains on (i) physical, (ii) psychosocial, and (iii) environmental wellbeing. The first comprised the sum of scores from questions 2–6, the second from questions 7–9, and the third from questions 10–12. Each child's overall score on HrQoL was built by summing up individual scores from questions 1–12. Domain and overall raw scores were further converted to a 100-point scale (formula: [(raw score−lowest possible score)/raw score range]×100) [43]. Cronbach's alpha coefficient was used to assess for internal consistency of the HrQoL scores. Overall HrQoL, domain, and VAS scores were subjected to analysis on floor and ceiling effects. Floor or ceiling effects (>15% of respondents achieved lowest or highest possible score) can indicate limited content validity and reduced reliability, whilst responsiveness may be limited since changes in respondents with lowest or highest possible scores cannot be measured [44]. The validity of the HrQoL instrument was further evaluated by assessing relationships of domain, overall HrQoL and VAS scores with symptoms reporting and clinical signs using Spearman rank correlation and linear regression analysis, as appropriate. In order to relate the questionnaire measures with self-reported and clinically assessed morbidity, additional summary variables providing the total number of self-reported symptoms and diseases (n = 19) and clinical signs (n = 7) for each child was generated, with possible ranges of 0 to 19 and 0 to 7, respectively.
Chi square (χ2), Fisher's exact, Student-t, Kruskal-Wallis, and Wilcoxon rank sum tests were applied, as appropriate, to investigate significant univariate differences between groups for sociodemographic, parasitological, clinical, and HrQoL indicators. Associations between the HrQoL outcome and parasitic infection, infection intensity, and clinical status were assessed using multivariate linear regression analysis with random effects to account for clustering within schools. In case of censored data, we additionally applied tobit regression models. Particular emphasis was placed on total HrQoL and physical wellbeing domain scores as outcome in order to make explicit the physical and non-physical impacts of the health conditions assessed. Explanatories of regression models included sociodemographic, parasitological, and clinical variables. The final models were built, following a stepwise backward elimination approach. Covariates were excluded from the model at a significance level of 0.20 or higher. Relationships between the outcome and remaining explanatory variables were expressed as adjusted mean differences with corresponding 95% confidence intervals (CIs).
A total of 94 schools across Côte d'Ivoire were visited during the study and 5,491 children invited to participate. Figure 2 depicts the study compliance and participation in the various assessments undertaken. The final sample used for in-depth analysis consisted of 4,848 children from 92 schools with a mean age of 9.8 years (range: 5 to 16 years). These children had complete questionnaire, parasitological, and clinical data and had not received deworming drugs within the past 4 weeks prior to the survey. There were slightly more boys than girls (2,579 versus 2,269). 72 schools were considered rural, whilst the remaining 20 (21.7%) were based in urban settings. 4,101 (84.6%) of the children belonged to the two targeted school grades, 3 and 4.
The data set is provided as supplementary information (Data set S1).
Table 1 summarizes overall prevalence and intensity of parasitic infections, clinical signs, and self-reported symptoms and diseases. Overall 3,635 of the 4,848 children (75.0%) harbored any malaria parasite. P. falciparum was the predominant species (74.1%), followed by P. malariae (3.9%) and P. ovale (0.3%). The latter two Plasmodium species occurred mainly as co-infections with P. falciparum. Helminth infections; namely, hookworm, S. mansoni, A. lumbricoides, and T. trichiura were observed in 17.2%, 3.7%, 1.8%, and 1.3% of the children, respectively. Microhematuria was found in 5.7% of the children. The majority (95.6%) of soil-transmitted helminth infections were of light intensity, whereas about half of the S. mansoni-infected children had moderate- to heavy-intensity infections (≥100 eggs per gram of stool). More than a fourth of all children were found to be anemic (28.7%) or malnourished (28.4%) and a mean number of 6.1 experienced symptoms or diseases were reported.
Detailed information on parasitic infections and clinically assessed and self-reported morbidity stratified by sex, age group, residential area, and ecozone are provided in Supporting Information Tables S2 and S3. Boys showed significantly higher infection rates for P. falciparum, hookworm, and S. mansoni (Table S1). Prevalence rates differed between age groups; while P. malariae was more often found in younger children, infections with Schistosoma and soil-transmitted helminths were more prevalent in children aged 11–16 years than in their younger counterparts. Plasmodium spp. and soil-transmitted helminth infections were most prevalent among the poorest and rural households (all p<0.001). Plasmodium spp. was more common in children living in the northern ecozone. Clinical morbidity, such as anemia and indicators for malnutrition, was more pronounced in boys than girls and in older children compared to their younger counterparts (Table S2). Splenomegaly was found to be more common in the younger age group (p = 0.014) and in children from rural and northern settings compared to children living in urban and southern environments (both p<0.001). Anemia (p = 0.049), splenomegaly (p<0.001) and stunting (p<0.001) were significantly lower in children from wealthier households. Furthermore, helminth (OR = 1.69, p<0.001) and Plasmodium (OR = 1.44, p<0.05) mono-infected as well as co-infected (OR = 2.0, p<0.001) children showed significantly higher odds ratios (ORs) for anemia than their non-infected peers in multivariable logistic regression analysis. Symptom and disease reporting was higher in girls compared to boys, in older compared to younger individuals, in children from northern regions compared to their counterparts living in the southern ecozone (all p<0.001), and in children from poorer households (p = 0.025).
Table 2 shows the results from the utility and validity analysis of the HrQoL measures. For the summary scores, floor and ceiling effects were negligible. In contrast, relevant ceiling effects were observed for single HrQoL domains and the VAS scores. Internal consistency of the 12-item HrQoL questionnaire was above the recommended threshold of 0.7 for Cronbach's alpha needed for comparison between groups. The item-rest correlations were all above 0.25, indicating that single items measured the same construct as the remaining ones and removal of a specific item would not have increased Cronbach's alpha.
Self-reported symptoms and diseases were reflected in the HrQoL. All HrQoL measures showed significant negative correlations and associations with increasing number of self-reported morbidities. For an incremental increase of 1 self-reported morbidity, the overall HrQoL decreased by 1.4 points (p<0.001). Clinical signs were mainly captured by the physical domain of the HrQoL tool, showing a decreased domain score of 1.2 points (p = 0.001) by each supplemental clinical morbidity observed. VAS scores showed a statistically significant correlation and association with self-reported symptoms and diseases (Table 2) and also a statistically significant correlation with overall HrQoL (all p<0.001), but the correlations were only weak (ρ = −0.22 and ρ = 0.30, respectively). The VAS results were not considered for in-depth analysis and calculation of DWs due to deviance between actual data collected and the original concept of the scale.
Univariate analysis showed several differences in overall HrQoL among groups with different sociodemographic factors and observed clinical signs (Table 3). Boys reported higher overall HrQoL scores, which were mainly driven by higher self-rated environmental wellbeing. Children from the most poor wealth quintile showed significantly lower scores for all three HrQoL domains. Lower scores for psychosocial and environmental wellbeing, and thus lower overall HrQoL scores, were observed in older children and children living in urban areas. Children from the northern regions reported higher physical but lower environmental wellbeing than their peers from the southern zone. Children's HrQoL with regard to parasitic infections mainly showed differences for the physical domain. Microhematuria negatively affected physical wellbeing, while light-intensity soil-transmitted helminth infections and low Plasmodium parasitemia were associated with fewer problems in this domain compared to non-infected counterparts. Comparison for Plasmodium-helminth co-infection categories and the number of concurrent parasitic infections (including malaria parasites) showed that children harboring two or more concurrent infections reported the highest physical wellbeing scores compared to their mono- or non-infected counterparts. Anemic children's HrQoL was considerably compromised compared to non-anemic children. A similar but less pronounced decrease in HrQoL was found in children with splenomegaly. Other observed clinical signs showed no significant effects on children's overall HrQoL, but wasted children reported a significantly increased psychosocial wellbeing, while generally malnourished children reported not only higher psychosocial but also higher environmental wellbeing.
Table 4 provides an overview on significant associations between sociodemographic, parasitological, and clinical variables on one hand and self-reported HrQoL on the other hand, placing emphasis on summary and physical wellbeing scores, derived from multivariate linear regression with a stepwise backward elimination procedure. Sex, socioeconomic status, anemia, Plasmodium spp. infection, Plasmodium-helminth co-infection, and number of concurrent parasitic infections remained significant predictors for overall HrQoL. If only physical wellbeing was considered, negative associations of clinical manifestations such as anemia and malnutrition were more pronounced. Interestingly, several single species parasitic infections (i.e., Plasmodium spp., and soil-transmitted helminths) and multiple species parasitic infections (i.e., Plasmodium-helminth, and number of concurrent infections ≥2) showed a significant positive association with self-reported physical wellbeing.
We present HrQoL measures among 4,848 school-aged children surveyed during a 3-month cross-sectional survey in the dry season in Côte d'Ivoire, and explore associations with parasitic infections and clinical and sociodemographic measures. Parasitological examination revealed a very high prevalence of Plasmodium spp. infection (75.0%). Helminth infections were considerably lower; 17.2%, 10.6%, 3.7%, 1.8%, and 1.3% for hookworm, S. haematobium (microhematuria), S. mansoni, A. lumbricoides, and T. trichiura, respectively. More than a quarter of the surveyed children showed clinical signs of anemia and malnutrition. Findings from multivariate linear regression analysis revealed that the children's self-rated overall HrQoL and physical wellbeing is lower among those affected by anemia and malnutrition compared to their counterparts without anemia and malnutrition. Surprisingly, associations between HrQoL and parasitic infection status were of positive rather than negative direction. Sociodemographic variables such as sex, age group, socioeconomic status, and setting characteristics had considerable influences on children's perceived HrQoL. The locally adapted HrQoL instrument employed showed acceptable utility considering minimal floor and ceiling effects and a robust internal consistency (Cronbach's α≥0.7). Significant correlations and associations between HrQoL scales and self-reported and clinically assessed morbidity were found and even though the effect sizes were weak, they may further support the concept of health measured by the HrQoL tool.
Interestingly, we could not identify significantly lower HrQoL scores in Plasmodium- and helminth-infected children compared to their non-infected peers. Possible explanations for this finding are offered for consideration. First, in Côte d'Ivoire 100% of the population is at risk of Plasmodium infection [3] and previous research concluded that malaria transmission is perennial [45], [46]. Constant exposure from early childhood onwards leads to naturally acquired immunity to malaria at an early age [47]. Thus, most of the Plasmodium infections we identified in the school-aged population surveyed were asymptomatic (>98%). Levels of transmission and endemicity of parasitic infections has been shown to influence children's HrQoL. For example, Kenyan school-aged children infected with S. haematobium from a high endemicity setting reported similar HrQoL measures than their non-infected counterparts, whilst infected children in a low prevalence village reported significantly lower HrQoL compared to non-infected children [23].
Second, our study focused on children who were present at school the day of the survey. Children experiencing a clinical disease episode, perhaps related to a parasite infection, and who might have expressed lowered HrQoL, were more likely to be absent from school than their healthier peers. Helminth infections, additionally, might still be at a less advanced stage with regard to disability in children compared to adolescents or adults. Smaller studies conducted in the People's Republic of China and Kenya also found no evidence of significant differences in self-rated HrQoL between helminth-infected and non-infected school children [24], [26]. Those studies that reported negative associations between HrQoL and helminth infections either focused on adult populations [22] or investigated chronic and advanced clinical stages of an infection [21], [48].
Third, polyparasitism, particularly Plasmodium-helminth co-infections, is common in Côte d'Ivoire [49]–[51]. It follows that interactions between multiple species parasitic infections and their influence on ill-health must be considered. Indeed, potentially beneficial effects from light-intensity helminth infections on clinical outcomes (i.e., anemia) and subtle morbidity (i.e., physical fitness) in school-aged populations from malaria co-endemic settings in Côte d'Ivoire have been indicated [46], [50], [51]. Underlying mechanisms might be seen in the immunomodulatory features of helminth infections that down regulate the pro-inflammatory immune response needed to combat intracellular parasites like Plasmodium. Consequently, this may negatively affect resistance but simultaneously promote tolerance to malaria-related pathology by controlling harmful associated inflammation [52]. The current study confirms these prior observations, as we observed a positive association between soil-transmitted helminth infections, Plasmodium-helminth co-infections, and two or more concurrent infections including malaria parasites, and reported physical wellbeing.
We found significantly lower HrQoL among anemic children compared to non-anemic children. Parasitic infections, most notably Plasmodium and hookworm contribute to the development of anemia [53]. Plasmodium and helminth mono- or co-infected children in our sample showed significantly higher odds ratios for anemia than their non-infected counterparts (all ORs>1.4). Consequently, we suggest the attribution of direct disease consequences (sequelae) – such as anemia due to specific parasitic infections – to the etiological cause in future burden estimates [14]. We found a 1-point lower HrQoL score overall and a 2-point lower physical wellbeing score on a 100-point scale. If divided by 100, these findings might translate to DWs of 0.01 and 0.02 on the DW scale that ranges from 0 to 1. Such DWs are within the range of recent DW estimates of the GBD 2010 Study, which were set at 0.005, 0.058, and 0.164 for mild, moderate, and severe anemia [15], [54].
The HrQoL concept attempts to evaluate the impact of diseases and injuries from a comprehensive point of view, incorporating psychological, social, and environmental wellbeing on top of physical health [12], [15]. We found that particularly psychosocial and environmental measures of wellbeing were significantly associated with sociodemographic variables like sex, age, socioeconomic status, and residential area. Associations of the physical component of HrQoL with parasite infections and clinical signs were observed to be more pronounced and indicated that the perceived health status varies between and depends importantly on different sociocultural settings. Our results are in line with previous observations [22], [23], [25] and highlight the importance of inclusion of social determinants for more integrative burden of disease assessments.
Our data stem from a large-scale nation-wide survey, which subjected almost 5,000 children to detailed clinical and parasitological examinations, coupled with a questionnaire. A major weakness of previous studies was their small sample sizes [22], [24], [25]. Further, we consider the setting-tailored, applied HrQoL tool as a useful and valid instrument. Its internal consistency was good (Cronbach's α>0.7) and floor and ceiling effects for the overall HrQoL were minimal, despite its shortness, including only 12 items compared to 26 questions in the WHOQOL-Bref, which was used as a template to develop our tool [18]. Particularly the ceiling effects were more pronounced when looking at single domain scores, which is, however, not surprising, considering the lower number of items in each domain. The ceiling effects found for the physical domain, were addressed by utilizing tobit regression analysis, which have been shown to provide more reliable estimates in censored data [55], in parallel to linear regression models. The negative associations and correlations between HrQoL and symptom and disease reporting followed the logic of lower self-rated HrQoL in simultaneously higher experienced morbidity and supported the construct that our instrument measures.
Data collection on a national scale entails several limitations. To respect the tight time schedule and in view of limited financial and human resources, all parasitological, clinical, and questionnaire information had to be collected within a single day at each location by dedicated field teams. Consequently, teachers of the selected schools were trained to administer our questionnaire and they assisted in the conduct of the interview. Given our time constraints and restricted resources, we were not able to assess inter-observer agreement and cannot exclude measurement errors due to variation between interviewers. Another limitation regarding the questionnaire was the difficult implementation of the VAS, as already observed elsewhere [24]. The concept of this scale, the range of 0 to 100, and the fact that children had to point out their respective health status on a sheet was poorly understood. As an adaptation, children were asked to rate and orally express their health status according to a scale they were more familiar with, a scale of school marks (ranging from 0 to 10). However, this procedure resulted in a categorical rather than an interval distribution of scores. Unfortunately, this limitations hindered us to fully exploit these data and derive DWs, which could have been compared with previous research conducted elsewhere focusing on chronic S. japonicum infection [48]. Hence, there is a pressing need for a culturally accepted alternative to the VAS.
We conclude that the assessment of HrQoL in school-aged children in areas where parasitic infections are still widespread tends to be difficult and may not be sensitive enough to capture subtle morbidities. Important factors blurring the picture might be the often asymptomatic course due to acquired immunity in malaria and more subtle morbidities at this age for helminth infections, which therefore may not be perceived as disabling by infected children. School absenteeism adds bias, as non-inclusion of children who might experience more measurable disability will not be part of the analysis. Importantly though, the applied instrument showed acceptable utility and validity and was able to identify significant disability of more chronic sequelae such as anemia. Further refinement and more rigorous reliability measurements of the tool are needed. Surveys in settings targeting specific parasite endemicity levels and efforts to include non-enrolled and otherwise absent school-aged children might resolve some of the limitations highlighted here. The aim of developing, validating, and applying setting-specific HrQoL tools that will allow comparison between areas and measuring changes over time remains – particularly as large-scale control efforts targeting malaria and the NTDs are underway.
|
10.1371/journal.pntd.0004052 | Tracking Dengue Virus Intra-host Genetic Diversity during Human-to-Mosquito Transmission | Dengue virus (DENV) infection of an individual human or mosquito host produces a dynamic population of closely-related sequences. This intra-host genetic diversity is thought to offer an advantage for arboviruses to adapt as they cycle between two very different host species, but it remains poorly characterized. To track changes in viral intra-host genetic diversity during horizontal transmission, we infected Aedes aegypti mosquitoes by allowing them to feed on DENV2-infected patients. We then performed whole-genome deep-sequencing of human- and matched mosquito-derived DENV samples on the Illumina platform and used a sensitive variant-caller to detect single nucleotide variants (SNVs) within each sample. >90% of SNVs were lost upon transition from human to mosquito, as well as from mosquito abdomen to salivary glands. Levels of viral diversity were maintained, however, by the regeneration of new SNVs at each stage of transmission. We further show that SNVs maintained across transmission stages were transmitted as a unit of two at maximum, suggesting the presence of numerous variant genomes carrying only one or two SNVs each. We also present evidence for differences in selection pressures between human and mosquito hosts, particularly on the structural and NS1 genes. This analysis provides insights into how population drops during transmission shape RNA virus genetic diversity, has direct implications for virus evolution, and illustrates the value of high-coverage, whole-genome next-generation sequencing for understanding viral intra-host genetic diversity.
| Dengue virus (DENV) is transmitted between humans through the bite of infected female Aedes aegypti mosquitoes. Virus populations experience significant drops in size and are subject to differing selection pressures as they cycle between human and mosquito hosts. Subsequent changes in viral intra-host genetic diversity may have consequences for the adaptability and fitness of the virus population as a whole but are poorly understood. To study the impact of human-to-mosquito transmission on DENV populations, we allowed mosquitoes to feed directly on patients with acute dengue infections, then deep-sequenced DENV populations from patient plasma samples and from the abdomens and salivary glands of corresponding mosquitoes. These matched samples allowed us to estimate the size of the population drop that occurs during establishment of infection in the mosquito, track changes in viral intra-host variant repertoires at different stages in transmission, and investigate the possibility of host-specific immune selection pressures acting on the virus population. These novel insights improve our understanding of DENV population dynamics during horizontal transmission.
| With 3.6 billion people at risk and nearly 400 million infections annually [1,2], dengue has become the most important mosquito-borne viral disease affecting humans today. Dengue virus (DENV) is a positive-sense, single-stranded RNA virus of the family Flaviviridae, genus Flavivirus. The ~10.7 kb DENV genome encodes three structural proteins (capsid [C], premembrane [prM], and envelope [E]) and seven non-structural (NS) proteins (NS1, NS2A, NS2B, NS3, NS4A, NS4B, and NS5).
DENV is transmitted between humans by the mosquitoes Aedes aegypti and Aedes albopictus. which acquire the virus by taking a bloodmeal from an infected human. Once ingested, DENV first infects and replicates in the mosquito midgut epithelium. It subsequently disseminates through the hemolymph to infect other organs such as the fat body and trachea, finally reaching the salivary glands, where it is secreted into mosquito saliva and injected into a human host during a subsequent blood-feeding event [3].
As is typically the case for RNA viruses, the DENV RNA-dependent RNA polymerase (RdRp, encoded by DENV NS5) operates at low fidelity, resulting in the accumulation of a population of closely-related but genetically-distinct viral genomes, organized around a consensus sequence, within each individual human or mosquito host. Sometimes termed a quasispecies, these intra-host variants are thought to interact cooperatively on a functional level, and to collectively contribute to the overall fitness of the virus population (reviewed in [4]). This has important implications for viral pathogenesis: High fidelity poliovirus mutants, for example, are highly attenuated, demonstrating the importance of genetic diversity and the ability to adapt to changing environments during infection [5,6].
Adaptability is especially important for mosquito-borne viruses, which encounter distinct selection pressures when cycling between vertebrate and invertebrate hosts. Even within a single host, intra-host variants generated through multiple rounds of virus replication are subject to evolutionary mechanisms such as bottlenecks, drift, and positive and negative selection pressures. In human-derived DENV populations, highly immunogenic E protein domains also display higher levels of intra-host genetic diversity, suggesting that selection pressures on low frequency population variants operate even during acute infection [7]. In mosquitoes, RNA interference (RNAi), a key antiviral defense mechanism in insects, has been proposed to be a driver of viral intra-host genetic diversity in the Culex-West Nile virus (WNV, family Flaviviridae) system [8], where higher intra-host diversity levels have been reported in the mosquito than in vertebrate hosts [9,10].
Host alternation also subjects arboviruses to frequent and significant drops in population size. Within the mosquito and vertebrate hosts, these drops occur during initial establishment of infection and subsequent spread through various tissues and organs, as well as during the process of blood-feeding itself, where only a small percentage of the total virus population circulating in the human actually seeds infection in the mosquito host. It is unclear how such drops shape the diversity and repertoires of intra-host virus populations, and the potential effect they have on transmission.
Studies on intra-host genetic diversity in DENV have focused on virus populations in the human, and with few exceptions [7], have been confined to the sequencing of one or two viral genes [11–14]. It is unclear, however, what happens to intra-host diversity during replication in the mosquito, and particularly after the virus population has gone through a significant drop, such as occurs when a mosquito feeds on a dengue patient that has a viremia around the 50% mosquito infectious dose (MID50).
In this study, we used whole-genome amplification and next-generation Illumina sequencing to characterize DENV intra-host genetic diversity in both patient- and matched mosquito-derived virus populations. Because mosquitoes in our study were infected by allowing them to feed directly on patients, we were able to track changes in viral populations during human-to-mosquito transmission. We used a rigorous variant-calling algorithm to identify low frequency single nucleotide variants (SNVs) within a viral population. This unique study design provides important insights into how population drops shape viral genetic diversity, as well as into the differing selection pressures between human and insect hosts.
Study protocols were reviewed and approved by the Scientific and Ethical Committee of the Hospital for Tropical Diseases (CS/ND/09/24) and the Oxford Tropical Research Ethical Committee (OxTREC 20–09). Written informed consent was obtained from all participants or from the parent or guardian of any child participants on their behalf.
The human viremic plasma and DENV2-infected mosquitoes used here originated from the study by Nguyen et al. [15], in which field-derived Ae. aegypti mosquitoes were allowed to feed on acute dengue cases. Just prior to blood feeding, venous blood was drawn and the plasma fraction stored. Blood-fed mosquitoes were reared until day 14 when the abdomen and salivary glands were dissected and stored.
Out of the 83 DENV2-infected patients described by Nguyen et al. [15], we selected 12 patient plasma samples and 36 paired blood fed mosquitoes (three per patient) for investigation of DENV sequence diversity in this study. The rationale for these patient-mosquito pairs was that six of them involved mosquito cohorts that blood fed when the plasma viremia was near to the calculated MID50 (S1 Fig) [15]. Hence the seeding of initial infection in these mosquitoes was most likely by a small or very small population of infectious virions, i.e. a major drop in DENV population size moving from human blood to mosquito. Nucleic acid was extracted from plasma samples and from mosquito tissues for quantification of the DENV RNA concentration as described previously [15].
Next-generation whole-genome sequencing of human- and mosquito-derived DENV samples was performed as described in [16]. Viral RNA was extracted from human plasma or mosquito abdomen / salivary gland samples using the QIAamp Viral RNA Mini Kit (Qiagen), and cDNA synthesis was carried out with the Maxima H Minus First Strand cDNA Synthesis Kit (ThermoScientific) using a single primer designed to bind to the 3' end of the viral genome. The entire DENV genome was PCR-amplified in 14 overlapping fragments, each about 2kb in length, using the PfuUltra II Fusion HS DNA Polymerase (Agilent Technologies). Primer sequences can be found in S5 Table. PCR products were run on an agarose gel and purified using the Qiagen Gel Extraction Kit (Qiagen).
For each sample, equal amounts of all PCR-amplified fragments were combined and sheared on the Covaris S2 sonicator (Covaris) to achieve a peak size range of 100–300 bp (shearing conditions: duty cycle—10%; intensity—5; cycles per burst—200; time—110 seconds). Samples were purified with the Qiagen PCR Purification Kit (Qiagen) and quality-checked on the Agilent 2100 Bioanalyzer with a DNA 1000 Chip (Agilent Technologies).
Library preparation was performed with the KAPA Library Preparation Kit (KAPA Biosciences). After end-repair, A-tailing, and adapter ligation, ligated products in the 200–400 bp range were gel-extracted with the Qiagen Gel Extraction Kit (Qiagen). Samples were subjected to 14 PCR cycles to incorporate multiplexing indices, quantified with the KAPA SYBR FAST qPCR Master Mix (KAPA Biosciences) on the LightCycler 480 II real-time thermocycler (Roche Applied Science), and pooled.
Paired-end, multiplexed sequencing (2 x 101 bp reads) of libraries was performed on the Illumina HiSeq (Illumina) at the Genome Institute of Singapore. Base calling was done with CASAVA 1.7; reads that did not pass Illumina’s chastity filter (CASAVA 1.7 user guide) were removed.
Illumina-generated FASTQ files were put through the Viral Pipeline Runner (ViPR, available at https://github.com/CSB5/vipr), which automates the following steps: Iterative mapping of reads against the DENV-2 reference sequence NC_001474 using the Burrows-Wheeler Aligner [17], generation of a consensus sequence, and calling of low frequency single nucleotide variants (SNVs) against the consensus using the LoFreq algorithm [18]. SNVs differ from single nucleotide polymorphisms (SNPs) in that the latter occur between the consensus sequences of different individuals.
LoFreq has previously been applied to DENV datasets, and its SNV predictions on these datasets have been experimentally validated down to 0.5% frequency [18]. To increase specificity and keep only conservative predictions, SNVs that fulfilled any of the following criteria were discarded: located within primer sequences, located adjacent to known homopolymer regions, coverage of <1000X, frequency of <0.01 (1%).
Patient DENV clinical samples from the Early DENgue study (EDEN) [19] were previously assigned into human-human transmission pairs thought to be separated by a single mosquito [20]. This was based on Sanger sequence similarity, distance between physical addresses, and time between dates of fever onset [20]; the spatiotemporal criteria are consistent with the National Environment Agency's criteria for active transmission [21]. Whole-genome Illumina sequencing and SNV calling for these samples was carried out as described above.
To identify mutational hotspots, a scanning window approach was used to scan the DENV genome (window size of 20, overlap of five nucleotides) for an excess of SNVs in a window compared with the genome-wide average (binomial test; Bonferroni-corrected p-value < 0.05). This was done on a per-sample basis. For coldspots, SNVs from all samples were pooled and scanned for windows (minimum size of 40) with a depletion of SNVs (binomial test; Bonferroni-corrected p-value < 0.05) [18].
Since haplotype reconstruction is a notoriously difficult problem, we resorted to a simple approach that estimates a lower bound of the number of haplotypes (viral genomes) present in a sample. This was done by greedily clustering SNVs based on their allele frequency confidence intervals (Agresti-Coull at the 0.05 level). SNVs were sorted by their allele frequency and the SNV with highest allele frequency seeds the first cluster. Variants were added to an existing cluster if their upper confidence interval limit was greater than the cluster minimum, otherwise they form a new cluster. The number of clusters represents a lower bound of the number of distinct viral genomes present in a sample. This clustering approach is implemented as one of the tools that come bundled with LoFreq [18].
The size of the infecting virus population was estimated by simulating 1000 samplings of varying size from the virus population in the human, following a normal distribution with mean SNV frequency matching that in the human. This sampling was carried out for a range of mean SNV frequencies. The sampling error rate was calculated from the simulated SNV frequencies, and p-values were computed using a two-tailed t-distribution (S3 Table).
Nguyen et al. [15] infected field-derived Ae. aegypti mosquitoes with DENV2 by allowing them to blood-feed directly on patients at the acute stage of infection. To track changes in viral intra-population genetic diversity during human-to-mosquito transmission, we performed whole-genome Illumina sequencing of DENV2 populations from 12 patient plasma samples and 36 infected mosquitoes (three per patient) derived from that study [15]. We successfully obtained genome-length DENV2 sequences from all 12 patient plasma samples, as well as from 25 mosquitoes (21 abdomen-derived and 25 salivary-gland-derived). We were unable to PCR-amplify virus from the remaining 11 mosquito samples. Successfully sequenced patient and mosquito samples are described in S1 Table.
Single nucleotide variants (SNVs) in each DENV2 population were called with the LoFreq variant calling algorithm [18]. At a conservative minimum cutoff of 1% frequency, we observed a total of 1,116 SNVs across all human- and mosquito-derived sequence sets. These were evenly distributed across the ~10.7 kb DENV2 genome (S2 Fig).
We examined four measures of intra-sample diversity, calculated on a per sample basis [7]: a) the number of SNVs, b) the sum of SNV frequencies, c) the average SNV frequency, and d) the standard error of the mean (SEM) SNV frequency (Fig 1). These measures capture different aspects of diversity: the number of SNVs indicates how many distinct variant positions there are along the viral genome; the sum and mean of SNV frequencies indicate how commonly these variants occur and whether they are distributed across many or a few positions; the SEM indicates whether sample diversity comes from dominant variant genomes, minor variant genomes, or a mix of both.
All four diversity measures varied widely among individual samples (Fig 1A–1D). No statistically significant differences in any of these four measures were observed between DENV populations from human plasma, mosquito abdomens and mosquito salivary glands (p > 0.05, Kruskal-Wallis test) (Fig 1A–1D), indicating that intra-sample diversity levels remain similar during replication in human and mosquito hosts.
Although we cannot experimentally follow transmission of DENV2 from mosquitoes into humans, we had access to patient plasma samples from the Early DENgue (EDEN) study carried out in Singapore from 2005–2007 [19]. DENV1 and DENV3 samples from this study were previously assigned to human-human transmission pairs thought to be separated by a single mosquito, based on Sanger sequence similarity and spatiotemporal relationships (distance between patients' homes and time between dates of fever onset) between members of a pair [20]. Illumina deep sequencing of these samples revealed that measures of intra-sample diversity also did not change significantly between members of a human-human transmission pair (S3A–S3D Fig), suggesting that passage through a mosquito does not affect intra-host viral diversity levels upon subsequent replication in the human.
Although levels of intra-host diversity remain unchanged, horizontal transmission may alter the SNV repertoire in a viral population if existing SNVs are lost and new ones generated. To examine this, we tracked individual SNVs during transmission from human to mosquito abdomen to mosquito salivary glands (Fig 2). Of 267 SNVs present in human plasma-derived DENV2 populations, only 26 (9.7%) were observed again in any of the associated mosquitoes (abdomen, salivary glands, or both), and only 37 of 478 SNVs present in the mosquito abdomen (7.7%) were observed again in the salivary gland (Fig 2A). This indicates that viral populations are able to quickly restore their diversity upon replication in a new compartment, but predominantly with a very different SNV repertoire that most likely arises through random mutation. It is possible that SNVs not observed in the mosquito abdomen or salivary gland had dropped below our conservatively-chosen level of detection (frequency < 1%).
We observed relatively few instances of SNVs being maintained across transmission stages (Fig 2B). These SNVs were present at a significantly higher frequency than those that were lost (Fig 3A and 3B), suggesting that higher frequency SNVs are more likely to be detectably transmitted. Maintained SNVs were never seen in more than one human-mosquito group, but were observed in more than one mosquito per human approximately half the time (12 out of 26 cases; in 8 of these cases, SNVs were maintained in all three mosquitoes that had fed on the same human) (Table 1). No significant difference in frequency was observed between SNVs maintained in one mosquito and SNVs maintained in more than one mosquito (Fig 3C).
SNVs may also be maintained because they play roles in increasing virus fitness, and selection for these SNVs may result in an increase in their frequencies over the course of transmission. We did not, however, observe this—the frequencies of most maintained SNVs (both synonymous and non-synonymous) remained similar in human-, mosquito abdomen-, and mosquito salivary gland-derived DENV populations (Fig 4), suggesting that most SNVs are neutral in the mosquito. There were several exceptions—for example, the frequency of 3989 T>C (NS2A) nearly doubles from 0.11 in the human to 0.20 in the mosquito, the frequency of 1745 T>C (E) nearly halves from 0.26 in the human to 0.16 in two separate mosquitoes, and the frequency of 507 T>C (M) increases 8-fold from 0.02 in the human to 0.16 in the mosquito abdomen, but drops back to 0.07 in the salivary gland (Fig 4). It is possible that these more dramatic changes in SNV frequency may be due to selection for fitter variants, but it is difficult to differentiate this from stochasticity resulting from drops in population size.
In the EDEN human-human transmission pairs, 15 of 130 SNVs (11.5%) present in the first member of the pair were observed again in the second member (Fig 2C and Table 2). The majority of these SNVs were from a single pair, where 12 out of 23 SNVs (52.2%) in the first member were maintained (S2 Table). Thus, similar to what we observed for human-mosquito pairs, linked human-human pairs of DENV clinical samples also yielded predominantly different SNV repertoires, although there may be exceptions, as observed here. While the majority of SNVs may drop below our detection limit during passage through the mosquito, it is possible that, once transmitted back into the human, selection pressures there may again drive up the frequencies of SNVs that impact virus fitness.
Given that ~90% of SNVs detected in the human were lost, presumably due to chance, during transmission to the mosquito, we wanted to estimate the size of the population drop at which this would occur. For an assumed range of plasma viremia levels from 102 to 107 infectious virions / ml, we estimated the maximum size of the virus population infecting the mosquito for which loss of SNVs could be attributed to random chance. This was done by simulating 1000 samplings of varying sizes from the viral population in the human.
For SNVs with frequencies from 0.01 to 0.03 (the vast majority of SNVs in our dataset), the maximum infecting virus population size for SNV loss to be attributable to chance was 100 virions (S3 Table). This represents the number of virions that establish a productive infection in the mosquito midgut. Since a 2 μl bloodmeal [22,23] taken from a patient with plasma viremia equivalent to the MID50 of 2x106 RNA copies / ml [15] would be expected to contain 4000 RNA copies, this estimate suggests that the viral population size actually infecting the mosquito midgut is reduced at least 40X, probably due to the bulk of viral genomes being non-infectious.
Low frequency DENV variants that arise during virus replication are subject to immune selection pressure. To examine differences in selection pressures between hosts, we compared the ratio of the number of non-synonymous to synonymous SNVs (#NS/#S) for each gene in human- versus mosquito-derived virus populations. Because of the small numbers of human plasma-derived DENV2 SNVs from this study, we pooled these with SNVs detected in DENV1 and DENV3 populations from the EDEN study plasma samples [19]. We are aware that selection pressures may differ across DENV serotypes, but this approach allowed us to broadly compare the vertebrate and invertebrate hosts.
Human #NS/#S ratios were significantly higher than mosquito ratios for the prM, E, and NS1 genes (Fig 5A). While non-significant, the ratio for C in the human was nearly double that in the mosquito (2.17 versus 1.18). It is striking that the products of these genes are actively targeted by the human antibody response (reviewed in [24]), while no antibody response exists in the mosquito immune repertoire. Although these ratios did not reach the levels formally required for positive selection, we note that positive selection is usually measured at consensus level over time scales much longer than a single transmission [25]. We are instead interested in selection pressures acting on low frequency variants generated over multiple virus replication cycles in each host.
We found no significant differences in #NS/#S ratios between mosquito abdomen and salivary gland for any of the viral genes (Fig 5B), suggesting that host might be more important than body compartment. The much smaller number of SNVs in these two datasets, however, makes it difficult to rule out distinct selection pressures acting in separate mosquito tissues.
We also used a scanning window approach to identify mutational hot- and coldspots in the DENV genome; i.e. regions containing a statistically significant excess or lack of SNVs compared with the genome-wide average (Fig 6). We identified a hotspot in E, the gene encoding the DENV envelope protein, in a single human-derived sample from Vietnam, but not in any of the mosquito-derived samples (Fig 6). An E hotspot was also found in a DENV3 human-derived sample from the EDEN study (S4B Fig). These hotspots were located in E domains (ED) II and I respectively, which are known targets of the antibody response [24]. Taken together with our finding that the E #NS/#S ratio is significantly higher in humans compared to mosquitoes (Fig 5A), we speculate that immune pressure on the E protein may play a bigger role in generating diversity in the DENV population in humans than in mosquitoes.
Hotspots in NS3, which encodes the viral serine protease and helicase, were detected in two mosquito-derived samples. While not significantly different between humans and mosquitoes (Fig 5A), the NS3 #NS/#S ratio in mosquitoes was significantly higher than the average across all viral genes (p = 0.027, exact binomial test), suggesting that NS3 may experience a stronger selection pressure than other viral genes in the mosquito. We also identified a 3'UTR hotspot in one mosquito-derived sample. Adaptation to mosquitoes has been proposed as the major driver of evolution in the 3'UTR of chikungunya virus (CHIKV) [27], an alphavirus of the family Togaviridae; it will be intriguing to test this hypothesis in DENV, a flavivirus.
The coldspot in prM detected in mosquito-derived samples corresponds to a conserved region of the gene, which has been reported to be important for the association of prM with E during viral assembly [18,28]. We also detected a coldspot in NS5 in mosquito-derived samples, which spans the junction between the methyltransferase and polymerase domains of the protein. The lack of SNVs in these regions suggests the presence of functionally important residues.
27 SNVs (2.4% of all SNVs) were found to encode low frequency premature stop codon mutations. These were spread across the coding region of the DENV genome in both human- and mosquito-derived DENV populations (S4 Table). Previous studies have proposed that defective RNA viruses can be transmitted through complementation by co-infection of host cells with functional virus [29,30]; however in our dataset we did not observe transmission of any of these SNVs.
To determine how SNVs were distributed into distinct viral genomes, we clustered SNVs in each sample based on frequency, reasoning that SNVs present on the same viral genome would be present at similar frequencies. This gave us an estimate of the minimum number of distinct variant viral genomes in each sample (Fig 7A); the actual number is likely to be higher in many samples, since different genomes could be present at the same frequency, and lower frequency SNVs tend to cluster together without good separation. Similar to the diversity measures in Fig 1, we found no significant difference in the minimum number of distinct variant genomes between virus populations derived from humans, mosquito abdomens, and mosquito salivary glands (p > 0.05, Kruskal-Wallis test) (Fig 7A).
We next used network diagrams to visualize transmission patterns of viral genomes (Fig 7B). In this example (patient-mosquito group 641), no consensus changes were observed as the virus was transmitted from human to mosquito abdomen to mosquito salivary gland. A maximum of only one SNV from each variant genome (shown as circles radiating out of the consensus sequence) was maintained across transmission stages (Fig 7B). This situation was similar across all our other patient-mosquito groups, where it was most common for only one SNV per variant genome to be maintained across transmission stages, with a maximum of two SNVs being maintained as a unit. This suggests that SNVs in each sample were spread out across many variant genomes instead of being clustered together on one genome, and that the actual number of distinct genomes per sample is higher than our predicted minimum number.
Here we report the use of next-generation, whole-genome DENV sequencing to follow within- and between-host differences in viral populations in naturally-infected humans and their infected Ae. aegypti mosquito counterparts.
We observed that the vast majority (>90%) of SNVs in a population were lost upon virus transmission from human to mosquito. Based on this, we estimated that a maximum of ~100 virions infect the mosquito midgut, with the actual number varying with the level of viremia in the human. This number is consistent with previous studies that used fluorescent reporter viruses to show that, even with high titer bloodmeals, oral infection with WNV or Venezuelan equine encephalitis virus (VEEV, family Togaviridae) resulted in a relatively small number of midgut cells being infected, on the order of 15 for WNV [31] and 100 for VEEV [32]. The VEEV study detected a higher than expected frequency of dually-infected cells, suggesting that only a small percentage of midgut cells are susceptible to infection [32].
A 2 μl bloodmeal from a human host where the DENV plasma viremia is 2x106 RNA copies per ml [15] would be expected to contain 4000 RNA copies. Our estimate suggests that the number of virions that actually infect the mosquito midgut is at least 40X lower. This drop, compatible with a bottleneck scenario, could be due to a number of factors, such as the presence of non-infectious particles, preference for certain cell types, and interference with receptor binding, perhaps by the midgut microbiota or proteins in the bloodmeal. For example, the particle-to-PFU (plaque-forming unit) ratio for DENV has been reported to be on the order of 103 to 104: 1 [33,34]. High particle-to-PFU ratios are typical for animal viruses [35], and suggest the presence of many non-infectious particles in a population.
Studies on the effect of population drops on arboviral intra-host genetic diversity have yielded varying results. WNV diversity has been reported to decrease during dissemination from the midgut of Culex pipiens to the salivary gland and saliva [36]. Another study reported that WNV genetic diversity was maintained during infection in the mosquito, and, unlike us, observed considerable maintenance of particular variants (cloned sequences) from one mosquito body compartment to another. The authors concluded that in the mosquito, population drops did not significantly impact WNV genetic diversity, and that the maintenance of diversity was more likely due to variation in the input virus population rather than the generation of new mutants [37].
In contrast, intra-host diversity in our DENV dataset was maintained through the generation of a new SNV repertoire at each transmission stage, such that population drops in the mosquito abdomen and salivary glands impact viral repertoire but not overall diversity levels. The high error rate (10−4 [4], approximately one mutation per 11 kb DENV genome) and burst size (~103−104 genomes per cell [38,39]) of the virus means that a new array of SNVs is quickly regenerated. It will be interesting to determine if genetic diversity is also maintained in virus populations in mosquito saliva, which this study did not sequence. Infectious DENV loads (PFUs) in saliva vary greatly from mosquito to mosquito, but are in general much lower (~10-100X) than in the salivary gland or carcass [40,41], suggesting that the population injected into a human will have gone through yet another drop. However, our observation that genetic diversity also remains similar across members of EDEN human-human transmission pairs suggests that diversity is quickly restored upon replication in the human.
We observed that SNVs that were maintained between transmission stages were present at significantly higher frequencies than SNVs that were not maintained. A previous next-generation sequencing analysis of human DENV clinical samples from Nicaragua found that intra-host variants that were also observed at the inter-host level (i.e. in consensus sequences of circulating Nicaraguan DENV strains) were present at higher frequencies than variants that were only observed at the intra-host level [7]. This suggests that the initial abundance of a SNV may affect its chances of being maintained and eventually becoming the consensus base. This initial abundance may be subject to stochastic processes; indeed, non-synchronous infection, where early infecting virions have an advantage, has been reported to be a major contributor towards genetic drift in human immunodeficiency virus (HIV) populations [42].
We were unable to follow transmission from the mosquito back into the human, but the EDEN human-human transmission pairs (thought to be separated by a single mosquito) allowed us to make hypotheses about this process. Similar to the human-mosquito pairs, paired human-human DENV clinical samples also showed predominantly different SNV repertoires. We estimate that the majority of SNVs present in a human sample will fall below our detection limit during passage through the mosquito, where most SNVs appear to be neutral. However, once transmitted to a second human host, selection pressures there may once again drive up the frequencies of SNVs that impact virus fitness in the human. Our observation that several viral genes display higher #NS/#S ratios in human- than in mosquito-derived samples is consistent with this idea, but SNVs would have to be followed over a chain of multiple linked transmissions in order to further investigate this.
Over multiple rounds of replication in the human or mosquito, viral populations grow exponentially and generate a range of low frequency variants that are subject to selection pressures. Our data suggest that variants in the prM, E, and NS1 genes experience stronger immune pressures in the human than in the mosquito, consistent with the fact that these gene products are known targets of the human antibody response (reviewed in [24]), which has no equivalent in mosquitoes. The envelope protein E is the main antigen on the virion surface and the target of neutralizing antibody [43,44]. Intra-host genetic diversity in the various E protein domains has been reported to positively correlate with immunogenicity in humans, suggesting that immune-driven selection occurs even during short-term, acute dengue infection [7]. We further observed E mutational hotspots in human- but not mosquito-derived DENV populations, leading us to hypothesize that the antibody response in humans is a major driver of viral diversity.
The mosquito immune response lacks an adaptive arm, but is able to mount potent responses against virus, bacteria, and fungi through the Toll, IMD, and JAK/STAT immune signaling pathways (reviewed in [45]). Compared to the situation in vertebrates, relatively little is known about the molecular mechanisms by which these pathways act against DENV, and a better understanding is required before selection pressures on DENV in the vector can be characterized.
In addition to the classical immune signaling pathways, RNAi is also a major antiviral defense mechanism in the mosquito [40,46]. WNV intra-host genetic diversity is thought to be driven by RNAi, with the parts of the WNV genome most likely to be targeted by RNAi also being the most diverse [8]. A separate study made use of artificially diverse WNV strains to show that high intra-host genetic diversity was associated with increased viral fitness in Culex mosquitoes [47]. Although it has been proposed that purifying selection is relaxed in the Culex mosquito host for WNV [10,48], these studies suggest that diversity may be actively selected for in insect vectors. This idea is supported by our observation that almost entirely new DENV SNV repertoires were generated at each transmission stage, but requires further testing in the DENV-Ae. aegypti system.
We found no evidence for differences in selection pressures between the mosquito abdomen and salivary gland. Still, because the SNV sample size was small, this does not rule out differences in immune pressures between body compartments, and #NS/#S ratios did diverge for several genes. A potential driver of differing immune pressures is the midgut microbiota, which has been shown to impact mosquito physiology and vector competence for human pathogens (reviewed in [49]). Depletion of the gut microbiota increases mosquito susceptibility to DENV [50]; this is thought to occur at least partly because the microbiota trigger a basal level of immune activity [50–52]. It will be interesting to examine the impact of the gut microbiota on DENV diversity.
After phasing SNVs into distinct viral genomes, we observed that in most cases, only one SNV from a predicted variant genome was maintained across transmission stages, with a maximum of two being maintained as a unit. This suggests that SNVs in a sample are spread out across many variant genomes instead of being clustered together on one genome, and is consistent with the RdRp error rate of 1x10-4 [4], which corresponds to approximately one mutation per 11 kb DENV genome.
Despite the growing number of studies characterizing intra-host genetic diversity, the impact of this diversity on virulence or disease severity for arboviruses in vertebrate hosts is not well understood. One study unexpectedly found that mouse morbidity and mortality was negatively correlated with WNV intra-host genetic diversity [10]. The authors suggest that this could be due to the parental virus clone being highly pathogenic, such that most mutations would result in a decrease in pathogenicity. Alternatively, adaptation to mosquitoes during passage could also have resulted in a loss of pathogenicity [10]. Another study found lower intra-host diversity in patients with severe dengue than with mild dengue [12]; however the authors state that it is difficult to know if these differences are the cause or the consequence of disease severity. In other studies, no association was found between DENV intra-host diversity and disease severity [7,14]. It will be important to address the question of whether disease severity is associated with the genetic diversity of the infecting viral population, rather than with diversity after the onset of symptoms.
|
10.1371/journal.pntd.0002387 | A Novel Clinical Grading Scale to Guide the Management of Crusted Scabies | Crusted scabies, or hyperinfestation with Sarcoptes scabiei, occurs in people with an inadequate immune response to the mite. In recent decades, data have emerged suggesting that treatment of crusted scabies with oral ivermectin combined with topical agents leads to lower mortality, but there are no generally accepted tools for describing disease severity. Here, we describe a clinical grading scale for crusted scabies and its utility in real world practice.
In 2002, Royal Darwin Hospital (RDH), a hospital in tropical Australia developed and began using a clinical grading scale to guide the treatment of crusted scabies. We conducted a retrospective observational study including all episodes of admission to RDH for crusted scabies during the period October 2002–December 2010 inclusive. Patients who were managed according to the grading scale were compared with those in whom the scale was not used at the time of admission but was calculated retrospectively. There were 49 admissions in 30 patients during the study period, of which 49 (100%) were in Indigenous Australians, 29 (59%) were male and the median age was 44.1 years. According to the grading scale, 8 (16%) episodes were mild, 24 (49%) were moderate, and 17 (35%) were severe. Readmission within the study period was significantly more likely with increasing disease severity, with an odds ratio (95% CI) of 12.8 (1.3–130) for severe disease compared with mild. The patients managed according to the grading scale (29 episodes) did not differ from those who were not (20 episodes), but they received fewer doses of ivermectin and had a shorter length of stay (11 vs. 16 days, p = 0.02). Despite this the outcomes were no different, with no deaths in either group and a similar readmission rate.
Our grading scale is a useful tool for the assessment and management of crusted scabies.
| Crusted scabies is a severe skin condition caused by a microscopic parasitic mite. It occurs in people whose immune system does not react properly to the mite and it leads to crusting and cracking of the skin and can cause death. The usual treatment for crusted scabies is a tablet called ivermectin combined with anti-scabies skin creams. However, there is no current method of measuring the severity of crusted scabies and thus deciding how long to continue the treatment for. We have developed a grading scale based on examination of the skin, which classifies patients as mild, moderate or severe, and uses this grading to suggest the duration of treatment. We have trialed this grading scale over an 8-year period in 49 episodes of crusted scabies requiring hospital admission, and have found that it leads to a shorter length of hospital stay and treatment, but equivalent outcomes compared to those who were treated without the use of the grading scale.
| Scabies is a parasitic infestation caused by the mite Sacroptes scabiei var hominis. Globally, over 300 million people are estimated to be affected [1]. The mite is endemic in disadvantaged and impoverished communities [2], [3]. In Australia Indigenous people suffer a significant disadvantage in health outcomes compared with non-Indigenous Australians [4], [5], and scabies is endemic in many Indigenous communities in northern Australia, with a recent survey demonstrating a mean prevalence of 13.4% in five remote Indigenous communities [6].
Crusted scabies (also known as “Norwegian scabies”) is hyperinfestation with the Sarcoptes scabiei mite, and is characterized by a non-protective host immune response, the development of hyperkeratotic skin crusts and skin fissuring [7]. It is a severe disease with a significantly higher mortality than ordinary scabies. Unlike ordinary scabies, where there are usually less than 20 mites on the host's entire skin, individuals with crusted scabies can have up to 4000 mites per gram of skin and are extremely infectious to others [8], [9]. Despite the severity of the disease there is significant variability in the clinical presentation, and there is currently no generally accepted method of describing the severity of a crusted scabies infection.
The optimal treatment for crusted scabies has not been subjected to a comparative trial and is generally based on expert opinion [10], [11]. However observational data suggest that the use of multiple doses of oral ivermectin as therapy for crusted scabies can lead to a significant decline in mortality [9], [12], [13]. In an attempt to formalize and improve the treatment of crusted scabies, we developed a grading scale, based on our clinical experience in managing such patients. This was introduced into routine clinical use at our hospital in 2002, and has been used since this time to titrate the duration of ivermectin and topical therapy to illness severity.
Here, we describe the grading scale and our experience with it over the first eight years of its use. We aimed to evaluate the utility of the grading scale, including its correlation with other putative markers of illness severity, the safety of its use and the effect on length of stay and relapse rates.
The study was approved by the human research ethics committee of the Menzies School of Health Research and Northern Territory Department of Health.
350 bed tertiary referral hospital in the tropical Northern Territory, Australia, serving a population of approximately 150,000 people spread over an area of 500,000 km2, including many remote Indigenous communities. Local policies encourage the hospitalization of patients with crusted scabies for clinical management, as well as environmental health input to address the risk of ongoing transmission in an index patient's household. The standard treatment protocol for crusted scabies includes prolonged hospitalization in a single room with contact precautions, the use of topical benzyl benzoate plus 5% tea tree oil 2–3 times per week [14], multiple doses of oral ivermectin (as described below), topical keratolytics, systemic antibacterial drugs where judged clinically necessary, and attention to medical comorbidities.
All patients admitted to our hospital with a discharge diagnosis of crusted scabies between 1st of October 2002 and 31st of December 2010 were included in the study. Crusted scabies was diagnosed based on the clinical opinion of an Infectious Diseases specialist, supplemented by skin scrapings demonstrating S. scabiei mites on microscopy.
The grading scale for crusted scabies is shown in Figure 1. It is based on clinical assessment in four key areas: the distribution and extent of crusting; the depth of crusting; the degree of skin cracking and pyoderma; and the number of previous episodes. This scale was developed in 2002 by two of the authors (JD and BC) for use with all patients hospitalised with crusted scabies. It was partly based on previous local experience that multiple doses of ivermectin in addition to topical treatment were more effective than topical treatment alone for the treatment of crusted scabies [9]. Other studies have confirmed the efficacy of the combination of ivermectin and topical therapy for crusted scabies [13], [15], [16]. During the study period medical staff managing patients with crusted scabies were encouraged but not compelled to use the grading scale to guide management. Therefore we were able to compare those patients in whom the grading scale was applied at the time of the patient's clinical presentation to those in whom the grading scale was not used and then calculated retrospectively by the authors.
We reviewed clinical notes, bedside charts and the hospital's clinical pathology database for each patient using a standardized case record form. We collected data on demographics, comorbidities, disease severity, grading scale and outcomes. Where the grading scale had not been prospectively documented, we calculated it based on the detailed clinical information found in the medical record. Each admission (rather than each individual patient) was counted as a discrete episode. Where a patient had more than one admission during the study period, it was only counted as a separate episode if at least 30 days had elapsed from the previous date of discharge. Iatrogenic immunosuppresion was defined as the use of any of the following medications within the past 3 months: prednisolone ≥0.5 mg/kg/day or equivalent for at least 14 days; immunosuppresion for solid organ transplant; cancer chemotherapy; immunosuppressive monoclonal antibody use; any other use of azathioprine, methotrexate, leflunomide, cyclosporine, mycophenolate, or cyclophosphamide. Hazardous alcohol use was defined as an average of >4 standard drinks per day for a man or >2 for a woman. Chronic renal disease was defined as an estimated glomerular filtration rate of less than 30 ml/min, or the need for dialysis.
Data were entered into a purpose-built database using Epidata v 3.0 and were analysed using Stata version 10 (Statacorp, College Station, Texas, USA). Categorical variables were compared using Fisher's exact test, and continuous using Mann-Whitney-U test. Correlations were assessed using Spearman's rank correlation. P values of <0.05 were considered significant
There were 49 admissions for crusted scabies in 30 patients during the eight year study period. Of the episodes, 49 (100%) were in Indigenous Australians, 29 (59%) were male and the median age at the time of the first admission within the study period was 45.4 years (Table 1). Most of the patients lived in remote areas, and iatrogenic immunosuppresion was rare.
All patients received at least one dose of oral ivermectin (with a mean of 5.2 doses, and a range of 2 to 10). 47 patients (95%) were treated with topical benzyl-benzoate in combination with 5% tea-tree oil, and the remainder with topical permethrin. In addition, all patients were treated with topical Calmurid (lactic acid and urea in sorbolene cream, used as a keratolytic). Systemic antibiotics were used in 38 (79%) of episodes.
According to the grading scale, 8 (16%) episodes were mild (grade 1), 24 (49%) were moderate (grade 2), and 17 (35%) were severe (grade 3). Seven episodes (14%) were complicated by bacteraemia, with the causative organism being Staphylococcus aureus in 6 patients, and a mixed infection with Group A streptococcus and Escherischia coli in 1. The disease severity according to the grading scale did not correlate with the proportion of patients with bacteremia, or with the peak plasma C-reactive protein during the admission (table 2). However, there was a non-significant trend towards lower nadir plasma albumin and longer hospital stay with higher severity (table 2). No patients in this cohort died during the hospital admission, but a substantial proportion (47%) required readmission for crusted scabies within the eight year study period. Readmission was significantly more likely with increasing disease severity, with an odds ratio (95% CI) of 5.9 (0.7–55.9) for moderate disease compared with mild, and 12.8 (1.3–130) for severe disease compared with mild.
There was no significant difference in age, gender, location of residence or comorbidities between those patients who had the severity score calculated at the time of admission (n = 29) and those who did not (n = 20). Episodes where the grading scale was calculated at the time of admission had a significantly shorter length of stay, and received fewer doses of ivermectin than those not managed using the grading scale (Table 3). Despite this their outcomes were no different, with no deaths in either group, and a similar readmission rate in the two groups.
This is the first published description of a clinical severity grading scale for use in patients with crusted scabies. The use of this grading scale in our setting is associated with good outcomes despite shorter hospital stays and less ivermectin use compared with those managed without the use of the grading scale.
Crusted scabies is a severe disease with significant morbidity and mortality which is more prevalent in communities such as remote-dwelling Australian Indigenous people [2]. Crusted scabies is usually reported as occurring in patients who are immunosuppressed, either iatrogenically [17], [18], [19], [20] or by retroviral infection [21], [22], [23]. In our cohort there was a high rate of hazardous alcohol use, diabetes and chronic renal disease, but only 16% of episodes were associated with iatrogenic immunosuppresion or HTLV-1 infection. This reinforces the findings of previous studies that, in Indigenous Australians, the majority of people with crusted scabies do not meet the generally accepted definitions of significant immunosuppression and suggests that the immune defect in patients with crusted scabies is subtle and probably multifactorial [8].
Our grading scale did not correlate with many of the putative measures of disease severity we used (CRP, ICU admission, bacteraemia). However, these factors are really measures of the sequelae of crusted scabies and there is no generally accepted single marker of disease severity in this setting (hence the need for the clinical grading scale). The degree of systemic inflammation and risk of bacteraemia are likely to relate to multiple factors, including the patient's immune responses, the depth of skin cracks, the degree of bacterial skin colonization and the patient's underlying comorbidities. Hence this lack of correlation does not necessarily imply that the grading scale does not reflect disease severity.
Long hospital stays (particularly those involving single rooms and contact isolation) are expensive to the health care system, and frustrating for patients. The 5 day decrease in length of stay which we observed with the use of the grading scale, with no increase in relapse rates, is substantial and represents a large cost saving. Another potential advantage of our grading scale is that it may help guide the duration and type of therapy for those clinicians who are less experienced in the management of crusted scabies. Given that crusted scabies is a rare condition in most settings, the utility of such a grading scale for the average clinician is a good reason for its use.
Ivermectin is an orally administered semi-synthetic macrocyclic lactone antibiotic. It is approved for the treatment of scabies in France but is not licensed for the treatment of scabies in the United States, United Kingdom or Australia. However is it commonly used off-label for the treatment of scabies in Australia. Ivermectin does not sterilize scabies eggs so multiple doses are recommended to kill newly hatched mites [10]. Ivermectin has been associated with adverse effects in some studies, which emphasizes the benefit of using a grading scale that allows for the titration of the total dose of ivermectin and in our study a possible reduction in number of doses in patients with milder disease. Intensive ivermectin use may also increase the probability of the mite developing resistance especially in patients with multiple relapses [24].
This study was planned prospectively, but the grading scale had to be calculated retrospectively in 40% of patients, introducing possible inaccuracies in the calculated scores. Fortunately, a detailed clinical assessment was recorded in the medical record for all patients, and thus we were able to calculate the score for all patients without having to interpolate missing data. Despite this, the score calculated at the time of clinical assessment is likely to be more accurate; retrospectively calculated scores may have underestimated severity in certain areas such as degree of crusting and shedding. However, this would not affect the overall conclusions regarding the use of the score, as the outcome measures were not the scores themselves, but objective measures including length of hospital stay and need for re-admission. Our population differs substantially from some others in whom crusted scabies has been reported to occur. Hence it is important for the grading scale to be studied in other populations before concluding that it is useful in all settings.
We have described a simple clinical grading scale to aid in the management of patients with crusted scabies. If validated in other settings, its use is likely to improve the management of crusted scabies and may lead to a decreased length of required hospital stay and of ivermectin treatment, without compromising outcomes.
|
10.1371/journal.pcbi.1000906 | Current Estimates for HIV-1 Production Imply Rapid Viral Clearance in Lymphoid Tissues | It has recently been estimated that a single HIV-1 infected cell produces between and more than viral particles over its life span. Since body-wide estimates of the ratio of free virus to productively infected cells are smaller than and much smaller than , individual virions must be cleared rapidly. This seems difficult to reconcile with the fact that most of the total body virus is trapped on follicular dendritic cells where it can survive for many months. It has also been difficult to reconcile the vast difference in the rates at which the virus is cleared from the blood in rhesus macaques and in chronically infected patients. Here we attempt to reconcile these seemingly contradictory observations by considering the virion clearance rate in various organs and the virion exchange rates between them. The main results are that the per capita clearance rate of free virus in lymphoid tissue should be fast, the virion exchange rate between lymphoid tissue and the blood should be slow, and the comparatively slow previous estimates for the virion clearance rate from the blood correspond to the rate of virion efflux from the blood to other organs where the virus is ultimately cleared.
| A human cell that is infected with the AIDS virus HIV-1 may produce more than new viral particles over its short life span. In patients chronically infected with HIV-1, one can estimate that on average there are much less than free viral particles per productively infected cell. This suggests that the rate at which individual virus particles are cleared from the body must be fast. Most of the virus is long-lived, however, because it is trapped on follicular dendritic cells. We attempt to reconcile these seemingly contradictory observations by estimating the virion clearance rate in various organs, and the virion exchange rates between them, using a mathematical modeling approach. We find that individual virus particles are cleared rapidly from the lymphoid tissue, and that the rate at which virus is exchanged between lymphoid tissue and the blood is slow.
| The major targets of HIV and simian immunodeficiency virus (SIV) infection are CD4 T cells [1]. During the acute stage of infection large numbers of resting and memory CD4 T cells disappear from the lymphoid tissues and mucosal layers, particularly in the gut, by direct infection and by bystander effects. Lymphoid tissue remains the primary site of infection after acute infection has resolved and the viral load approaches a steady state called the “set-point” [1]. Chronic infection is characterized by a set-point viral load, and rapid turnover of productively infected cells [2]–[4]. To maintain this steady state requires a balance between virus production and clearance, and between target cell production and death. Combining image analysis with in situ hybridization in lymphoid tissue from patients chronically infected with HIV-1, the total number of productively infected CD4 T cells has been estimated to be cells, and the total number of HIV-1 particles has been estimated to exceed virions [5], [6]. At steady state, one could naively conclude that a single productively infected CD4 T cell should therefore account for a viral load of approximately 500 virions. We shall show below that the situation is more complex.
To understand viral production and clearance better, one needs to consider the current quantitaive estimates of viral production and clearance, as well as where these processes are occurring in the body. Most of the total body virus is located in lymphoid tissues, typically in association with follicular dendritic cells (FDCs) [1], [5]. FDCs trap virus and retain it on their surface for many months [7]–[11]. The FDC associated virus pool fills up during early infection, i.e., not later than a few days after onset of symptoms, and does not expand over the course of chronic HIV-1 infection [12]. Although most of the virus resides in this fairly constant storage on FDCs, the store rapidly declines during antiretroviral treatment (ARV) [13], suggesting the existence of a quasi steady state between free and FDC associated virus in the lymphoid tissue [8]–[10], [14]. In a mouse model it was shown that a small fraction of HIV-1 persisted on FDCs, and remained infectious over a period of 9 months [7]. A recent study confirmed the long-lived nature of this reservoir in humans [11].
Viral clearance rates have been estimated in the blood, and different techniques have yielded a variety of estimates [4], [15]–[17]. Rapid clearance rates with half-lives of 3–4 minutes were found after infusion of SIV into the blood of uninfected and infected rhesus macaques [16], [17]. In patients chronically infected with HIV-1, more than 10-fold slower clearance rates were found using plasma apheresis to increase viral clearance, and hence perturb the viral set point [15]. By plasma apheresis approximately particles were removed over a period of two hours, and this reduced the viral load in blood, with a nadir that was approximately half of the original viral load [15]. The mere fact that the removal of less than 1% of the total body virus lead to an observable decline in the plasma virus load [15], suggests that the exchange of virus between the lymphoid tissues and the blood cannot be rapid [14]. Further, sequence analysis of virus in splenic white pulps suggests that virus trapped on FDC is produced locally [18], supporting the notion of slow viral exchange between blood and lymphoid tissue.
Virus production rates have also been estimated by several techniques. Because in other lentivirus infections, most notably visna virus, intracellular viral DNA levels increase approximately exponentially and then virus appears to be released rapidly [19], the term burst size is commonly used to describe the total amount of virus produced by an infected cell [20]. If one knows the burst size, , and the average lifespan of a productively infected cell, , then the viral production rate, , is given by . Note that is then the average rate of virion production over the lifespan of a productively infected cell. Current estimates suggest d [21]. Thus, the burst size corresponds to the daily viral production rate.
Recent studies have combined image analysis with in situ hybridization to estimate burst size. Assuming that the maximal HIV RNA count in a cell corresponds to the burst size, Haase et al. [5] estimated a production rate of approximately a hundred particles over the life span of a productively infected cell. Hockett et al. [22] quantified more precisely the amount of viral RNA (vRNA) per cell by a PCR technique. They found an average of 3900 (range 3162–5011) vRNA copies per infected cell, and because of limited variation in the number of copies per cell, they concluded that viral production is a few thousand virions per cell, and also assumed that bursting was an all-or-none phenomenon [22]. However, the estimates by Haase et al. [5] and Hockett et al. [22] are based on measuring HIV RNA at a single time point. If infected cells continue to produce virus over an extended period, then these estimates would be underestimates of the true total cellular production of virus.
One can also attempt to measure burst size by directly imaging the extracellular viral particles surrounding an infected cell [23]. Using this method, Reilly et al. [23] found and copies of HIV-1 RNA surrounding infected activated and resting CD4 T cells, respectively, in the lymphoid tissue of acutely SIV-infected rhesus macaques. They then fitted a five parameter model to this data, with three of the parameters describing the rate of viral production as a function of time since infection (see Methods), and the remaining parameters describing the rate of exponential decay of cells producing virus, and the rate of loss of viral particles. Using this model, they estimated that the half-life of virus located around CD4 T cells producing virus in lymphoid tissue was approximately three hours. However, this half-life combined diffusion of virus out of the local area and true virion clearance [23]. Even if all loss was due to clearance, a three hour half-life corresponds to a per virion clearance rate of d. With this estimate, Reilly et al. [23] calculated median production rates of approximately 1500 and 1400 viral particles per activated cell, and of approximately 650 and 3400 viral particles per resting cell, depending on two different assumptions for the half life of productively infected resting cells (see Methods).
Finally, the most direct estimates for the total amount of virus produced per infected cell was achieved using single-cycle SIV to infect PBMC which were placed back in uninfected rhesus macaques. By measuring the total amount of virus produced and accounting for clearance, this experiment yielded a total production of approximately (range –) virions per infected cell [24]. Because productively infected cells have a lifespan of about one day, the cellular burst size estimates of Chen et al. [24] imply daily production rates of approximately virions.
Summarizing, the latest production rate estimates converge on a few thousand to approximately virions per productively infected cell [22]–[24]. The production rate estimates of Reilly et al. and Chen et al. depend on the viral clearance rate, . The 10-fold range in the estimated production rates is at least partly due to differences in the clearance rate used in the calculations. Reilly et al. [23] estimate that d in lymphoid tissue, while Chen et al. [24] used a previous estimate of d in the blood [15]. Since, our main result will be that the clearance of free virus in lymphoid tissue should be fast, and that the observed clearance from the blood is not clearance but the rate of efflux to other organs, we will vary the production rate in our analysis and consider to particles per cell as potential realistic estimates. Finally, note that different cell types, e.g., infected macrophages, may have different production rates than infected T cells. Here we consider that the vast majority of virus is produced by infected CD4+ T-cells [4], [25], and hence use estimates of production from those cells.
In one earlier modeling study a production rate of several thousand particles per cell was shown to be consistent with a viral half-life of 3–4 hours in the lymphoid tissue [14], suggesting that the recent estimates of 10-fold higher production rates [24] imply even shorter half-lives. However, large total viral production per infected cell [22]–[24] and the short viral half-lives they imply [14], seem difficult to reconcile with the suggestion that most of the virus in the lymphoid tissue is long-lived and in association with FDCs [5]. The problem of balancing production with clearance can be introduced by a simple calculation that assumes the body is a single well-mixed compartment. An order of magnitude estimate for the total number of productively infected cells in a human is cells [5], [6]. For a human with a viral load of approximately particles ml of plasma, and an estimated total of 15 liters of extracellular body water in which virus could distribute, one estimates that there are a total of free virus particles in extracellular fluids (i.e., only about 2% of the estimated total body load) [4]. Requiring steady state in the conventional model for virus production, i.e., , with a production rate of viral particles per infected cell, , per day, and a steady state of free virus particles and productively infected cells, one would need a per virion clearance rate of d, which is much higher than published estimates for the viral clearance rate in humans [15], but resembles the rapid clearance rate observed in rhesus monkeys [17]. Even if d then d is needed to balance production, which is still larger than the current clearance rate estimate in humans [15].
In this paper we attempt to reconcile the various estimates for the viral clearance rate, the viral production rate, and the amount of long-lived virus trapped on FDCs within one modeling framework in order to test whether there is a consistent interpretation explaining all observations. To do so we introduce compartmental models to analyze recent experimental data on viral clearance in various organs, and estimate the rates at which virus is exchanged between them.
A simple and direct approach to estimate the clearance rate of virus from the blood is to infuse virus particles into the blood, and monitor their disappearance by taking frequent blood samples. Zhang et al. [16], [17] administered large amounts of SIV to infected and uninfected rhesus macaques by an intravenous bolus injection ( to viral particles), or by constant intravenous infusion ( viral particles min). From the rate at which virus was lost from the plasma afterwards, plasma half-lives of 3–4 minutes were estimated [16], [17]. These half-lives were similar in infected and uninfected monkeys. Virus did not appear to be lost from the plasma by binding to erythrocytes, PBMCs, granulocytes, or platelets because there was no evidence of virion binding to these cell types [16]. In another experiment, viral clearance in various organs was tracked by injecting radioactively labeled SIV into macaques, and measuring the percentage of radioactivity and of SIV RNA persisting in various organs after two hours [17]. Because 30% of the radioactivity, and only 0.053% of the injected SIV RNA, was recovered from the liver (see Table 1), it was concluded that the liver plays a major role in viral degradation [17].
If most of the viral degradation indeed takes place in organs such as the liver [17], most of the measured clearance from the blood would be efflux from the blood into the organs. Therefore, we write the following simple model for the amount of radioactive virus in the plasma, , and in a particular organ, ,(1)where is the daily efflux from the blood, is the fraction that arrives in the particular organ, and is rate of clearance in the organ. We neglect the flux from the organ back to the blood during this 2 hr experiment because this simplifies the analyses, but also because the results of the bolus injection and the continuous virus infusion experiments [16], [17], yielded similar disappearance of virus from plasma in uninfected and already infected monkeys, indicating that this flux is very small in these short-term experiments.
If is the total amount of virus injected into the plasma, then(2)After two hours, the fraction of infused viral RNA found in the organ is(3)where the converts 2 hrs into a per day timescale. The fraction of radioactivity ending up in the organ after the two hour experiment, , is assumed to be equal to the fraction of virus entering (and presumably degraded in) that organ, i.e.,(4)
Using the estimated clearance rate from the blood of 0.2 min [17] as the efflux rate, d, we estimate the fraction, , and the clearance in the organ, , from the Zhang et al. [17] data shown in Table 1. Since in Eq. (4) the term , the fraction of measured radioactivity in the organ, , determines the parameter in the model. Substituting the fraction of SIV RNA in the organ, , the estimated efflux, , and the fraction of radioactivity, , into Eq. (3), one can numerically solve for the clearance rate constants, , in the four organs (Table 1). The estimated clearance rates in the various organs vary from d in the liver to d in the lymph nodes. The latter is only 2-fold faster than the clearance rate of d estimated by Reilly et al. [23] for lymphoid tissue.
Only 40.5% of the total radioactivity was recovered in the monkey 2 hr after injection. This could be due to a loss of radioactivity by viral degradation and removal of labeled molecules, or to accumulation of SIV in other body compartments, such as the gastrointestinal tract, that were not examined [17]. The former we can correct for by re-normalizing the radioactivity data so that the total is 100%. This correction doubles the estimate clearance rate in lymph nodes, and has a smaller effect on the other clearance rates (see Table 1). If the virus unaccounted for by the radioactivity data is ending up in other organs, the clearance rates based upon the uncorrected radioactivity data should be valid. Interestingly, percentages of SIV RNA were also measured in ileum, cecum, duodenum and rectum in other monkeys, and adding these data from the gut to the “Others” class hardly increased the amount of SIV RNA in that class [16]. This would argue that only a minor fraction of the injected virus ends up in the gastrointestinal tract, and/or that the clearance rate in the gastrointestinal tract is much larger than in the other organs so that SIV RNA is not found there. Unfortunately, there is no radioactivity data for the gastrointestinal tract to distinguish between these two possibilities, and we estimate the viral half-lives in the various organs by the ranges indicated in Table 1, as obtained from the corrected and the uncorrected radioactivity data, respectively. Summarizing, modeling the radioactivity data provides estimates of viral clearance rates between d and d in various organs (Table 1).
HIV-1 clearance rates from plasma have been estimated in chronically infected patients by plasma apheresis over a period of two hours. Plasma was removed at a rate of 39 mL per min, and was replaced by an equivalent volume of isotonic saline containing 5% albumin. On average, this procedure removed a total of approximately particles, and resulted in a nadir of virus equal to half the initial viral load [15]. The following model, formally equivalent to the one presented in Ramratnam et al. [15], was used to fit the data:(5)where is the rate of virus influx from the lymphoid tissue, is the normal efflux from the blood, and is the additional rate of virus removed from the blood due to plasma apheresis. It was assumed that over the course of the two hour experiment the flow of virus into the blood, , remains constant, and it was found that the rate of viral efflux in four patients ranged from to 36 d, with an average of d [15]. The rate of virus influx, , at steady state is estimated by multiplying the plasma efflux rate, , by the initial virus load, and varies from to particles d, with an average of particles d. The mere fact that the removal of less than 1% of the total body virus over a period of two hours led to significant declines in the plasma virus load (Table 2 and Ramratnam et al. [15]), suggests that the plasma virus pool is not rapidly replenished from the lymphoid tissue or other organs[14]. This observation also supports our neglecting virus return from organs back into the plasma in Eq. (1).
It is worth noting that the estimate for the efflux rate in humans of HIV-1 from plasma is more than 10-fold slower than the estimated plasma efflux rate of SIV in rhesus monkeys [15]–[17], which could reflect a true difference between these two species. Alternatively, it could be that the plasma clearance rate in the four patients studied by Ramratnam et al. [15], all of which had high viral loads, is slower than in the monkeys studied which had much lower viral loads [16], [17]. A potential mechanism for the more rapid clearance in monkeys with low viral load could be the rapid attachment of virus to various receptors on blood born cells, whereas in patients with chronic high viral loads these receptors could be saturated and bind less virus (see Discussion).
Since most virus production takes place in the lymphoid tissue, we modify a previously published compartmental model [14] to rewrite the fixed source in Eq. (5) into a term depending on the amount of virus in lymphoid tissue (LT). We proceed by considering four viral compartments: virus in organs other than LT, , virus in the plasma, , free virus in the lymphoid tissue, , and virus bound to FDCs in lymphoid tissue, . The model has two clearance rates, and , for the rate of clearance of free virus in LT, and in other organs (like the liver and lung), respectively. As before, there is no clearance in blood. Virus bound to FDCs is considered to be long-lived [7], and virus in the plasma is considered to be lost by migration to organs or LT. We allow for influx of free virus into the plasma from the LT with rate constant , because now we are modeling a long term process, and efflux from the plasma with rate constant . A fraction of the virus leaving the plasma will return to the LT, the rest is cleared in organs like the liver and lung. This compartment model is represented by the following equations:(6)(7)(8)(9)where is the number of productively infected cells in the LT.
In this model we make a distinction between clearance and efflux. We speak of efflux when the number of virus particles is conserved, and speak of viral clearance in an organ only if the virus is being degraded there. Thus, these equations assume that there is no viral clearance from the blood; there only is efflux to the lymphoid tissue and to other organs that have viral clearance rates, and , respectively. Mixing SIV with fresh blood taken from monkeys provided no evidence for viral degradation within plasma ex vivo [17], and for several viruses most of the clearance takes place in liver and spleen [16]. The model can easily be modified to allow for viral clearance from the blood, e.g., by decreasing the term in Eq. (6). This would not affect our results, however, because Eq. (6) represents a sink that does not affect the other three compartments of the model. Technically, the “viral clearance” rates from the blood estimated previously by the apheresis [15] and infusion [16], [17] experiments, cannot distinguish between clearance and efflux, and we will let these estimates represent the efflux rate, , in Eq. (7).
The parameter in Eqs. (8) and (9) is the average dissociation rate of virus bound to FDCs. is the maximum number of binding sites on FDCs that HIV-1 can attach to, and is the rate constant for the association of virus with an FDC binding site. These binding sites include complement receptors and Fc receptors [8], and possibly other receptors like DC-SIGN [26]. The dissociation process is complicated and depends on the number of bonds by which virus is bound to the FDC [8]. When most virus particles have multiple bonds holding the virus to FDC, which makes the dissociation slow, and probably accounts for the long half-life of a fraction of the bound virus [7], [8], [11]. When is large most viruses will have few bonds holding them to FDC and dissociation should be more rapid. We can describe this phenomenologically with a Hill-function such that the effective virion dissociation rate constant increases with (10)where and are constants with , and when .
In a typical patient the FDC pool seems saturated, i.e., , and most virus is expected to be monovalently bound with a dissociation rate estimated as /sec [8]. Image analysis combined with in situ hybridization suggested that most of the virus in the lymphoid tissue is associated with the FDC network [5]. The FDC associated virus, , fills up early in infection [12]. During chronic infection we therefore assume , and this large pool of FDC associated virus is viewed as a filled store in quasi steady-state that contributes little to the total body viral clearance. When the FDC pool is close to being saturated, the steady state of Eqs. (6) to (9) corresponds to(11)where if , and if .
To consider the total virus load we add Eqs. (6)–(9) yielding,(12)where is the total body virus load. For a chronically infected patient assumed to be in a steady state we substitute from Eq. (11) to obtain(13)where the total body clearance rate, , is the total steady state rate of clearance taking place in organs like the liver and the lymphoid tissues. Since our aim is to estimate , we rewrite Eq. (13) as(14)where we have again used from Eq. (11).
We can use the relationships between and the steady state levels in Eq. (14) to study what clearance rates would be required to balance production in a number of typical situations. Since there is ambiguity on the rate of efflux from the blood, , we use the efflux estimates from the plasma apheresis experiments [15], i.e., to 36 d, and from the rhesus monkeys [16], [17], i.e., d, as lower and upper bounds to create examples of how viral production and clearance could be balanced in hypothetical patients. If we use a middle value for the fraction of plasma virus returning to the LT, e.g., , we obtain an upper estimate of d and take a lower estimate of d. To estimate the ratios between the variables in Eq. (14) we pick an example of a patient in a chronic steady state with total body counts of productively infected cells, virus particles in the peripheral blood, and virus particles in the lymphoid tissue. Finally, because Hockett et al. [22] did not detect virus associated with FDCs in almost half of their patients, and measured 10-fold higher total amounts of virus in lymph nodes from those patients where they could detect virus on FDC, we consider two possibilities. To model a “typical” patient where most of the virus is associated with FDCs, we let 90% of the lymphoid tissue virus be associated with FDCs, and obtain that . To model patients with a smaller pool of FDC associated virus, we also consider the possibility that half of the LT virus is bound to FDCs, i.e., , which amounts to . This allows us to study how the estimates for the viral clearance rate in LT depend on the fraction of virus bound to FDCs.
For the more “realistic” example, Fig. 1a, where most of the virus is associated with FDCs, our estimate of the per capita clearance rate in LT, , depends strongly on the production rate , and a large production rate, e.g., virions per cell per day, requires rapid clearance of virus in the lymphoid tissue, i.e., per day to maintain a steady state level of virus (Fig. 1a). The recently proposed production rates of more than viral particles per infected CD4 T cell would require LT clearance rates of per day (Fig. 1a). When Chen et al. [24] estimated these high production rates they were conservatively assuming that d. Because their estimated production is proportional to the assumed clearance rate, our new results suggest that the true production could be even higher. In cases where less virus is associated with FDCs (Fig. 1b), we find a similar relation between the clearance rate in LT, , and the production rate, , but the required clearance rate is approximately 5-fold smaller because we allow for 5-fold more free virus, i.e., (Fig. 1b). In this case, for realistic virion production rates per cell, e.g., per day, the estimated clearance rate is fairly independent of rate of efflux from the blood (); see Fig. 1.
Analysis of the plasma apheresis experiments in humans provided estimates for the influx of virus from the LT into the blood (Table 2). This can also be calculated from the quasi steady state of Eq. (7), i.e., for the influx of virus from the LT into the blood, one obtains that(15)where the efflux has the two estimates of d and d in patients and macaques, respectively. For the case when 90% of the lymphoid tissue virus is associated with FDCs, i.e., , this means that d. Assuming an equal distribution of free and FDC-bound virus in lymphoid tissue, i.e., , one obtains d. Taking these two cases as extremes, the daily influx of virus from the lymphoid tissue into the blood would be between virus particles per day, i.e., about to particles per hour. These estimates are in good agreement with the daily influx estimated from the plasma apheresis experiments (Table 2). Finally, for these estimates of , the clearance rate in LT should approach as the term in Eq. (14) is negligible.
To test whether the full model (Eqs. 6–9) is consistent with the plasma apheresis experiments, we make reasonable guesses for the other parameters of the model. Allowing rapid filling of the pool of virus bound to FDCs we set the number of FDC binding sites and d. With these parameters the initial “on” rate when all FDC sites are free is d (10 s). To have 100-fold more free virus in LT than in blood, i.e., (see Eq. (11)), we set d and d. To have about free virus particles in the LT, we set particles d and d (see Eq. (11)). During a two hour plasma apheresis experiment we transiently add an estimated efflux of d during apheresis to Eq. (7) (like we do in Eq. (5)). For these parameters the model results mimic the plasma apheresis experiments in the blood (Fig. 2a), reducing the viral load in plasma by 30% and accumulating a total of virus particles. Free and bound virus in LT are hardly affected (Fig. 2b). Very similar results are obtained when we increase production and viral clearance in the LT 10-fold to particles d and d (not shown). Explaining the plasma apheresis experiments in humans therefore indeed requires a viral efflux half-life from the plasma of about half an hour [15].
Choosing the efflux rates estimated in monkeys [16], [17], we have also set d and d, which delivers a very similar steady state as that shown at time zero in Fig. 2. Simulating plasma apheresis by setting d for two hours a total of virions are removed, but the plasma virus load decreases by 5% only (not shown). This is a natural result because the additional clearance of d is small compared to the normal efflux from the plasma of d. We therefore predict that if plasma apheresis experiments were repeated in SIV infected rhesus macaques, the viral load in the plasma would hardly be affected.
We have shown that the balance between viral production and viral clearance implies rapid per capita viral clearance rates in lymphoid tissues. The larger the viral production rate per productively infected cell, and the more virus that is bound to FDCs (see Fig. 1), the larger the per capita clearance rate of free virus in lymphoid tissue must be to balance viral production. Recent estimates of a burst size up to virions per cell and a productively infected cell life span of about a day [21], [24], imply viral clearance rates in the lymphoid tissue of to d (Fig. 1).
In our modeling work the rate of virion clearance, , is assumed to be a constant. This is equivalent to assuming that clearance occurs by a first-order process or that virus clearance can be described by an exponential decay. However, it is possible that viral clearance obeys more complex laws. Comparing viral loads in lymph nodes and plasma from 9 patients at relatively advanced stages of disease, Hockett et al. [22] demonstrate that the plasma viral load increases faster than proportional with the number of productively infected cells. One possible explanation is that the viral clearance rate decreases or saturates when the viral load increases. However, this explanation remains speculative as it requires a 100-fold decrease in the clearance rate at their highest viral load, and it seems unlikely that there is such a tremendous variation in the clearance rate [22].
The efflux and/or clearance rate of SIV from the blood of uninfected monkeys and of infected monkeys with a low viral load [16], [17] is about 10-fold higher than that of HIV-1 measured by plasma apheresis experiments in chronically infected patients [15]. Because the additional removal ( in Table 2) realized in the plasma apheresis experiments is also 10-fold smaller than these efflux rates in monkeys, plasma apheresis would have hardly any effect if humans were to have efflux rates similar to these monkeys (this expectation was confirmed by computer simulation). As discussed above this 10-fold difference in the estimated efflux rate from the plasma could reflect a true species difference. A speculative alternative is that efflux from the plasma hinges upon attachment of virus to various receptors on blood born cells, like CCR5 and CD4 on various cell types, gp340 on macrophages [27], DC-SIGN on dendritic cells [26] and DARC on red blood cells [28], [29]. Because the monkeys in these experiments had much lower viral loads than the four patients studied by apheresis [15]–[17], most of the receptors could be free in monkeys and occupied with HIV-1 in chronically infected humans with a high viral load. However, this remains speculative because Zhang et al. [16] found negligible amounts of virus on erythrocytes, peripheral blood mononuclear cells (PBMC), granulocytes, and platelets. Moreover, note that the estimated clearance rate in the liver, the organ that appears to be responsible for most of the peripheral viral degradation [17], is reasonably close to the high clearance rates of free virus in lymphoid tissue that we estimate to be required for balancing the total body virus production.
Finally, it may seem that the notion of a large FDC store of bound virus is incompatible with the rapid viral clearance rates seen in spleen and lymphoid tissues of uninfected macaques [16], [17]. In other words, if the FDCs were trapping and maintaining the virus in lymphoid tissues, then after radiolabeled virus was injected, the percentages of radioactivity and SIV RNA found in these tissues should have been similar, whereas a 2-fold difference was observed, i.e., 3% vs. 1.4% (Table 1). There are at least two possible explanations for this difference. First, virus is reversibly bound to FDC [8] so radiolabeled virus could dissociate and the virus could then be degraded. Alternatively, the clearance after bolus infusion of SIV in monkeys was studied over a time window of just two hours [16], [17], and one could speculate that most of the added virus in these short experiments fails to bind FDCs, and could therefore be cleared rapidly.
Rapid viral clearance in lymphoid tissue is not surprising. The lymphoid tissue contains more than CD4 T cells, i.e., the ratio of virus to CD4 cells in the LT is approximately 1∶1, and virus particles will bind CD4 T cells and macrophages, and defective virus particles will be “cleared” by non-productive infection. Phagocytic cells that are also abundantly available in LT may clear virus via various types of receptors, like Fc, complement, and DC-SIGN on dendritic cells [26]. Since the process of virus binding cell surface receptors is relatively fast, it can readily account for the rapid clearance rates that we derive from balancing total body production with total body clearance. Finally, during chronic HIV-1 infection the long-lived pool of virus on FDCs should be in steady state, and thus not contribute to the actual clearance rate of the virus in lymphoid tissue (see Eq. (12)).
In a previous paper we showed that the rate of viral clearance in lymphoid tissue would markedly affect the estimated life span, , of productively infected cells deduced from antiretroviral drug therapy (ART) experiments, if the clearance rate in lymphoid tissues were sufficiently slow [14]. Since we now estimate even higher clearance rates than we did previously, it becomes even more likely that the clearance from lymphoid tissue is sufficiently fast to not affect the accuracy of current estimates of , the death rate of productively infected cells. Having , one expects the loss of productively infected cells to be the dominant slope of viral decline during the first week or two of ART [14]. The amount of virus in the blood, free virus in lymphoid tissue, and virus on FDCs should be in quasi steady state with the loss of productively infected cells, which is in good agreement with the rate of about 0.5 per day at which virus in lymphoid tissue declines during ART [13]. When the amount of virus on FDCs has dropped significantly, most of the remaining virus will be attached by multiple bonds [8], which can account for the observed long half lives of virus on FDCs during ART [7].
Summarizing, we have provided new estimates of viral efflux and clearance rates in various organs, including blood and lymphoid tissue. We have confirmed that the exchange rate from lymphoid tissue to the blood should be slow [14]. Whenever viral production rates exceed virus particles over the life time of a productively infected cell [22]–[24], we estimate clearance rates in lymphoid tissue of 10–100 d for typical situations where most of virus in lymphoid tissue is associated with FDCs.
Reilly et al. [23] fitted data obtained by in situ hybridization in lymphoid tissue, including measurements of the number SIV RNA copies on the surface of and in the vicinity of activated and resting CD4 T cell in lymphoid tissue of acutely SIV-infected rhesus macaques, with averages of and copies per cell. Although the data were static, i.e., they were single snap-shots of different cells not containing any information on the time since infection of the cells, the data was fit to a dynamic model with a Bayesian approach using simulated annealing to find the most likely parameter values [23]. The model consisted of a three parameter viral production function (see Fig. 3, i.e., intercept, up-slope, and saturation time, and two parameters for the exponential decay of cells producing virus, and loss of viral particles, respectively [23]. The prior distribution for the half-life of activated infected cells was fixed to a mean of 1.5 days, whereas having little information on the expected life span of productively infected resting cells, two different prior distributions for the half-life of resting infected cells were chosen, with means of 4 days (Fig. 3a & b) and 14 days (Fig. 3c & d), respectively.
To estimate viral production rates per virus producing cell from this data, one has to integrate the virus production function over the life span of the cells. Reilly et al. [23] estimated the median production rate by integrating up to the estimated half-life of the cells (see the heavy lines in Fig. 3), and this yielded median production rates of 1479 and 1395 viral particles per activated virus producing cell for the two prior distributions (Fig. 3b & d, respectively), and of 644 and 3405 viral particles per resting virus producing cell (Fig. 3a & c, respectively).
Note that the estimates obtained in Fig. 3 are independent of the estimated time to saturation, and largely depend on the intercept and slope of the virus production function (Fig. 3). In three of the four panels the heavy black line, reflecting the period over which virus production was integrated, stops well before the saturation point. This is reassuring because the 95% credible intervals on the saturation time are very large. Indeed, these data can hardly support the existence of a saturation time in the production curve because the estimated saturation time is much larger than the estimated half-life of the cells. Thus, the data must have had virtually no cells that became old enough to breach the saturation time, and therefore the data hardly contains any information on possible saturation effects. This implies that the estimated saturation times were largely determined by the prior distribution of the Bayesian parameter estimation procedure. Fortunately, for the interest of this paper, eliminating the saturation barely affects the estimated production rates.
|
10.1371/journal.pntd.0002909 | Proteomic Analysis of Mecistocirrus digitatus and Haemonchus contortus Intestinal Protein Extracts and Subsequent Efficacy Testing in a Vaccine Trial | Gastrointestinal nematode infections, such as Haemonchus contortus and Mecistocirrus digitatus, are ranked in the top twenty diseases affecting small-holder farmers' livestock, yet research into M. digitatus, which infects cattle and buffalo in Asia is limited. Intestine-derived native protein vaccines are effective against Haemonchus, yet the protective efficacy of intestine-derived M. digitatus proteins has yet to be determined.
A simplified protein extraction protocol (A) is described and compared to an established method (B) for protein extraction from H. contortus. Proteomic analysis of the H. contortus and M. digitatus protein extracts identified putative vaccine antigens including aminopeptidases (H11), zinc metallopeptidases, glutamate dehydrogenase, and apical gut membrane polyproteins. A vaccine trial compared the ability of the M. digitatus extract and two different H. contortus extracts to protect sheep against H. contortus challenge. Both Haemonchus fractions (A and B) were highly effective, reducing cumulative Faecal Egg Counts (FEC) by 99.19% and 99.89% and total worm burdens by 87.28% and 93.64% respectively, compared to the unvaccinated controls. There was no effect on H. contortus worm burdens following vaccination with the M. digitatus extract and the 28.2% reduction in cumulative FEC was not statistically significant. However, FEC were consistently lower in the M. digitatus extract vaccinates compared to the un-vaccinated controls from 25 days post-infection.
Similar, antigenically cross-reactive proteins are found in H. contortus and M. digitatus; this is the first step towards developing a multivalent native vaccine against Haemonchus species and M. digitatus. The simplified protein extraction method could form the basis for a locally produced vaccine against H. contortus and, possibly M. digitatus, in regions where effective cold chains for vaccine distribution are limited. The application of such a vaccine in these regions would reduce the need for anthelmintic treatment and the resultant selection for anthelmintic resistant parasites.
| Parasitic worms infecting the intestines of grazing livestock cause economic losses and welfare problems. Infection is predominantly controlled by wormers, the indiscriminate use of which has led to drug-resistance problems in the worms infecting livestock on which many of the world's resource-poor farmers are dependent. New and cheap control methods are needed. Vaccination with protein extracts from the parasite Haemonchus contortus reduces the burden of infection and some work has indicated that cross-protection between closely related parasites is possible. Typically, these extracts are made using relatively sophisticated centrifugation and chromatography equipment as well as needing refrigeration capabilities. In this study, the authors show that equally efficacious extracts can be prepared using a very simplified protocol not requiring these specialist facilities. Proteomic analyses demonstrated the close similarity between protein extracts from both H. contortus and Mecistocirrus digitatus and vaccine trials in sheep showed that the simplified extraction protocol resulted in an equally efficacious vaccine compared to the more complex methods described prior to this work. Antigenic cross-reactivity was demonstrated between extracts from the two species; the M. digitatus extract gave a slight reduction in worm egg output when used to vaccinate sheep challenged with H. contortus.
| Infections with blood-feeding gastrointestinal nematodes, such as Haemonchus contortus and Mecistocirrus digitatus, cause significant animal welfare and production losses globally [1], [2]. The latter is an important blood-sucking nematode of cattle in Asia and Central America [3]. In Asia and Africa, where resource-poor small holder farming is more common, gastrointestinal nematode infections of livestock are ranked in the top twenty diseases of livestock affecting the farmers ability to maintain food security and contribute to economic growth [4]. Control of these parasites is currently achieved by the regular use of anthelmintics: However, this approach leads to the inevitable development of anthelmintic resistance [5]. In Tamil Nadu, India, a recent survey (Dicker et al, unpublished) found evidence of widespread inefficacy of albendazole, levamisole and ivermectin against H. contortus in sheep and goats. As such, novel control strategies, such as vaccines, are urgently needed to enable resource-poor small-holder farmers in Tamil Nadu to control parasite infections in their livestock to ensure their food security.
Substantial progress has been made in identifying several antigens from H. contortus which, in their native form, stimulate sufficiently high levels of protective immunity (70–95% reductions in faecal egg output) in the ovine host to indicate that vaccination is feasible [6]–[9]. Much previous work by other authors has focused on proteins or protein complexes expressed on the surface of the worm gut which are exposed to the blood meal and, hence, antibody ingested with it. The antigens generally, but not in all cases, show protease activity and the antibody is thought to mediate protective immunity by blocking the activity of enzymes involved in blood meal digestion within the parasite [10]. The recent increase in genomic data for nematodes such as Caenorhabditis elegans, H. contortus, hookworms and the trematode Fasciola hepatica has allowed identification of novel candidate vaccine antigens whilst proteomics analysis has aided in the identification of post-translational modifications which affect protein folding and protein immunogenicity [11]–[13] (http://nematode.net/NN3_frontpage.cgi). Compared to H. contortus, very little is known about M. digitatus, with only 25 nucleotide sequences, 12 genes and 38 protein entries present on NCBI (http://www.ncbi.nlm.nih.gov/, 31st May 2013) and no information on the potential for vaccination as an alternative to anthelmintics. Proteomics has been used to investigate other potential vaccine candidates such as excretory/secretory products from adult Ostertagia ostertagi and H. contortus and larval Teladorsagia circumcincta [14]–[17]. No proteomic comparison has been made between extracts from different blood-feeding nematodes but this approach should readily identify if potential vaccine candidates are shared by related species.
Given that anaemia and a reduction in weight gain caused by the haematophagous activity of adult stages seem to be the most important pathogenic effects of M. digitatus infection in calves and are similar to those observed during infection with Haemonchus placei in calves and H. contortus in sheep and goats [2], [18], we sought to compare native protein vaccine preparations, enriched for intestinal surface proteins by Concanavalin A lectin affinity binding [19] from H. contortus and M. digitatus using proteomics, and to evaluate the protective efficacy of the latter against H. contortus challenge in sheep as a prelude to vaccine trials in buffalo in India. Cross-protection has been previously shown to occur in trials conducted in sheep which had been immunized with native Ostertagia protein fractions but challenged with H. contortus; the Ostertagia antigens cross-protected efficiently against Haemonchus [20], as such cross-protection between M. digitatus and H. contortus was believed to be likely.
All experimental procedures were approved by the Moredun Research Institute Experiments and Ethics committee (Experiment number E31/12) and were conducted under the legislation of a UK Home Office License (60/3825) in accordance with the Animals (Scientific Procedures) Act of 1986.
Adult M. digitatus were collected post mortem from abomasa of cattle collected at an abattoir in Salem, India. Adult Haemonchus contortus were obtained from a donor lamb following standard methods as described in Smith & Smith [8]. All parasites were stored at −20°C in 1 X PBS until required.
Protein extraction from M. digitatus was carried out in India with the resulting protein extract transported to Moredun Research Institute (MRI) whilst maintaining a cold chain; H. contortus proteins were purified at MRI. Proteins were extracted using a protocol based on the method in [21]. The parasites were washed several times in 1 X Tris buffered saline (TBS) at a ratio of 10 ml per g dried worms and the worm pellet then homogenised on ice using a chilled pestle and mortar followed by a chilled glass hand homogeniser directly in a 1.0% v/v Triton X-100 buffer. The homogenate was then centrifuged at 2500 X G for 20 mins at 4°C. The supernatant was removed and filter sterilised through a 0.45 µM filter before being mixed with ConA lectin-agarose (Vector laboratories) on a rotary mixer at 4°C for 1 hour. The Protein-ConA-agarose complex was allowed to settle under gravity at 4°C and the supernatant removed. This washing procedure was repeated on 3 occasions using a 0.25% v/v Triton X-100 buffer and then bound proteins were eluted by using a buffer containing 200 mM α Methyl-D-mannopyranoside and 200 mM α Methyl-D-glucopyranoside. The resultant protein solutions were subsequently passed through a 0.22 µM filter. The H. contortus extract made using this method was named Hc extract A. A second H. contortus extract was made following the method in [20], [19], and is referred to as Hc extract B. Briefly, with centrifugation between each step, adult parasites were extracted in PBS to remove water soluble proteins, then the resultant pellet extracted in PBS/Tween 20 to solubilise membrane-associated proteins with the final pellet solubilised with PBS containing 1.0%v/v Triton X-100. The solution was pumped through a column containing ConA lectin-agarose (Vector laboratories). After thorough washing in a 0.25% v/v Triton X-100 buffer, the column bound proteins were eluted using a buffer containing 200 mM α Methyl-D-mannopyranoside and 200 mM α Methyl-D-glucopyranoside.
Protein concentration was estimated using a Pierce BCA Protein Assay Kit (Thermo Scientific), according to the manufacturer's instructions. An aliquot of each of the M. digitatus and H. contortus protein elutions were concentrated using the Amicon Ultracel centrifugal filters (Millipore) with a 10 KDa cut off, before a final estimation of protein concentration was obtained.
To determine the complete protein profile from each parasite, individual non-reduced Novex NuPAGE 4–12% Bis-Tris gels (Life Technologies) for M. digitatus and H. contortus in 1 X MOPS buffer (Invitrogen) were run at 200 V for 45 mins following the manufacturers' protocols. 2.19 µg and 3.29 µg M. digitatus protein and 2.11 µg and 3.17 µg Hc extract A were loaded in NuPAGE LDS sample buffer with the PageRuler Unstained Broad Range Protein Ladder (Fermentas) loaded alongside to allow estimation of protein band size. The gel was stained with SimplyBlue Safestain (Invitrogen) and de-stained with distilled water according to the manufacturer's instructions.
Liquid chromatography-electrospray ionisation-tandem mass spectrometry (LC-ESI-MS/MS) was carried out on the proteins contained in one complete lane from each species at the MRI Proteomics facility using the method as described previously in Wheelhouse et al [22] to provide an estimate of the relative abundance of each protein. Mascot generic files were generated and submitted to a local database server, utilising ProteinScape version 2.1 (Bruker Daltonics), to perform database searches against the NCBI non-redundant eukaryotic database (http://www.ncbi.nlm.nih.gov/) and the NEMBASE4 nucleotide database (http://www.nematodes.org/nembase4/index.shtml), using the MASCOT (Matrix science) search algorithm. The carbamidomethyl (C) modification was fixed whilst the Deamidated (NQ) and Oxidation (M) modifications were variable, peptide and fragmentation mass tolerance values were set at 0.5 Da, allowing for a single 13C isotope.
Peptide matches were compiled into a protein list compilation (PLC) search result and the quality of proteins inspected manually. Proteins with three or more peptides, or two peptides and with a Molecular Weight Search (MOWSE) score greater than or equal to 90, were deemed significant if at least two different peptides were observed to contain an unbroken run of 4 ‘b’ or ‘y’ ions. The NCBI or NEMBASE protein hit identity for all selected proteins (those passing the quality checks) was determined and the number of identical proteins in each of the databases for both H. contortus and M. digitatus determined. Proteins identified as mammalian, trypsin or keratin were removed from the analysis.
Subsequently, to determine the identity of individual bands of interest, visible bands were excised from a second gel which had been loaded with 3.29 µg M. digitatus extract and 3.17 µg Hc extract A in NuPAGE LDS sample buffer and run as described above to provide identification of individual bands. These individual bands were subjected to LC-ESI-MS/MS and MASCOT searches against both the NCBI non-redundant eukaryotic database and the NEMBASE4 nucleotide database carried out, as described previously. The quality of the protein matches was manually inspected as described for the PLC results.
A vaccine trial comparing the efficacy of the Hc extract A and M. digitatus protein extract against Hc extract B and an unvaccinated control group was undertaken, following standard methods as described in Smith and Smith [8]. Briefly, groups (n = 7) of indoor housed, parasite free lambs, matched for sex and weight, were vaccinated sub-cutaneously three times, three weeks apart with a dose of 40 µg/mL of protein extract (either Hc extract A, Hc extract B or M. digitatus extract) in TBS with VAX Saponin adjuvant (Guinness Chemical Products Ltd) at a final concentration of 1 mg/mL. The unvaccinated control group received VAX Saponin adjuvant in TBS only. On the third vaccination day all lambs were challenged with 5000 H. contortus L3s suspended in water per os. From fourteen days post challenge, twice weekly faecal egg counts (FECs) using a modified technique as described in Jackson [23], were carried out on faecal samples obtained per rectum. Individual cumulative FEC were estimated by utilising the area under the curve calculation with the linear trapezoidal rule. The mean cumulative FEC for each group was subsequently calculated. Sheep were euthanized on day 35 post challenge, when it was anticipated that all worms present had reached patency, and worms recovered following methods described in Patterson et al [24]. Mean total, male and female worm burdens were calculated for each group.
Statistical analysis of the FEC and worm burden results was carried out following the guidelines set out in Coles et al [25], data was analysed using Minitab (version 15). The non-parametric Kruskal-Wallis test, followed by Pairwise Mann-Whitney tests, with adjusted P values for multiple comparisons (Bonferroni correction), was used to determine whether statistically significant differences in worm burdens and cumulative FEC were present between the vaccinated groups compared to the unvaccinated control group. S.E.M., range and percentage efficacy (P.E.) for the group mean worm burdens and group mean cumulative FEC were calculated; the P.E. for each vaccinated group was calculated relative to the unvaccinated controls.
2.85 µg M. digitatus extract and 2.83 µg Hc extract A in NuPAGE LDS sample buffer were heated at 70°C for 10 mins, then loaded onto a 4–12% Bis-Tris gel (Life Technologies) in 1 X MES buffer (Invitrogen) and run at 200 V for 50 mins following the manufacturers' protocols. 8 µL PageRuler Prestained Protein Ladder (Fermentas) was run alongside the samples, before being removed with a scalpel. The gel was transferred onto a nitrocellulose membrane (Invitrogen) for 1 hour using the XCell blot module, washed twice in a 50 mM Tris, 2.5M NaCl, 0.25% Tween20 pH 7.4 buffer (TNT) before being blocked overnight in TNT. 200 µL sera from each Md extract vaccinated lamb taken 7 days after the third vaccination was pooled together then diluted 1 in 200 in TNT. The blot was incubated in the diluted sera for 1 hour then washed for 10 mins three times in TNT. Monoclonal mouse anti goat/sheep IgG-HRP1 (Sigma-Aldrich) was diluted 1 in 1000 and the blot incubated in it for 1 hour, followed by three 10 mins washes in TNT. Finally the blot was visualised by incubation in DAB reagent (Sigma-Aldrich) until bands became visible.
The relative abundances of different proteins are shown in Figure 1, whilst the identities of individual protein detected in H. contortus and M. digitatus (Figure 2) are shown in Tables 1 and 2 respectively. The most frequently identified hits in the H. contortus NCBInr database (representing 22.7% of the results each) were for aminopeptidase, such as H11 and apical gut membrane polyproteins including the P100GA and P46GA2 proteins (Figure 1). Aminopeptidases were also the most prevalent hit in the M. digitatus NCBInr database (37.5%), and the third most prevalent search result (12.9%) in the H. contortus NEMBASE nucleotide database. Protein disulphide isomerases and aminopeptidases was the most frequently identified hit (25% each) in the M. digitatus NEMBASE nucleotide database. The most frequently identified hit (32.3%) in the H. contortus NEMBASE nucleotide database was protein disulphide isomerase, which also accounted for 13.6% of the protein identities from the H. contortus NCBInr database search. Zinc metallopeptidases were identified more often from the H. contortus database searches, representing 13.64% and 16.1% of the significant hits from the NCBInr and NEMBASE nucleotide searches, respectively, compared to the M. digitatus database searches (0% and 6.3%, respectively). In both the H. contortus and M. digitatus database search results, homologues of other potential vaccine candidates were identified frequently (Figure 1) and included glutamate dehydrogenase [26] and the P100GA proteins [27]. Potential vaccines candidates only identified from H. contortus include a 24 kDa excretory/secretory protein [28], aspartyl protease precursor [9] and P46GA2 [27]. Only one potential vaccine component was solely identified from the M. digitatus whole lane analysis; a galectin protein 5 identified from the NEMBASE4 nucleotide database search. Between 16 and 41 peptides identified as cysteine proteases (including cathepsins) were present in database searches from both M. digitatus and H. contortus; however none of the proteins passed the quality checks.
The proteomic analyses of the individual bands from H. contortus and M. digitatus protein extracts, excised from a 4–12% Bis Tris gel (Figure 2A) emphasised the similarity between the two parasites, an observation enhanced by the demonstrable antigenic cross-reactivity of the two extracts (Figure 2B). The putative identities for these protein bands are shown in Tables 1 and 2; further details of the proteomic results for these bands can be found in Tables S1 and S2, respectively. Hc1 and 2 and Md4 are all putative zinc metallopeptidases whilst Hc4, Md3, 5 and 6 are all microsomal aminopeptidases or H11: It is possible that Md3, at 220 kDa, is a dimer of either Md5 or 6. Hc7, at approx 47 kDa, and Md9, at approx 45 kDa share a putative protein identity of P100GA whilst Hc8 and Md10 both migrated at approximately 40 kDa and had putative identities of aspartyl proteases. However, other proteins also gave significant matches to these bands indicating that each band may comprise more than one protein, so the identity of these bands could not be confirmed. Hc9 was identified as a Galectin 5, whilst no M. digitatus bands were identified as galectins. Combined with the previous, whole lane, analysis this indicates Galectin 5 may be present in both species. Two H. contortus bands (Hc3 and 11) and five M. digitatus bands (Md1, 2, 11, 12 and 13) returned results either as no significant hits or hypothetical or uncharacterized proteins.
Figure 3 shows the average FECs for each group obtained from twice weekly per rectum faecal samples from day 14 to day 34 post challenge, with the error bars representing the standard error of the mean (S.E.M.) for each group. Both the H. contortus vaccine preparations (Hc extracts A and B) elicited similar levels of significant protection against H. contortus challenge. The percentage efficacy of the Hc extracts A and B, as determined by group average cumulative FEC, was 99.19%, and 99.89% respectively (Table 3) with the reductions in total worm burdens being 87.28% and 93.64% respectively (Table 4). Both the H. contortus vaccine preparations appeared to be more effective against females than males, reducing the worm burden by 94.23% and 79.54% (Hc extract A females and males) and 98.46% and 88.48% (Hc extract B females and males) compared to the unvaccinated controls (Table 4). All cumulative FEC and worm burden reductions for Hc extract A and B vaccinated groups were statistically significant (P<0.0167) compared to the unvaccinated controls, with the P value adjusted for multiple comparisons using Bonferroni correction.
Figure 3 shows that, although there was an indication that vaccination with M. digitatus derived proteins against H. contortus challenge reduced the group mean FEC slightly; the cumulative FEC P.E. of 28.20% (Table 3) was not statistically significant. The Md extract vaccinated group average FEC was always lower than the controls; between day 67 and 76 post infection, the FECR was between 24.3% and 39.2% compared to the unvaccinated controls (Figure 3). Average group male and total worm burdens were higher (i.e. negative P.E.) in the Md extract vaccinated group compared to the unvaccinated controls whilst the 14.62% reduction in group mean female worm burden was not statistically significant (Table 4).
The main aim of this work was to compare, using proteomic analyses, the major gut membrane proteins from the closely related haematophagous nematodes, H. contortus (affecting sheep and goats) and M. digitatus (affecting cattle and buffalo), both of which impose significant constraints on livestock production in tropical and subtropical regions of the world. Gut membrane proteins have proven vaccine efficacy in H. contortus [6], and a recent study showed that vaccination of calves with native parasite gut membrane glycoproteins obtained from H. contortus conferred protection against both H. placei and H. contortus [29]. This work, and a previous study [20], indicate that good cross-nematode species protection could be stimulated by vaccination with gut membrane proteins derived from closely related species. Due to their similar haematophagous life cycle, it is possible that cross protection against M. digitatus could be achieved utilising a vaccine developed against H. contortus [2], [29]. If vaccination is going to be a viable method of control for both H. contortus and M. digitatus for resource-poor small-holder farmers then the vaccine production method should be cheap without a requirement for specialist and expensive laboratory equipment. This enables local production with minimal dependency on an extensive cold chain. In this paper we describe such a method. Then, the resulting protein extracts from both M. digitatus and H. contortus were analysed by proteomics to determine whether similar, known candidate vaccine antigens were present in each extract. Finally the protein extracts were tested in a vaccine trial in sheep against H. contortus challenge.
Several reports have described the purification of protective antigens from the intestine of Haemonchus, and they share the need for successive saline and detergent extractions, high speed centrifugation steps and specialist chromatography equipment [19], [30]–[33]. Redmond et al [21] used a simple detergent extract followed by affinity chromatography over Con A lectin to isolate a variety of glycoproteins from C. elegans. ConA lectin has a strong affinity for the microvillar surface of the Haemonchus intestine [6], [8] and was used in [21] to isolate similar antigens from C. elegans, antibody to which cross reacted with numerous H. contortus gut antigens. Here, we used the same technique to prepare antigen extracts from H. contortus and M. digitatus with the modification that the Con A lectin chromatography step was replaced by a simpler protocol of mixing the extract directly with the affinity medium, with an incubation and wash protocol, as described in the materials and methods. The proteomic analyses described underline that this simple method is effective for the isolation of an extract which is highly enriched for intestinal proteins from nematodes and that its composition did not differ obviously from extract B, by comparison to an analysis described by Sherlock [33].
The protein components of each vaccine were determined by both LC-ESI-MS/MS of whole gel lanes (Figure 1) and on individual bands which were excised from the gels (Figure 2A) and then analysed using LC-ESI-MS/MS (Table 1, Haemonchus and 2, Mecistocirrus). Analysis of the whole gel slices was performed to ensure all proteins present in the extracts would be identified, not just those in sufficient abundance to create a visible band on the gel, whilst the identity of individual bands was determined by excising these from a second gel. There was a broad agreement between these datasets but there tended to be more unidentified or hypothetical protein bands in the analyses from M. digitatus compared to H. contortus, probably as a result of a lack of specific sequence data for the former. The precise identification of M. digitatus proteins is likely to be hampered by low genomic coverage of M. digitatus and resultant peptide matches with lower percentage sequence coverage and MOWSE scores [34]. Nonetheless, the results of this analysis indicate that similar proteins were extracted from adult H. contortus and M. digitatus and, as such, further validation work on the individual proteins identified from M. digitatus would be a worthwhile step towards developing a multivalent native vaccine for use in areas where co-infection of livestock with Haemonchus species and M. digitatus occurs.
Despite some differences in the migration pattern of the protein bands from M. digitatus and H. contortus, as shown in Figure 2, the analyses here indicate that both extracts are quite similar in terms of protein functions identified. For example, aminopeptidases including H11, zinc metallopeptidases, and protein disulphide isomerases were prominent in both extracts. Many of these proteins have been associated with varying levels of protection against H. contortus and other nematodes in vaccine trials [6], [9]. H11 is an insoluble gut membrane glycoprotein of approx 110 kDa involved in blood meal digestion in H. contortus and is the most effective vaccine candidate in H. contortus, giving greater than 90% reduction in worm burden [30]. The presence of aminopeptidases, including H11, in M. digitatus indicates that, as a blood feeder like H. contortus, it may also be amenable to the gut antigen vaccination approach. In addition, both extracts contained homologues of zinc metalloproteases, which had been shown previously to be a major component of a host-protective protein complex H-gal-GP [35]. Zinc metallopeptidases were prevalent in the H. contortus database searches, representing 13.64% and 16.1% of the significant hits from the NCBInr and NEMBASE nucleotide searches, respectively and were barely detectable in the M. digitatus equivalents (0% and 6.3%, respectively). Somewhat surprisingly, there were very few hits to cysteine proteases (none of which passed the proteomic quality checks) despite their apparent abundance in EST datasets from the intestine of H. contortus [36], [37]. This anomaly may reflect differences in transcript abundance compared to translation into an actual protein. Gut derived cysteine proteases have been shown to be useful immunogens in H. contortus and in human hookworms, which are also blood-feeders [38], [39].
Protein disulphide isomerases were particularly prominent in both datasets; they play important roles in protein folding, catalysing thiol-disulphide interchange which leads to protein disulphide bond formation. In C. elegans, they have a role in the formation of cuticular collagen network [14]. Although protein disulphide isomerases have been found in several nematodes and have been detected in ES [14], [15], [40], no report has linked them to the induction of protective immune responses as yet.
Other putative vaccine candidates which have been studied in less detail in parasitic nematodes and which have been identified in this current study include: Glutamate dehydrogenase, the apical gut membrane polyproteins (P100GA and P46GA2), and a 24 kDa excretory/secretory (E/S) protein. Glutamate dehydrogenase has been identified from H. contortus and T circumcincta using Thiol-Sepharose affinity chromatography but, despite being the major 60 kDa component of the TSBP extract [7], it did not provide protection against infection [31]. Three proteins, P46GA1, P52GA1 and P100GA1 are encoded by the same gene in H. contortus, initially being expressed as a polyprotein [27] and together giving 60% and 50% reductions respectively in worm burden and faecal egg outputs in goats [41]. In this analysis, P100GA2 was identified from both H. contortus and M. digitatus whilst P46GA2 was only identified from H. contortus. Finally, a 24 kDa excretory/secretory protein was identified only from the H. contortus protein extract. In H. contortus, this 24 kDa E/S protein, together with a 15 kDa E/S protein reduced FECs by 32–77% and adult worm burden by 64–85% when tested as a vaccine in sheep [28].
Subsequent to the proteomic analysis, a vaccine trial, comparing the efficacy of the Md extract, Hc extracts A and B against H. contortus challenge, was undertaken. Both H. contortus vaccine preparations gave statistically significant levels of protection against homologous H. contortus challenge, compared to the unvaccinated controls as measured by cumulative FEC and worm burden. The levels of protection (reductions in FEC of 99.19% and 99.89% with Hc extract A and B respectively) exceeding figures of 80% efficacy in 80% of the flock predicted by Barnes et al [42] as necessary to provide better protection against infection and disease than standard anthelmintic based control strategies. However, this prediction was made using a computer model based on Trichostrongylus colubriformis, a less fecund nematode. These data indicate that the simplified method used to produce Hc extract A could form the basis for a locally produced vaccine against H. contortus in regions, such as India, where effective cold chains for vaccine distribution are limited, with the proviso that sufficient worm biomass can be harvested, either from donor animals or abattoir material. All the vaccine preparations tested here were more effective against female worms than their male counterparts, probably reflecting the extra nutritional demands imposed by egg production [43].
M. digitatus is not native to the U.K. so performing a protection trial using a crude M. digitatus protein extract against homologous challenge was not possible. Therefore, a heterologous challenge with H. contortus, which is part of the same Trichostrongylidae family and is also haematophagous, was chosen to determine whether protection could be achieved with M. digitatus protein extracts [18]. Previously, cross protection trials have been carried out using protein extracts from Ostertagia ostertagi and Teladorsagia circumcincta against H. contortus challenge and using H. contortus proteins against T. circumcincta, Trichostrongylus axei and Cooperia oncophora challenge [20], [44]. O. ostertagi proteins cross protected against H. contortus challenge by reducing FEC by 81–97% and worm burdens by 57–84% [20]. In comparison the results of cross-protection trials against T. circumcincta, Tr. axei and C. oncophora were mixed as H. contortus proteins did not provide any protection against infection with any of the aforementioned species, yet T. circumcincta proteins caused a significant reduction in FEC (though no effect on worm burdens) following challenge with H. contortus [44]. In this current trial, there was no effect on H. contortus worm burdens following vaccination with a M. digitatus vaccine extract and the 28.2% reduction in cumulative FEC was not statistically significant. However, it is notable that the FEC were consistently lower in the M. digitatus extract vaccinates compared to the challenge controls from 25 days post-infection. A trial is now in progress in India to determine whether M. digitatus proteins provide protection against M. digitatus challenge. The failure to obtain evidence of effective cross-protection was somewhat surprising given the evidence from prior studies discussed above and that the parasites are closely related and obligate blood feeders. Moreover, the immunoblot data shown in Figure 2 confirms that there is strong antigenic cross reactivity between the H. contortus and M. digitatus extracts. Perhaps the explanation lies in the relative abundance or absence of specific components when the two vaccine extracts are compared. For example, there is solid evidence that the zinc metallopeptidases contribute to vaccine-induced immunity against Haemonchus but these were much less abundant in M. digitatus (Figure 1) and the 24 kDa Haemonchus ES protein, associated with strong protective immune responses [28] was not detected in the M. digitatus extract.
|
10.1371/journal.pgen.1000486 | HIF-1 Modulates Dietary Restriction-Mediated Lifespan Extension via IRE-1 in Caenorhabditis elegans | Dietary restriction (DR) extends lifespan in various species and also slows the onset of age-related diseases. Previous studies from flies and yeast have demonstrated that the target of rapamycin (TOR) pathway is essential for longevity phenotypes resulting from DR. TOR is a conserved protein kinase that regulates growth and metabolism in response to nutrients and growth factors. While some of the downstream targets of TOR have been implicated in regulating lifespan, it is still unclear whether additional targets of this pathway also modulate lifespan. It has been shown that the hypoxia inducible factor-1 (HIF-1) is one of the targets of the TOR pathway in mammalian cells. HIF-1 is a transcription factor complex that plays key roles in oxygen homeostasis, tumor formation, glucose metabolism, cell survival, and inflammatory response. Here, we describe a novel role for HIF-1 in modulating lifespan extension by DR in Caenorhabditis elegans. We find that HIF-1 deficiency results in extended lifespan, which overlaps with that by inhibition of the RSKS-1/S6 kinase, a key component of the TOR pathway. Using a modified DR method based on variation of bacterial food concentrations on solid agar plates, we find that HIF-1 modulates longevity in a nutrient-dependent manner. The hif-1 loss-of-function mutant extends lifespan under rich nutrient conditions but fails to show lifespan extension under DR. Conversely, a mutation in egl-9, which increases HIF-1 activity, diminishes the lifespan extension under DR. This deficiency is rescued by tissue-specific expression of egl-9 in specific neurons and muscles. Increased lifespan by hif-1 or DR is dependent on the endoplasmic reticulum (ER) stress regulator inositol-requiring protein-1 (IRE-1) and is associated with lower levels of ER stress. Therefore, our results demonstrate a tissue-specific role for HIF-1 in the lifespan extension by DR involving the IRE-1 ER stress pathway.
| Dietary restriction (DR) is one of the most robust environmental manipulations that extend lifespan in various species. DR has also been shown to slow the onset of a number of age-related diseases. Studies in model organisms like C. elegans can be used to uncover biological mechanisms that determine the beneficial effects of DR. Previous studies suggest that the nutrient-sensing target of rapamycin (TOR) pathway is required for DR-mediated lifespan extension. However, the downstream mechanisms by which TOR modulates lifespan remain unclear. In mammalian cells, TOR and the downstream S6 kinase (S6K) activate expression of the hypoxia-inducible factor-1 (HIF-1), which is frequently up-regulated in various tumors. Using C. elegans as a model system, we characterized novel functions of HIF-1 in aging. We find that inhibition of HIF-1 extends lifespan under rich nutrient conditions, whereas enhanced levels of HIF-1 only allow partial lifespan extension by DR. We also demonstrated that increased lifespan by hif-1 or DR depends on the endoplasmic reticulum (ER) stress regulator inositol-requiring protein-1 (IRE-1) and is associated with lower levels of ER stress, which is caused by overloading of misfolded/unfolded proteins to ER. Thus, our results support the idea that HIF-1–mediated changes in protein homeostasis play a key role in the lifespan extension by DR.
| Dietary restriction (DR) has been shown to extend lifespan in various species. It also slows the onset of a number of age-related diseases in rodents. Conservation of signaling pathways in multiple species and the rapidity with which lifespan studies can be carried out in simple model organisms make them powerful tools to understand aging and age-related diseases in humans. Identification of genes involved in DR response will therefore provide potential targets for treatments of age–associated diseases and the extension of healthful lifespan in humans. The quest to understand the mechanisms of DR-induced lifespan extension has led to intensive studies in the primary genetic model organisms Saccharomyces cerevisiae, Caenorhabditis elegans, and Drosophila melanogaster, in which the robust effects of lifespan extension by DR can be observed [1]–[5]. Evidence from previous studies has identified the TOR pathway as a key mediator of nutrient-modulated lifespan changes in flies [6], yeast [7] and worms [8], although the involvement of TOR in response to DR in C. elegans is still controversial [8],[9].
TOR is a conserved protein kinase that plays essential roles in regulating growth and metabolism in response to nutrients and growth factors [10]. TOR interacts with the regulatory associated protein of TOR (raptor) to permit transduction of nutrient signals to downstream cellular processes, including mRNA translation, ribosome synthesis, expression of metabolism-related genes and autophagy [11]. TOR promotes protein synthesis by activating S6K and inhibiting eukaryotic translation initiation factor 4E-binding protein (4E-BP). Recent studies have shown that regulation of mRNA translation plays critical roles in lifespan determination in multiple species [9], [12]–[19]. In C. elegans, mutations in the TOR ortholog let-363 lead to developmental arrest at the third larval stage and intestinal atrophy [20]. Inhibition of let-363 by RNAi extends lifespan [8],[12],[21]. The larval arrest phenotype was also observed from homozygous mutants of daf-15, which encodes the C. elegans ortholog of raptor. Heterozygous mutants of daf-15 have increased lifespan [22]. Inactivation of rsks-1, which encodes the C. elegans ortholog of S6K [23], significantly extends lifespan [12],[13]. Although it is well established that the C. elegans TOR pathway modulates lifespan, it is still unclear whether TOR affects DR-mediated lifespan extension. There have been two reports with opposite results on whether inhibition of let-363 by RNAi further extends lifespan of the eat-2 mutant, which serves as a genetic mimic of DR [8],[9].
Multiple studies in mammalian and Drosophila cells have implicated the transcription factor HIF-1 as a target of the TOR pathway [24]–[27]. HIF-1 is regulated at the mRNA translation level, and enhanced levels of HIF-1 are associated with increased TOR and S6K activities under both normoxic and hypoxic conditions [24],[26],[28],[29]. HIF-1 is a heterodimeric transcriptional complex that contains HIF-1α and HIF-1β. It plays essential roles in oxygen homeostasis [30] and is also regulated by other physiological stimuli like heat acclimation [31], acidosis [32], nitric oxide [33], inflammation [34] and oxidative stress [35],[36]. Under normoxia, specific proline residues of HIF-1α are hydroxylated by the PH superfamily of dioxygenase encoded by egl-9 in C. elegans. Hydroxylated HIF-1α is subject to von Hippel Lindau (pVHL)-mediated proteasome degradation. Under hypoxia, the hydroxylation modification declines and HIF-1α is stabilized for its transcriptional activities [37]. HIF-1 helps cells adapt to low-oxygen stress by regulating angiogenesis, glycolysis, and cell survival [30]. HIF-1 overexpression is frequently detected in solid tumors due to intratumoral hypoxia and genetic mutations, and inhibition of HIF-1 can prevent tumor growth [38],[39]. In C. elegans, hif-1 encodes the HIF-1α ortholog, which has been connected to multiple biological processes, including hypoxia response [37],[40], adaption to heat stress [31], oxygen preference [41],[42], and neuronal migration [43].
Despite the well-characterized links between HIF-1 and cancer, metabolism, and cell survival, it is not known whether HIF-1 is directly involved in organismal aging. In this study, we investigated the role of HIF-1 in lifespan determination in C. elegans. We demonstrate that HIF-1 functions downstream of S6K to modulate DR-dependent lifespan extension in specific neurons and muscles via the IRE-1 ER stress pathway.
To characterize the role of HIF-1 in organismal aging, we examined lifespan phenotypes of hif-1 and egl-9 mutants. The egl-9 mutant has significantly increased HIF-1 protein levels and transcriptional activities [37],[44], thus serving as a HIF-1 gain-of-function mutant. A deletion mutant of hif-1 extended lifespan by 24%, whereas the egl-9 deletion mutant did not affect lifespan significantly under standard lab culture conditions (Figure 1A; Table 1 and S1). Inhibition of hif-1 by RNAi also extended adult lifespan (Figure 1B; Table 1 and S1). Thus, HIF-1 is a novel lifespan determinant in C. elegans.
In order to characterize the mechanism of lifespan extension by hif-1 deficiency, we performed genetic epistasis experiments to test interactions between hif-1 and other known longevity pathways. We tested mutations from the insulin/insulin-like growth facor-1 (IIS) pathway, which is a conserved pathway that modulates lifespan in nematodes, flies and mammals [3]. Mutations in the DAF-2/IGF-1 receptor double C. elegans lifespan [45],[46], and this lifespan extension is suppressed by mutations in the downstream DAF-16/FOXO transcription factor [47],[48]. Lifespan extension by hif-1 did not require DAF-16 (Figure 1C; Table 1 and S1). Inhibition of hif-1 by RNAi extended lifespan of N2 and a daf-16 null mutant to similar levels (18% lifespan extension, p<0.0001). Consistently, hif-1 RNAi further extended the lifespan of a daf-2 mutant (Figure 1D; Table 1 and S1). These results suggest that HIF-1 might modulate lifespan through mechanisms that are distinct from those used by the IIS pathway.
Next we tested the interaction between HIF-1 and the TOR-S6K pathway since previous studies from mammalian cells have shown that S6K promotes hif-1 expression [24],[26],[28],[29]. We found that inhibition of hif-1 by RNAi did not further extend lifespan of a daf-15 heterozygous mutant (Figure 1E; Table 1 and S1). The lifespan extension by a deletion mutant of rsks-1 was fully suppressed by egl-9, and hif-1 did not further extend rsks-1 lifespan (Figure 1F; Table 1 and S1). Taken together, these experiments suggest that lifespan extension by inhibition of rsks-1 and hif-1 takes place through overlapping mechanisms. We hypothesize that HIF-1 acts downstream of the nutrient-responsive TOR-S6K pathway to determine lifespan in C. elegans.
Next we investigated whether HIF-1 is involved in nutrient-mediated lifespan extension using a DR paradigm modified from the previously described solid DR (sDR) method [49]. Lifespan of animals fed with bacterial food at different concentrations (1.0×108–1.0×1012 cfu/ml) was examined on solid agar plates. The differences between our DR paradigm and the sDR method include that peptone was excluded and antibiotics were added to prevent bacterial growth, 5-fluorodeoxyuridine (FUdR) was used to prevent progeny from hatching, and differential food treatment was started from Day 1 instead of Day 4 post-reproductive adulthood. We have termed this method as msDR for modified solid DR. As shown in Figure 2A, animals treated with E. coli food at 1.0×1011 cfu/ml had a lifespan similar to those under standard culture conditions, whereas animals fed with E. coli at 1.0×109 cfu/ml showed the most significant lifespan extension (47% lifespan extension compared to 1.0×1011 cfu/ml, p<0.0001). We also observed that reductions in bacterial concentration led to increased heat stress resistance and decreased fecundity (Figure 2B and 2C), consistent with the phenotypes observed from other model organisms under DR [50],[51]. Based on the tradeoffs between fecundity and optimal lifespan extension and stress resistance, bacterial concentrations at 1.0×1011 cfu/ml and 1.0×109 cfu/ml were considered as ad libitum (AL) and dietary restriction (DR), respectively. Consistent with published results using other DR protocols such as eat-2 [52], liquid DR [53],[54] and food deprivation [55], lifespan extension by msDR does not require DAF-16 (Figure S1). Previous studies indicated that DAF-16 is required for lifespan extension by sDR [49] and is partially required for that by intermittent fasting (IF), another method of DR [56]. The reason for the different involvement of DAF-16 between sDR and msDR is unknown, but it could be due to the timing of DR induction or the differences in bacterial growth between the two methods.
We then examined longevity phenotypes of hif-1 and egl-9 mutants under different nutrient conditions. The hif-1 mutation extended lifespan under AL but did not cause further lifespan extension under DR, whereas lifespan extension under DR was diminished by a mutation in egl-9 (Figure 3; Table 2 and S2). We used both a random effects linear model [57] and a generalized estimating equation approach [58] to test whether the curve slopes for each mutant (hif-1 and egl-9) are identical to that of N2 in Figure 3F. These methods showed that both hif-1 and egl-9 mutants have significantly reduced slopes in mean lifespan versus food concentrations relative to N2 (p<0.05). Thus, our data suggest that under high nutrient status, decreased HIF-1 levels may cause a shift to the DR state, and overexpression of HIF-1 partially inhibits the lifespan extension effect of DR. This phenotype is unlikely to be caused by behavior defects of the hif-1 mutant since hif-1 animals have a normal brood size and pumping rate (Figure S2).
Recent studies have shown that various protocols of DR extend lifespan through different mechanisms in C. elegans [49], [53]–[55],[59]. To further test whether HIF-1 is involved in DR-mediated lifespan extension, we examined the effect of egl-9 on another form of DR using the eat-2 mutant. eat-2 encodes a subunit of nicotinic acetylcholine receptor, and mutations in eat-2 cause significantly reduced pharyngeal pumping and extended lifespan. eat-2 mutants have been widely used as a genetic mimic of DR in C. elegans [52]. We made the eat-2; egl-9 double mutant and examined the adult lifespan. The egl-9 mutation significantly suppressed the lifespan extension by a strong loss-of-function allele of eat-2 (Figure 4; Table 2 and S2). Our results suggest that EGL-9 is an important regulator of longevity due to a genetic mimic of DR by the eat-2 mutant.
It has been shown that the FOXA transcription factor PHA-4 is required for lifespan extension by liquid DR [54] and by a mutation in rsks-1 [60]. To characterize the genetic interaction between hif-1 and pha-4, we tested lifespan of N2 and hif-1 animals treated with either control RNAi or pha-4 RNAi. We found that pha-4 RNAi slightly reduced lifespan in both N2 and hif-1 backgrounds, but hif-1 extended lifespan of animals treated with control or pha-4 RNAi to a similar level (Figure S3). Thus, unlike rsks-1, lifespan extension by hif-1 does not require PHA-4, suggesting HIF-1 is not the only downstream effector of RSKS-1.
A fundamental question about DR is how animals sense reduced nutrients in the environment and adjust physiology of the whole organism for extended survival. Previous studies suggest that the intercellular communication between the nutrient sensing and the major metabolic tissues coordinate physiological changes upon DR [53]. To define the sites where HIF-1 acts to regulate DR-dependent lifespan extension, we utilized previously published transgenic animals that express egl-9 cDNA in various tissues in the egl-9 mutant background [41].
Expression of the egl-9 cDNA from its endogenous promoter, which drives egl-9 expression in virtually all cells, completely rescued the shortened lifespan of egl-9 under DR (Figure 5A, Table 3 and S3). Simultaneous expression of egl-9 in pan-neuronal and uv1 uterine-vulval cells also rescued the mutant phenotype (Figure 5B, Table 3 and S3). Pan-neuronal expression alone was sufficient for the rescue (Figure 5C, Table 3 and S3), whereas uv1 cell expression showed very little rescuing effect (Figure 5D, Table 3). Further analyses indicated that egl-9 expression in the serotonergic subset (ADF, NSM) rather than the soluble guanylate cyclase (sGC) subset (URX, AQR, PQR) of neurons that regulate the oxygen preference phenotype [61] was required for the lifespan extension by DR (Figure 5E and 5F; Table 3 and S3).
The egl-9 cDNA expression driven by the myo-3 promoter in body wall and vulval muscles also rescued the mutant phenotype (Figure 5G; Table 3 and S3). Vulval muscle expression alone did not rescue (Figure 5H; Table 3 and S3), suggesting that egl-9 expression in body wall muscle, in addition to serotonergic neurons, is important for HIF-1-mediated lifespan extension via DR. The muscle tissue plays an important role in C. elegans aging. Previous studies have shown an age-related decline in the muscle structure and function in C. elegans, resembling human sarcopenia [62]. Pharyngeal muscle expression of egl-9 by the myo-2 promoter partially rescued the mutant (Figure 5I; Table 3 and S3). The pharynx is the food intake organ, and myo-2 is a direct target of the PHA-4/FOXA transcription factor [63], which has been shown to be important for DR-dependent lifespan extension in C. elegans [54]. Expression of egl-9 in other tissues, including the hypodermis, XXX endocrine cells, pharyngeal gland and pharyngeal marginal cells, did not rescue the lifespan phenotype under DR (data not shown). We also tested lifespan phenotypes of these transgenic animals under AL conditions. Tissue-specific rescue of egl-9 did not significantly affect lifespan under AL (Figure S4; Table S4).
A recent study showed that HIF-1 acts both in neurons (ADF, NSM, URX, AQR, and PQR) and in gonadal endocrine cells (uv1) to regulate oxygen preference in C. elegans [41]. Our results indicate that HIF-1 acts in multiple cell types to modulate the longevity phenotypes of DR, and the lifespan and oxygen sensing effects of HIF-1 are determined by partially overlapped cell types.
ER stress is caused by a mismatch between the load of unfolded/misfolded proteins to ER and the capacity of the cellular machinery to cope with this load [64]. ER stress activates the unfolded protein response (UPR) through three signaling pathways transduced by IRE1, PERK, and ATF6. We speculated that DR and the hif-1 mutant might extend lifespan through ER stress pathways since high nutrients have been shown to increase ER stress in rodents and activation of TOR also leads to increased ER stress [65]. Furthermore, the lifespan extension by mutations in the yeast TOR pathway is mediated through GCN4, which regulates the UPR during ER stress [17],[18].
We first examined whether the ER stress pathway is also involved in the lifespan extension by DR in C. elegans. ire-1 encodes an ER transmembrane protein that senses misfolded proteins in the ER lumen and then activates the downstream transcription factor XBP-1 to regulate target gene expression and reduce ER stress [64]. We measured lifespan of a deletion mutant of ire-1 with different concentrations of bacterial food. The ire-1 mutant showed significantly reduced lifespan extension by DR (Figure 6A; Table 2 and S2). Using both a random effects linear model [57] and a generalized estimating equation approach [58], we found that the ire-1 mutant has significantly reduced declines in mean lifespan versus food concentrations relative to N2 (p<0.05). Next, we tested whether the lifespan extension by hif-1 is impacted by the ER stress pathway by measuring lifespan of the ire-1; hif-1 double mutant. The ire-1 mutation fully suppressed lifespan extension by hif-1 both under AL and under DR conditions (Figure 6B; Table 4 and S5). Although the ire-1 mutant is short-lived, the suppression of hif-1 lifespan by ire-1 is not due to sickness in general, but is specific for this longevity pathway. Previous studies showed that inhibition of the translation initiation factor-4G (ifg-1) extends lifespan in C. elegans [9],[12],[13]. ifg-1 RNAi and the hif-1 mutation have additive effects on lifespan (data not shown), suggesting they modulate longevity through different mechanisms. We observed that lifespan extension by ifg-1 RNAi is not dependent on ire-1 (Figure S5). We also found that RNAi knocking-down of xbp-1, an essential transcription factor downstream of IRE-1 for ER stress response, suppressed the lifespan extension by hif-1 (Figure S6).
pek-1 encodes the C. elegans PERK homolog that functions in another branch of ER stress signaling. A deletion mutant of pek-1 has a normal lifespan, and it does not affect hif-1 lifespan both under AL and under DR conditions (Figure S7). Taken together, our results indicate that the IRE-1 ER stress pathway is a key effector of both hif-1 and DR with respect to lifespan extension.
In order to determine whether HIF-1 modulates ER stress in response to nutritional variations, we examined the mRNA levels of hsp-4, which encodes an ER chaperone BiP ortholog in C. elegans. Under ER stress, hsp-4 transcription increases dramatically, and it is widely used as an indicator of unfolded/misfolded protein overload [66]. We used quantitative RT-PCR (qRT-PCR) to measure hsp-4 mRNA levels in wild-type N2, hif-1, and egl-9 animals under different nutrient conditions. DR significantly reduced hsp-4 transcription (Figure 6C), supporting the hypothesis that high nutrient levels correlate with high ER stress. The hif-1 mutant under AL, which has extended lifespan, also showed reduced hsp-4 mRNA compared to N2 under the same condition. With the DR treatment, egl-9 animals, which have a shortened lifespan, showed higher hsp-4 mRNA levels than N2 and hif-1 (Figure 6C). C14B9.2 encodes a protein disulfide isomerase, and it was identified as a target of the IRE-1 pathway in C. elegans from previous studies [67]. Under ER stress, C14B9.2 mRNA levels increase dramatically, and the increased transcription is dependent on IRE-1 and XBP-1, but not PEK-1 [67]. Using qRT-PCR, we found that DR reduces C14B9.2 transcription, and egl-9 animals have increased C14B9.2 mRNA levels under DR (Figure S8). Thus, lower mRNA levels of ER stress reporters, which indicate reduced ER-associated protein misfolding, accompanied genetic and environmental conditions that extend lifespan. These results support the role for ER signaling in lifespan extension both by DR and by hif-1. Since the ER stress markers do not totally correlate with lifespan phenotypes, e.g., egl-9 under AL has higher hsp-4 and C14B9.2 mRNA levels but a normal lifespan, there might be ER stress-independent factors functioning downstream of HIF-1 in lifespan determination.
We propose a genetic model that depicts a pathway of lifespan modulation by nutrients which links the TOR pathway, HIF-1 and ER stress in C. elegans (Figure 7). This model indicates that HIF-1 acts downstream of the nutrient-responsive TOR-S6K pathway, as egl-9 animals with elevated HIF-1 show diminished lifespan extension both under DR and by a mutation in S6K. Nutrients modulate ER stress, and the importance of ER signaling in determining lifespan is demonstrated by the finding that ire-1 not only suppresses HIF-1-dependent lifespan extension, but also shows significantly diminished DR-dependent lifespan extension. We also find that reducing nutrient intake lessens unfolded protein damage in ER as measured by ER stress markers in C. elegans. Interestingly, inhibition of HIF-1, which extends lifespan under the AL condition, was also able to reduce the ER stress under this condition. Conversely, the egl-9 mutants, which fail to show maximal lifespan extension by DR, had elevated levels of the ER stress markers under DR. Together, these experiments suggest that HIF-1 functions through the IRE-1 ER stress pathway to modulate lifespan extension by DR. Since DR-dependent lifespan extension cannot be fully suppressed by either increased HIF-1 or the mutation in ire-1, there are potentially other genes that function in parallel to hif-1 to determine DR-dependent lifespan extension.
Previous studies have identified HIF-1 as a key regulator in various cellular processes, including stress resistance, glucose metabolism, angiogenesis and cell death. Despite the well-characterized HIF-1 functions in age-associated physiological processes and the regulation of HIF-1 by known lifespan determinants, such as TOR and S6K, it has not been clear whether HIF-1 is involved in organismal aging. In this study, we found that a deletion mutant of hif-1 shows significant lifespan extension in C. elegans. Genetic epistasis experiments indicate that HIF-1 functions downstream of the nutrient-responsive TOR-S6K pathway to modulate lifespan. Using a modified sDR regime, we found that HIF-1 is involved in DR-mediated lifespan regulation, with overexpression of HIF-1 diminishes lifespan extension under DR.
Recent studies in C. elegans have identified novel genetic pathways that determine the beneficial effects of DR [49], [53]–[55],[68]. Using different regimens to restrict nutrients intake, previous studies have identified key regulators of DR-dependent lifespan extension, including transcription factors PHA-4, SKN-1, HSF-1 and the cellular energy homeostasis regulator AMP-activated protein kinase (AMPK). It is very likely that the maximal lifespan extension by DR is not achieved by regulating a single genetic pathway but that multiple pathways act together to mediate the lifespan effect of DR [69],[70]. The TOR pathway has been shown to play important roles in DR-dependent lifespan extension in D. melanogaster [6] and S. cerevisiae [7]. Our study underscores the importance of the TOR-S6K pathway as a conserved mediator of lifespan extension by DR in multiple species.
One of the most important questions about DR is how animals sense reduced nutrients to adjust gene expression, metabolism and behavior for extended survival. It has been shown that the SKN-1 transcription factor functions in ASI neurons to mediate liquid DR-induced longevity by an endocrine mechanism [53]. Using transgenic animals that express egl-9 cDNA with various tissue-specific promoters in the egl-9 mutant background [41], we found that restoring EGL-9 function in serotonergic neurons (ADF, NSM), body wall muscle and pharyngeal muscle can rescue lifespan phenotypes of egl-9 under DR. Serotonin is a neurotransmitter that regulates feeding, reproduction, and fat metabolism in C. elegans. [71],[72]. Recently, serotonin signaling has also been connected to aging in C. elegans [73],[74]. It will be interesting to examine whether HIF-1 regulates serotonin signaling, and whether serotonin signaling is involved in DR-dependent lifespan extension in C. elegans. A previous study has shown that muscle decline is one of the major physiological causes of C. elegans aging [62]. Whether DR and HIF-1 affect muscle structure and function during aging needs to be further investigated. The pharynx is the food intake organ of C. elegans. The FOXA transcription factor PHA-4 plays essential roles in pharyngeal development [75]. Interestingly, PHA-4 is required for lifespan extension by DR [54] and by a mutation in S6K [76], but lifespan extension by hif-1 is not dependent on PHA-4. These experiments suggest that there may be multiple pathways downstream of TOR/S6K that mediate lifespan extension in response to DR.
Abundant evidence has indicated that ER stress and protein homeostasis are important for aging [17],[59],[77],[78]. ER stress is caused by an overload of unfolded/misfolded proteins to ER [64]. ER stress activates the unfolded protein response through three signaling pathways transduced by IRE1, PERK, and ATF6. We showed that lifespan extension by DR or hif-1 requires functional IRE-1 and XBP-1 ER stress signaling. Reduced nutrients are associated with lower ER stress, which is consistent with findings in mice that increased nutrient intake or TOR activation is linked to increased ER stress [65],[79]. Our results implicating the role of ER stress in DR-mediated lifespan extension in C. elegans are also consistent with findings from S. cerevisae, in which the lifespan extension by DR was found to be partially dependent on GCN4 [17], which functions downstream of PERK in the unfolded protein response [80]. Our work describes a novel role for ER signaling in aging and DR-dependent lifespan extension, examination of which may help explain how protein homeostasis determines lifespan and age-related diseases.
DR has not only been shown to extend lifespan in rodents but also is one of the most robust methods to reduce tumorigenesis in mice [81]. However, the mechanisms by which DR causes this protection have not been elucidated. Our results implicate HIF-1 as a potential target in mediating the protective effects of DR on tumorigenesis in mammals. HIF-1 overexpression is frequently detected in various tumors due to intratumoral hypoxia and genetic mutations. HIF-1 owes its oncogenic properties to pleiotropic effects on a variety of cellular processes, including survival under hypoxia, angiogenesis, metastasis and glucose metabolism. Inhibition of HIF-1 has been proven to be an efficient way to prevent tumor growth, and HIF-1 is being extensively studied as an important target in cancer therapy. Our study suggests an important role for HIF-1 as an oncogene in the context of aging and nutrient sensing, which are both key risk factors in tumor formation.
Strains were cultured under standard lab condition as described [82]. Strains used in this work include N2, ZG31 hif-1(ia04) V, JT307 egl-9(sa307) V, GR1329 daf-16(mgDf47) I, CB1370 daf-2(e1370) III, DR1439 unc-24(e138) daf-15(m634) IV/nT1[let-?(m435)] (IV;V), XA8223 rsks-1(ok1255) III, XA8206 rsks-1(ok1255) III; hif-1(ia04) V, XA8208 rsks-1(ok1255) III; egl-9(sa307) V, DA1116 eat-2(ad1116) II, PKL21 eat-2(ad1116) II; egl-9(sa307) V, CX8628 egl-9 (sa307) V; kyEx1525 [H20::egl-9::gfp], CX8630 egl-9 (sa307) V; kyEx1527 [myo-3::egl-9::gfp], CX8632 egl-9 (sa307) V; kyEx1529 [tph-1::egl-9::gfp], CX8756 egl-9 (sa307) V; kyEx1593 [egl-9::egl-9::gfp], CX8792 egl-9 (sa307) V; kyEx1616 [myo-2::egl-9::gfp], CX8832, egl-9 (sa307) V; kyEx1639 [gcy-36::egl-9::gfp], CX9646 egl-9 (sa307) V; kyEx2109 [eak-4::egl-9::gfp], CX9778 egl-9 (sa307) V; kyEx2159 [col-19::egl-9::gfp], CX9779 egl-9 (sa307) V; kyEx2160 [hlh-6::egl-9::gfp], CX9807 egl-9 (sa307) V; kyEx2161[ttx-1::egl-9::gfp], CX9889 egl-9 (sa307) V; kyEx2215 [hum-5::egl-9::gfp], CX10002 egl-9 (sa307) V; kyEx2254 [unc-31::egl-9::gfp], CX10090 egl-9 (sa307) V; kyEx2288 [tdc-1::egl-9::gfp], CX10149 egl-9 (sa307) V; kyEx2321 [H20::egl-9::gfp, tdc-1::egl-9::gfp], RE666 ire-1(v33) II, XA8234 ire-1(v33) II; hif-1(ia04) V, PKL3 pek-1(ok275) X, and PKL6 hif-1(ia04) V; pek-1(ok275) X.
The msDR method was modified from previously described [49]. Overnight culture of E. coli OP50 grown at 37°C was centrifuged at 3,000 rpm for 30 minutes to collect bacteria cells. The bacterial pellet was washed with the S buffer, and the bacterial concentration was adjusted to 1.0×1012 cfu/ml. Serial dilutions were performed to achieve bacterial concentrations of 1.0×1011, 1.0×1010, 1.0×109, and 1.0×108 cfu/ml. Diluted bacterial cultures were spotted onto DR agar plates, which were modified from the standard nematode growth media (NGM) plates by excluding peptone and increasing agar from 1.7% to 2.0%. Carbenicillin (50 µg/ml) was added to the agar plates to further prevent bacteria growth. Synchronized L4 larvae growing under standard lab conditions (NGM plates with OP50 food, 20°C) were transferred to fresh NGM plates with OP50 food and 5 µg/ml of FUdR, and were incubated at 25°C overnight. Day 1 adult animals were then transferred to DR agar plates seeded with OP50 at different concentrations.
In the first week of lifespan experiments and heat stress assays, 5-fluorodeoxyuridine (FUdR) at 50 µg/ml was also added into the agar plates to prevent progeny from hatching.
Lifespan assays using standard lab conditions were performed as previously described [16]. Late L4 larvae growing at 20°C were transferred to fresh NGM plates with FUdR (5 µg/ml) and incubated at 25°C. The first day of adulthood is Day 1 in survival curves. Animals were scored as alive, dead or lost every other day. Animals that failed to display touch-provoked movement were scored as dead. Animals that died from causes other than aging, such as sticking to the plate walls, internal hatching or bursting in the vulval region, were scored as lost. Animals were transferred to fresh plates every 3–6 days. All lifespan experiments were performed at 25°C. Survival curves were plotted and statistical analyses (log-rank tests) were performed using the Prism 4 software (Graphpad Software, Inc., San Diego, CA, USA).
Wild-type N2 L4 larvae growing at 20°C were transferred to plates seeded with E. coli food at different concentrations. Those animals were transferred every day to fresh plates and progeny produced during that 24-hour period were counted.
Adult wild-type N2 animals grown on DR plates with bacterial food diluted at different concentrations for 48 hours since Day 1 adulthood were used for heat stress assays. The temperature was shifted from 20°C to 35°C and survival was scored. Animals that failed to display touch-provoked movement were scored as dead. Survival curves were plotted using the Prism 4 software (Graphpad Software, Inc., San Diego, CA, USA).
Adult N2, hif-1, and egl-9 animals grown under either AL or DR for 5 days since Day 1 adulthood were collected for total RNA preparations using the Trizol reagent (Invitrogen). The first strand cDNA was synthesized using the reverse transcription system (Qiagen). SYBR Green dye (Quanta) was used for qRT-PCR. Reactions were performed in triplicate on an ABI Prism 7000 real-time PCR machine (Applied Biosystems). Relative-fold changes were calculated using the 2−ΔΔCt method [83]. qRT-PCR experiments were performed twice with consistent results using two independent RNA preparations. The sequences of primers used were act-1 forward, CAA TCC AAG AGA GGT ATC CTT ACC CTC; act-1 reverse, GAG GAG GAC TGG GTG CTC TTC; hsp-4 forward, GGA AGC ATA TGC CTA TCA GAT G; hsp-4 reverse, CAG ATT CAA GTT CCT TCT TTT GC; C14B9.2 forward, GTT GTT CTC GCC AAG ATG GAC; and C14B9.2 reverse, GAT TGG TTC ACT CTT CTT TCC AGC.
|
10.1371/journal.pbio.1000119 | Transcriptional Dysregulation in NIPBL and Cohesin Mutant Human Cells | Cohesin regulates sister chromatid cohesion during the mitotic cell cycle with Nipped-B-Like (NIPBL) facilitating its loading and unloading. In addition to this canonical role, cohesin has also been demonstrated to play a critical role in regulation of gene expression in nondividing cells. Heterozygous mutations in the cohesin regulator NIPBL or cohesin structural components SMC1A and SMC3 result in the multisystem developmental disorder Cornelia de Lange Syndrome (CdLS). Genome-wide assessment of transcription in 16 mutant cell lines from severely affected CdLS probands has identified a unique profile of dysregulated gene expression that was validated in an additional 101 samples and correlates with phenotypic severity. This profile could serve as a diagnostic and classification tool. Cohesin binding analysis demonstrates a preference for intergenic regions suggesting a cis-regulatory function mimicking that of a boundary/insulator interacting protein. However, the binding sites are enriched within the promoter regions of the dysregulated genes and are significantly decreased in CdLS proband, indicating an alternative role of cohesin as a transcription factor.
| Appropriate segregation of chromosomes to daughter cells depends upon proper cohesion of sister chromatids during mitosis. The multiprotein cohesin complex and its regulators are key factors in this process. Intriguingly, recent work has shown that the cohesin complex also has other cellular roles, including a role in regulating gene expression. Additionally, mutations in cohesin structural and regulatory components have been linked to human multisystem developmental disorders such as Cornelia de Lange Syndrome (CdLS), but the role cohesin is playing in the pathogenesis of this disorder is unknown. To define the role that cohesin plays in regulating gene expression in human cells, we analyzed gene expression and genome-wide cohesin binding patterns in cells from normal subjects and from CdLS probands with mutations in the cohesin regulator NIPBL or in the cohesin structural component SMC1A. We found a strikingly conserved pattern of gene dysregulation in these different cell lines that correlates with disease severity and a significant correlation between gene dysregulation and cohesin binding around misexpressed genes. The observed pattern of binding and misexpression is consistent with cohesin having a putative role as a boundary/insulator interacting protein or transcription factor, the activity of which is disrupted in CdLS probands.
| Cohesin is an evolutionally conserved multisubunit protein complex consisting of an SMC1A and SMC3 heterodimer, and at least two non-SMC proteins SCC1 (also known as RAD21 or MCD1) and SCC3 (also known as SA or STAG). Cohesin controls sister chromatid cohesion during S phase with Nipped-B-Like (NIPBL) facilitating its loading and unloading [1]. ESCO2 possesses acetyltransferase activity and is involved in the establishment of cohesion [2]. Cohesin is loaded onto chromatin during G1/S phase in budding yeast and during telophase of the preceding cell division in vertebrates. Loading of cohesin also happens during G2/M phase when double strand DNA breaks are generated [3]. Removal of cohesin from chromosome arms begins during prophase and completes by separase-mediated dissolving of the remaining cohesin from centromeres during anaphase [3]. Although no consensus DNA sequence for cohesin binding has been demonstrated, cohesin is enriched at heterochromatin [4] and DNA double-strand breaks (DSBs) [5]. A large amount of intact and free cohesin is associated with chromosomes for most of the cell division cycle because of a separase independent mechanism [6]–[8]. A noncanonical role for cohesin as a key regulator of gene expression has been proposed [9]. The Drosophila NIPBL homolog, Nipped-B, alleviates the gypsy insulator function by assisting in long distance enhancer–promoter interactions to activate cut and Ultrabithorax expression. Nipped-B and cohesin colocalize and bind preferentially to active transcription units [9]. Recently, CTCF was reported to colocalize with cohesin and required for cohesin binding to chromatin [10],[11]. CTCF is the only protein known to bind to all vertebrate chromatin insulators and was initially identified as a transcription factor that binds to mammalian c-MYC promoters [12]. In addition, CTCF is well studied for its role in regulating genomic imprinting, and is proposed to regulate higher order chromatin structures such as intra- and interchromosomal association [13],[14]. CTCF is required for Tsix transactivation and involved in maintaining both X inactivation and escape domains, it stabilizes the repetitive sequences in several genetic disorders, and has been suggested to act as a tumor suppressor gene [15]. CTCF binding sites have been mapped in the human genome [16].
Cornelia de Lange syndrome (CdLS, Online Mendelian Inheritance in Man [OMIM] 122470, 300590, and 610759) is a dominant disorder with multisystem abnormalities including characteristic facial features, hirsutism, upper extremity defects, gastroesophageal dysfunction, growth, and neurodevelopmental delays. The incidence is about one in 10,000, with most cases being sporadic. There is equal gender distribution with a high degree of phenotypic variability. About 60% of the probands with CdLS have heterozygous mutations in the NIPBL gene, whereas 5% have mutations in the SMC1A gene, and one patient was found to have a mutation in the SMC3 gene [17],[18]. Other multisystem developmental disorders have been found to be caused by mutations in cohesin-related genes, such as Roberts-SC phocomelia (RBS, OMIM 268300) an autosomal recessive disorder caused by either homozygous or compound heterozygous mutations in the ESCO2 gene [19]. These disorders have collectively been termed “cohesinopathies.” Given the paucity of sister chromatid cohesion defects observed in individuals with CdLS [20], we hypothesize that it is the newly established role of cohesin in gene regulation that results in the multisystem phenotype when disrupted. To study the effects of disruption of cohesin on gene expression in human cells we have utilized lymphoblastoid cell lines (LCLs) from individuals with CdLS that harbor known heterozygous mutations in the cohesin regulator NIPBL and cohesin structural component SMC1A and applied a genome-wide approach to analyze gene transcription and cohesin binding.
LCLs from 16 severely affected CdLS probands with NIPBL protein-truncating mutations as well as 17 age, gender, and race matched healthy controls were used as training samples for assays on the Affymetrix HG-U133 plus 2.0 expression arrays, six additional individuals including one CdLS proband, one healthy control, two RBS probands (a related cohesinopathy), and two Alagille syndrome (AGS) probands (an unrelated multisystem dominant developmental disorder caused by disruption in the Notch signaling pathway) served as testing samples (Table S1A and S1B). 27,995 probe sets (12,740 nonredundant genes) were considered to be expressed in human LCLs. Unsupervised sample clustering by principle component analysis (PCA) of all the expressed probe sets was able to separate probands from controls in the training set indicating these two groups have different gene expression patterns (Figure 1A). Differential expression of these 27,995 probe sets was ranked by F = (between group variance)/(within group variance). Permutation analysis was performed 100 times and false discovery rate (FDR) was calculated for each F score, whereas redundancy was collapsed by keeping the ones with the highest F scores. We have identified a group of 1,915 probe sets (1,501 nonredundant genes) with FDR<0.05 and 420 probe sets (339 nonredundant genes) with FDR<0.01 (Tables S2 and S3) that are differentially expressed in CdLS. Heatmap representation of the expression levels of these genes clearly demonstrates that the expression of the 420 probe sets is remarkably different between CdLS probands and controls (Figure 1B). NIPBL itself had the highest ranking, with FDR = 0 and a fold change of −1.34.
In order to examine whether CdLS probands could be differentiated from controls through expression profiling, Leave-One-Out cross validation was performed on the training set. The top 400 probe sets were selected on each of the 33 rounds that corresponded to an FDR≈0.01. The left-out samples were successfully classified into two distinct groups using nearest centroid method with the exception of two controls and one proband that were misclassified (Figure 1C). The area under the receiver operating characteristic (ROC) curve is 0.985 with test accuracy of 91% (95% confidence interval = 76%–98%). Nearest centroid classification method was further performed on the six testing samples based on the identified 420 probe sets (FDR<0.01). The one healthy control and two individuals with AGS were classified as controls; one CdLS and two RBS probands were classified as probands (Figure 1C and Table S4). RBS is due to the mutations in the ESCO2 gene that also regulates cohesin, whereas AGS is an independent genetic disorder due primarily to mutations in the JAG1 gene, a member of the Notch signaling pathway. Thus, although limited to only two samples, it appears that RBS shares a similar transcription profile with CdLS, consistent with these two disorders being caused by disruption of the cohesin pathway. It is of interest that whereas the two RBS probands were classified as CdLS, their discriminant scores (DS) were actually midway between the scores of CdLS probands and the controls suggesting RBS might have an intermediate transcription profile to CdLS and controls (Figure 1C).
Clustering-based feature selection was carried out on the 339 nonredundant genes (420 probe sets, FDR<0.01) to identify independent pathways or functional groups. Five clusters were discovered (Table S5) and 32 genes (Table S6) were chosen for further custom array validation according to smaller FDR, bigger fold change, and less redundancy.
A cohort of 101 samples including individuals with different phenotypes (healthy, CdLS, or other disorders) and various gene mutations (Table S7) were measured on custom arrays carrying 56 probes mapped to the 32 selected genes (Tables S6 and S8). We have followed a step-wise procedure to identify classifiers according to different CdLS subgroups, and applied nearest centroid classifications on all 101 samples (Table S9A and S9B). Detailed analysis is described in Text S1. A 23-gene classifier can be used to categorize CdLS probands with NIPBL mutations from the rest of the samples including non-CdLS, CdLS probands with SMC1A mutations, and CdLS probands without an identifiable gene mutation. This indicates that the expression of these 23 genes is tightly correlated to NIPBL function. To improve the generality of the classifier, we randomly selected 15 mild probands with NIPBL mutations as a new training group. Expression of ten of the 23 genes was significantly different between this group and the original 17 controls and was also capable of subclassifying all CdLS probands from non-CdLS controls, regardless of the gene mutations or clinical presentations of the probands. This suggests that expression of these ten genes is affected by a CdLS specific disease process. For both classifications, the expression levels of the classifier genes are tightly correlated to disease severity. A clear progression of increasing discriminant scores (DS) can be seen from healthy controls through mild, moderate, and severe CdLS probands (Figure 2A and 2B). In addition, we have identified three genes NFATC2, PAPSS2, and ZNF608 that could be used as biomarkers for CdLS (Figure S1A and S1B). The phenotype associated gene expression profiles strongly suggest either direct or indirect roles for the identified genes.
Cohesin is a multisubunit complex constructed from SMC1A, SMC3, RAD21, and SCC3 subunits. Mutations in SMC1A or SMC3 and the cohesin regulator NIPBL lead to the human developmental disorder CdLS. To test the hypothesis that cohesin regulates transcription through its chromatin binding activity and that this association is regulated by NIPBL activity we undertook whole genome mapping of cohesin binding sites in LCLs from two healthy controls and one severely affected CdLS proband with an NIPBL mutation. Because of our inability to identify an effective antibody with high specificity against NIPBL or SMC1A, we chose an antibody against RAD21 (one of the other key components of the cohesin complex) to map genome-wide cohesin binding sites. Chromatin immunoprecipitation (ChIP) using a polyclonal antibody against human RAD21 was performed and products were hybridized on Affymatrix 2.0 tiling arrays. The score of model-based analysis of tiling-arrays algorithm (MAT) was calculated and probes were mapped to genomic positions. Peaks representing genomic regions bound by hRAD21 were identified with a p<10−6 and FDR<0.01.
The 54,675 probe sets on Affymetrix HG-U133 plus 2.0 expression array can be unambiguously mapped to 15,162 unique RefSeq mRNAs including 10,378 transcribed and 4,784 nontranscribed genes in LCLs. 78% of the 15,162 mapped genes do not harbor intragenic cohesin sites (Here, “intragenic” means genomic region from the transcription start site [TSS] of a gene to the transcription termination site [TTS] of the same gene), and cohesin binds at variable distances outside those genes. 22% of the 15,162 mapped genes harbor intragenic cohesin sites, this number is reduced to 19.0% in the silent nontranscribed genes (p≤7.2e−6) and no change in the disease neutral genes (22.9%, p≤0.0864); on the contrary, more of the differentially expressed genes harbor cohesin sites (27.0%, p≤7.44e−5) (http://145.18.230.98/Service/Statistics/Binomial_proportions.html) (Table S10) suggesting a correlation between intragenic cohesin binding and gene expression. For the 22% of genes with intragenic cohesin sites, cohesin preferentially binds to a narrowed region surrounding the TSSs or the TTSs with frequency at the TTSs only half of that at the TSSs. The 100-kb regions spanning upstream and downstream of the genes have only background levels of cohesin binding (Figure 3A and 3B). Among controls, the degree of cohesin binding within +/− 1 kb of the TSSs is greatest for those genes that are actively transcribed and especially in those genes that are differentially expressed in the NIPBL mutant cell lines, whereas the silent nontranscribed genes have the same, or lower level, of enrichment as the background level (Figure 4A). In addition, cohesin binding is enriched at the 5′-UTRs only for actively transcribed genes, with no binding difference at exons, introns, or 3′-UTRs between the actively transcribed and silent genes (Figure 4B). Identification of overrepresented cohesin binding near promoters suggests that cohesin may regulate gene expression as a transcription factor. In spite of this, the majority of the expressed genes (78%) do not harbor any cohesin binding sites in their intragenic regions, indicating most of the genes in the human genome may be regulated by cohesin independent pathways or cohesin is involved in their expression regulation through an alterative mechanism. We further evaluated 13 genomic loci based on their gene expression alterations to validate their cohesin binding status by the more sensitive method of ChIP-quantitative PCR (qPCR). Out of these 13 loci, two regions are equally bound by cohesin in both healthy and CdLS cells, two regions are not bound by cohesin in either healthy or CdLS cells, and the remaining nine loci demonstrated significant loss of cohesin binding in CdLS cells as compared to control cells. The ChIP-qPCR results are consistent with cohesin binding alterations detected by ChIP array studies (Figure S2A and S2B, Table S11).
The total number of cohesin binding sites is reduced by 29.7% (9,530 versus 13,560) in the examined CdLS proband, but the total number of binding sites at TSSs is reduced by 43.4% (448 versus 792) in the same proband, suggesting that cohesin is more likely to be removed from TSSs (43.4% versus 29.7%, p = 5.9e−8) (Table 1). The 10,378 genes expressed in LCLs have been statistically ranked for their misexpression in CdLS probands as described above. In the controls, there exist 666 LCL expressed genes that have cohesin binding sites mapped to the +/− 1-kb vicinity of TSSs, and 107 of them are identified as differentially expressed in CdLS (FDR<0.05). In CdLS, only 376 such genes have cohesin sites around their TSSs (376 versus 666, reduced by 43.5%), only 53 of the 107 differentially expressed genes still maintain their TSS/cohesin association, whereas the rest have all lost their TSS cohesin binding sites (53 versus 107, reduced by 50.5%) in the proband (Table 2). At the TSSs, the number of cohesin sites on differentially expressed genes is significantly reduced in CdLS, whereas the reduction is moderate for the nondifferentially expressed genes, and only minimal for those silent nontranscribed genes (Figure 4A). The binding between cohesin and TSSs of expressed genes is highly correlated to the CdLS phenotype. In our identified panel of differentially expressed genes in CdLS, Fisher testing on the ChIP data shows that these genes tend to attract more cohesin to their TSSs in control cells under the healthy condition (p = 10e−4) than the neutral genes do, whereas under the diseased condition in CdLS cells, those genes tend to lose their capability to recruit cohesin and associate with cohesin at a similar level to the neutral genes and have lost their statistically significant difference (p = 0.1) (Table 2). This 2-kb region (+/− 1 kb surrounding the TSS) was further analyzed for the entire group of 10,378 genes expressed in LCLs that have been ranked for their differential expression in CdLS probands as described above. Cohesin enrichment was clearly identified in control cells for the top ranked genes and a dramatic decrease in binding is demonstrated in the CdLS cells, suggesting that the genes that harbor more cohesin sites around the promoter regions are more likely to be misexpressed in CdLS (Figure 4C). Moreover, this difference was even more remarkable if we narrowed the analyzed region to the +/− 100-bp central area surrounding TSSs (Figure S3).
To summarize, in control LCLs cohesin preferentially binds to transcribed genes at the TSSs as compared to the silent nontranscribed genes. The binding sites are even more enriched for the differentially expressed genes. In CdLS, cells tend to lose cohesin binding globally, however the cohesin sites at TSSs are more likely to be lost, most notably for the differentially expressed genes where loss of cohesin binding at the TSSs results in a binding frequency approaching the background level. The preferential binding to promoter regions suggests cohesin may play a role as a transcription factor.
Recently cohesin has been functionally linked to CTCF, an insulator capable of blocking enhancers or preventing the spread of epigenetic signals [15]. In our study, the ion transporter protein ATP11A is significantly downregulated in CdLS (FDR = 0.027), although the fold change is small (−1.24). ATP11A locates within ENCODE region ENr132 on Chromosome 13 with four other genes. Therefore, the ENCODE datasets obtained from GM6990, a similar EBV-transformed human B cell line (http://genome.ucsc.edu/, http://www.sanger.ac.uk/PostGenomics/encode/data-access.shtml), were able to be adapted for our analysis [21]. There are six CTCF and two cohesin binding sites in this area, both cohesin sites overlap with CTCF sites. In controls, this area can be split into three chromatin regions according to multiple histone modification makers (Figure 5A and 5B) [22]–[24], and cohesin and CTCF colocalize at the border. Region 1 harbors only one gene C13orf35, which is not expressed in LCLs. Region 3 harbors three genes, MCF2L, F7, and F10, which are all expressed comparably in LCLs from both controls and probands. The ENCODE study has shown that chromatin-silencing marker H3K27me3 is enriched in region 3, but open chromatin markers H3K4 me1/me2/me3, H3K36me3, and H3K79me3, and DNaseI sensitive sequences are underrepresented, indicating chromatin in this region is condensed and transcription repressed [22]–[24]. In region 2, on the other hand, H3K4 is highly methylated, H3 tails are vastly acetylated, and multiple DNaseI sensitive sites appeared; meanwhile, H3K27 methylation level is quite low indicating region 2 is an active open chromatin domain. ATP11A is the only gene located in region 2 and differentially expressed in CdLS. Of note, the cohesin binding site between regions 2 and 3 at Chromosome 13: 112,645,000–112,645,600 is lost in CdLS (Figure 5A and 5B). ChIP-qPCR was then performed using specific primers to amplify this binding locus in an expanded sample set including three healthy controls and three CdLS probands (Figure 5C). Two of the three probands, PT2 and PT12, have NIPBL truncating mutations with severely affected clinical features and have been included in the whole genome expression array studies as described above; the third proband CDL-017 has a mutation in the SMC1A gene and manifests a much milder phenotype (Tables S1 and S7). Cohesin binding site 1 (Chromosome 3: 79653256–79653385), which was not lost in CdLS according to our qualitative array analysis was therefore used as a positive binding control. By quantitative PCR assays, the enrichment of cohesin bound to site 1 was not found to be changed between controls and the probands, which is consistent with the array findings. However, cohesin binding was dramatically reduced, within Chromosome 13: 112,645,000–112,645,600 among CdLS probands including the individual with the SMC1A mutation (Figure 5C). Although cohesin binding was not completely lost in CdLS by ChIP-qPCR, the result is consistent with the missing binding peak seen in the qualitative ChIP array analysis. Although this dataset is limited, it suggests that cohesin possesses a function as an insulator/boundary protein, in addition, functional NIPBL is required for this process. With disruption in the NIPBL mutated or cohesin subunit SMC1A mutated human cells, the silent chromatin signals from region 3 appear to be able to cross the boundary and migrate into region 2 to inhibit ATP11A transcription. Cohesin and CTCF may function cooperatively at this locus owing to their colocalization. In addition, both CTCF binding sites remained intact in CdLS, which may explain why the downregulation of ATP11A was not dramatic (−1.24).
Ingenuity Pathways Analysis (IPA) (Ingenuity Systems, Inc., http://www.ingenuity.com) was used to analyze the identified differentially expressed genes. Out of the 339 genes with FDR<0.01, 150 genes are documented in cancer, neurological, hematological, skeletal and muscular, and dermatological diseases; 150 genes are identified as major players in embryonic and tissue development, hematological and immune system development and functions; in addition, 153 genes have well established cellular and molecular functions in cell death, cell proliferation, and cell cycle regulation. We have further analyzed the biological functions and canonical pathways mediated by the 23- and 10-gene classifiers and the three biomarkers as validated by target array (Table S12). Interestingly, more than 60% (15 out of 23) of the identifier genes harbor intragenic cohesin binding sites, which is much higher than the average genome level (22%). Moreover, some of these genes have completely lost their cohesin association in CdLS (Table S12). Both groups of classifier genes are tightly related to pathways of cell death, cellular development, and tissue morphology. 12 out of these 23 genes are involved in 47 known biological functions or disease conditions. Five of these 12 genes are also part of the 10-gene classifier, including two genes, NFATC2 and PAPSS2, which are the identified biomarkers for CdLS. The 23-gene classifier could differentiate CdLS probands with NIPBL mutations suggesting the expression of these 23 genes are tightly controlled by NIPBL; whereas the 10-gene classifier is less powerful and only able to identify CdLS probands without the ability to differentiate subgroups of probands with different gene mutations, suggesting that these ten genes are related to terminal events during CdLS pathogenesis. Therefore, the common five genes, PAPSS2, NFATC2, MAP3K5, LTB, and PHF16, which are involved in multiple reported events by IPA and are shared by the two classifiers, might be involved in cellular functions that are universally affected in CdLS. Presumably mutations in NIPBL, SMC1A, or SMC3, or as of yet unidentified CdLS causing gene mutations, will all result in alterations of the related biological functions controlled by these five genes. On the other hand, the seven unique genes with functional roles that were excluded from the 10-gene classifier, KIFAP3, AIM1, BBS9 (PTHB1), TSPAN12, TRERF, ARHGAP24, and ID3, probably represent cellular functions affected more specifically by NIPBL mutations. Four genes, PAPSS2, NFATC2, MAP3K5, ADCY1, were identified to be involved in 32 canonical pathways by IPA; they also regulate multiple biological functions as mentioned above. ADCY1 is the single gene out of the above four genes that exists in the 23-gene classifier but is excluded from the 10-gene classifier; thus the specific canonical pathways regulated by ADCY1 (i.e. B cell receptor signaling, RAR activation, sulfur metabolism, and endoplasmic reticulum stress pathways), could largely depend on normal functions of NIPBL. Two out of the three biomarkers, NFATC2 and PAPSS2, are reported to be involved in multiple biological functions and canonical pathways by IPA analysis. The third biomarker ZNF608 is a novel protein with very minimal known functions. However, the zinc finger protein family members are known to be the major players in many molecular and cellular pathways.
One of the biomarkers, NFATC2, is involved in multiple signaling pathways during development and affecting skeletal myogenesis, chondrogenesis, axon growth, and guidance [25],[26]. Two NFATC negative regulators both locate to the Down syndrome critical region of human Chromosome 21, Nfatc2−/− and Nfatc4−/− double-knockout mice have physical and cognitive features resembling human Down syndrome [27]. Dysregulation of NFATC2 in the postnatal nervous system may contribute to mental deficiency in CdLS. Another biomarker, PAPSS2, plays a pivotal role in the biosynthesis of sulfate donors for sulfotransferase reactions. Its activity is important for normal skeletal development; recessive mutations on PASS2 cause the genetic disorder spondyloepimetaphyseal dysplasia (SEMD), Pakistani type and degenerative osteoarthritis [28]. Papss2−/− knockout mice have shortened limbs, reduced axial skeletal length, and complex facial features. Its transcripts were also present in the heart and brain in mouse embryos [29].
Cohesin consists of four major proteins SMC1A, SMC3, SCC1, and SCC3. NIPBL plays a role in shuttling cohesin onto and off the chromatin, although the exact mechanism of its action is poorly understood. All proteins in this pathway are evolutionally conserved from yeast to human [30]. Cell-cycle related sister chromatid cohesion, DNA repair, and homologous rearrangement are well established roles for the cohesin apparatus. A role for cohesin in regulating gene expression has also been proposed and appears to be more sensitive to subtle dosage alterations of the cohesin apparatus and its regulators than its canonical function in sister chromatid cohesion [31]. In both yeast and Xenopus, the loading of cohesin onto chromatin in G1 phase is functionally separable from the establishment of sister chromatid cohesion at S phase [32],[33]. In Drosophila, Nipped-B mediates interactions between the promoter and remote enhancers for cut and Ultrabithorax; heterozygous Nipped-B mutants diminish cut expression in the emergent wing margin without showing cohesion defects indicating sister chromatid cohesion is independent from cohesin regulated gene expression [34]. In mice, Pds5b mutants have multiple CdLS-like defects without flawed sister chromatid cohesion [35].
In humans, CdLS is caused by heterozygous loss-of-function mutations in the NIPBL ortholog of Nipped-B and, in a smaller percent of cases, by mutations in the SMC1A or SMC3 cohesin subunit genes [17],[18],[36],[37]. Given the constellation of developmental abnormalities present in individuals with CdLS, with only a subset showing minor cohesion defects [20],[38], it is likely that the alterations of cohesin regulation and structure seen in these individuals result in gene expression dysregulation. We chose to use an easily accessible but a seemingly developmentally irrelevant tissue, LCLs, for these studies. We hypothesized that congenital genetic disorders might arise, in part, through dysregulation of expression of specific genes and that expression differences between affected and unaffected individuals might be present in tissues other than disease presenting tissues. These cells also provide an invaluable resource of naturally occurring mutant cohesin proteins (both structural and regulatory components of cohesin) that can be used to study the cellular processes regulated by this complex and specifically the impact on regulation of gene expression. Surprisingly these cells may also provide valuable insight into human developmental processes as well.
We have identified specific gene expression profiles for CdLS that are capable of classifying probands and tightly correlate with disease severity. Cohesin preferentially binds to promoter regions of the actively expressed genes suggesting a role as a general transcription factor. These binding sites are significantly reduced in NIPBL mutant CdLS samples. This result is likely due to NIPBL's direct role in cohesin loading on chromatin, which in turn affects transcriptional regulation at specific loci and would contribute to the CdLS pathogenesis. Out of the 339 dysregulated genes with FDR<0.01, 202 were upregulated (59.6%) and 137 were downregulated (40.4%), more genes were reactivated than inhibited with mutations in NIPBL (59.6% versus 40.4%, p = 3.44e−17) suggesting that NIPBL and cohesin can result in both negative (as transcriptional repression) and positive (as transcriptional activation) effects on expression. A similar percentage of upregulated and downregulated genes was also observed among the 1,501 dysregulated genes with FDR<0.05. Moreover, 71 of the above 339 genes (20.9%) and 207 of the 1501 genes (13.8%) have fold changes larger than 1.5, whereas the highest fold changes are −4.2 and +4.6, respectively. Although the majority of expression levels seemed only mildly perturbed, developmental deficits in CdLS are likely due to a cumulative change in multiple genes. Another reason for less remarkable expression differences could be the LCL tissue type used for this study, with bigger fold changes in more genes possibly present in more directly affected tissues of, e.g., brain or limb, and at specific times during embryonic development. However, it is also possible that the transcriptional dysregulation may be directly mediated by NIPBL through a yet uncharacterized mechanism and the reduced cohesin binding may be a secondary effect. In our study, a 30% reduction in NIPBL message was able to trigger a 29.7% (9,530 versus 13,560) reduction in cohesin binding sites in CdLS probands and further affects the transcription of specific genes.
The central components for sister chromatid cohesion, RAD21 (SCC1), SMC1A, SMC3, STAG2, ESCO1, ESCO2, and PDS5A (also known as SCC-112), are all expressed similarly between controls and CdLS probands with NIPBL mutations. However, STAG1, PDS5B (also known as APRIN), MAU2L (KIA0892), as well as several other genes with functions related to sister chromatid cohesion were significantly dysregulated in NIPBL mutant CdLS probands (FDR<0.05) (Table S13), suggesting that the cohesin pathway itself is affected by mutant NIPBL. MAU2 (KIAA0892) is the putative human homolog of scc4 in Caenorhabditis elegans [31],[39]. It forms an essential loading complex with NIPBL that regulates cohesin-chromatin association, sister-chromatid pairing, and mitotic checkpoints in HeLa cells. Physical association between NIPBL and MAU2 is indispensable for their stability, as depletion of either of the two proteins would subsequently diminish the cellular level of the other one [39]. In our study, decreased NIBPL transcription (−1.33, FDR = 0) was able to upregulate the transcription of MAU-2 (+1.11, FDR = 0.026), suggesting a functional compensation may exist for cohesin loading in CdLS. A cohesin-independent mechanism has also been suggested to exist. Condensin complexes [40], origin recognition complexes (ORCs) [41], centromere complexes [42], and DNA catenation [43] have each been reported to play a role in mediating cohesin-independent sister chromatid cohesion. Genes involved in these functions are also found to have dysregulated expression in NIPBL mutant individuals (Table S13). This finding indicates a subset of genes regulated by NIPBL are tightly involved in sister chromosome segregation events, but expression alteration may be required to pass a certain threshold in order to induce visible cohesion defects. This observation could explain why cell lines derived from CdLS probands did not demonstrate significant sister chromatid pairing problems. In contrast to CdLS, cohesion defects have been reported in three human developmental disorders: RBS (OMIM 268300) [19], Rothmund–Thomson syndrome (RTS, OMIM 268400) [44], and α-Thalassemia/mental retardation syndrome, X-linked (ATRX, OMIM 301040) [45]. Interestingly, although the expression of the RBS disease causative gene ESCO2 was not dysregulated in CdLS cell lines, the other two disease genes, ATRX and RECQL4, both demonstrated dysregulation in NIPBL mutant cell lines (Table S13).
Several cohesin targets have been identified. Steroid hormone ecdysone receptor (EcR), which is the Drosophila homolog of human NR1H3, was suggested to be regulated by Smc1, and Runx3 was identified as a direct target of Rad21 in zebrafish [46],[47]. The fact that both of these genes were also significantly dysregulated (FDR<0.05) in CdLS probands with NIBPL mutations indicates that NIPBL may first affect cohesin proteins and subsequently dysregulate cohesin targets. Surprisingly, we did not find that cohesin directly binds to these two genes in the cell line studied, which raises the possibility that cohesin may regulate their expression over long distances. When comparing ChIP-on-chip results for Nipped-B and/or SMC1A binding sites in three different Drosophila cell types [48], homologs of 20 differentially expressed human genes in CdLS probands (FDR<0.05) were also found to be bound by NIPBL and cohesin (unpublished data). Eight of these 20 genes are also bound by cohesin in humans suggesting they may be cohesin targets in both Drosophila and humans. It also suggests that cohesin mediated transcription is a conserved biological event. Moreover, most of the binding sites were lost in CdLS cells indicating dysregulated gene expression correlates with loss of cohesin binding. Among the eight genes, KMO, ELL2, and ARHGAP17 have cohesin binding at TSSs; ROBO1, UBE2H, MED13L, RASA3, and PDPK1 had cohesin binding within intronic regions. One of these genes, ROBO1 (homolog of lea in Drosophila), is of particular interest as it was found to have a fold change of 4.6, which is the largest among all the genes on the array. ROBO1 has been associated with dyslexia, a neurocognitive disorder of language and graphic processing that could be due to the abnormal migration and maturation of neurons during early development.
We have identified groups of 23, 10, and 3 genes as CdLS classifiers or biomarkers that are capable of differentiating CdLS from non-CdLS samples. The expression levels of these genes also correlate to the phenotypic severity of this disorder, although it is not clear at this time how the dysregulation of these particular genes might contribute to the phenotypes. More than 60% of the identifier genes harbor intragenic cohesin binding sites with some of them lost in CdLS proband. The obvious overrepresentation of genes carrying intragenic cohesin binding sites among the CdLS classifier genes further suggests that expression of the dysregulated genes is tightly related to the availability of cohesin binding. Overall, the majority of genes do not carry known cohesin binding sites, indicating that cohesin may play an upstream role in regulating human genes, or cohesin may enact regulation on some of the genes through distal cis- or trans-regulatory elements. The potential role for cohesin independent NIPBL regulation can not be excluded.
Cohesin has recently been found to be physically and functionally associated with the vertebrate insulator protein CTCF. In our study cohesin binds to only ∼20% of genes intragenically. This distribution does not change much between expressed genes and silent genes, and between differentially expressed genes in CdLS and disease neutral genes. Cohesin could be involved in gene regulation, like CTCF, by either binding to promoter elements and having a direct influence on the transcriptional machinery or by binding to intergenic cis-elements such as insulators to control gene expression from remote distances [49]. In our study, we have detected a potential boundary effect of cohesin at the ATP11A gene locus that suggests, for the first time in humans, that cohesin may bind to insulators and regulate transcription. Reduced cohesin binding at this locus was further validated in three additional CdLS probands by the more sensitive ChIP-qPCR including probands with either NIPBL mutations or cohesin subunit SMC1A mutation. However, cohesin does not exactly mimic the function of CTCF, at least in LCLs. Some CTCF target genes, such as PIM-1 [50] and APP [51], although expressed in LCLs, are neither dysregulated in CdLS nor do they lose cohesin binding at their regulatory regions. On the other hand, the CTCF target gene, BRCA1 [52], was downregulated in CdLS (−1.2, FDR = 0.017) but without losing cohesin binding sites. Additional quantitative analysis or ChIP-qPCR to study more genomic loci will delineate a clearer picture of cohesin and CTCF effects on transcriptional regulation. The role cohesin plays in imprinting and X inactivation remains unclear [53].
In summary, we have undertaken a genome-wide approach to study gene expression and cohesin binding in NIPBL mutant human samples. On the basis of our data and previously reported studies, we propose that NIPBL may be involved in modulating cohesin function through various mechanisms. Besides its canonical role in regulating sister chromatid segregation proposed by Haering et al. [54] (Figure 6A), cohesin may also regulate transcription (1) as an insulator protein by acting alone or with CTCF, or (2) as a transcription factor by binding to promoter elements. While regulating transcription, NIPBL may also serve as a cohesin shuttle to chromatin that leads to decreased cohesin binding when NIPBL is mutated. Data from this study are quite consistent with this role. Whether this loading mechanism either partially overlaps with, or is completely independent from NIPBL-mediated sister chromatid cohesion remains unknown (Figure 6B). NIPBL and cohesin may very well form one protein complex binding to regulatory elements of target genes, with NIPBL mutations affecting the regulatory capacity of this complex (Figure 6C). The colocalization of NIPBL and cohesin seen in Drosophila studies could be consistent with this model [9]. A third possibility is that NIPBL is able to maintain an accessible chromatin structure for cohesin binding whereas defective NIPBL leads to reduced accessibility for cohesin at specific chromosomal loci (Figure 6D).
This study was conducted according to the principles expressed in the Declaration of Helsinki. The study was approved by the Institutional Review Board of The Children's Hospital of Philadelphia and Misakaenosono Mutsumi Developmental, Medical, and Welfare Center. All patients provided written informed consent for the collection of samples and subsequent analysis.
All participants were evaluated by one or more experienced clinicians. Gene mutations were confirmed by sequencing, and most of the cases have been previously reported by our laboratories [17],[55],[56].
LCLs were grown uniformly in RPMI 1640 with 20% fetal bovine serum (FBS), 100 U penicillin/ml, 100 µg streptomycin/ml sulfate, and 1% L-glutamine. To identify differentially expressed genes between CdLS probands and controls, age and gender matched samples from 16 normal controls of European descent and 17 clinically severely affected probands of European descent with NIPBL protein-truncating mutations (nonsense or frameshift) were chosen as the training set for the discriminate analysis. To validate the expression pattern obtained from the training set, six samples including one healthy control, one Egyptian CdLS proband, two Roberts syndrome probands, and two Alagille probands were used as the testing set. All 39 cell lines were grown anonymously and the processing of these 39 cell lines were randomized by genotypes to eliminate batch effects that may contribute to genotype-specific gene expression. Samples are listed in Table S1A and S1B with detailed description. For custom array analysis, detailed information of these samples is listed in Table S7. Out of these 101 samples of European descent, the training set included 17 healthy controls and 14 severely affected CdLS probands with NIPBL protein-truncating mutations. All 31 samples were also used for the training in Affymatrix array analysis. For the testing set, all new samples were selected, which included four healthy controls, six severely affected probands, 13 moderately affected probands (nine have NIPBL mutations and four do not have an identifiable mutation), and 34 mildly affected probands (26 have NIPBL mutations and eight do not have an identifiable mutation). We have also included nine CdLS probands with SMC1A mutations, as well as four samples with different genetic diagnoses (two AGS, one Roberts syndrome, and one unknown multisystem genetic disorder). As above, samples were processed anonymously and randomly.
5×106 exponentially growing cells were seeded in 15 ml media in a 75-ml Falcon flask, and fed exactly after 24 h. After an additional 24 h on day 3, 8 ml of the media was removed and cells were pelleted by centrifuge and RNA extraction was performed immediately.
Total RNA from each sample was extracted with the RNeasy Mini-kit (Qiagen), synthesis of double-stranded cDNA was performed using SuperScript Double-Stranded cDNA Synthesis kit (Invitrogen), and cleaned up with GeneChip Sample Cleanup module (Affymetrix). The resulting products were then used to synthesize biotin-labeled cRNA with Enzo Bioarray High Yield RNA Transcript Labeling kit (Enzo Life Sciences) and further fragmented to 35–200-bp oligos. All procedures were done according to manufacturer's instructions. 30 µl fragmented cRNA at the concentration of 500 ng/µl was sent for hybridization in the microarray facility at The Children's Hospital of Philadelphia. Microarray hybridizations were performed by using HG-U133 plus 2.0 GeneChips (Affymetrix). The HG-U133 plus 2.0 contains ∼54,000 25-mer probe sets that covers approximately 47,000 transcripts and variants out of which 38,500 are well-characterized human genes. After hybridization and washes, arrays were scanned and analyzed both for genes that were present and for expression level using Microarray Analysis suite (MAS) 5.0 using default settings according to manufacturer's instructions.
The same RNA isolation process was performed as above. 32 genes were selected by clustering-based feature selection and 59 probes were designed (Table S8). Probe designing, RNA labeling, and hybridization were conducted using the Ziplex workstation (Xceed Molecular, http://www.xceedmolecular.com/). In brief, concentrations of the isolated RNA were determined by measuring the absorbance at 260 nm. All total RNA samples were of acceptable purity (ratio of the absorbance at 260 nm to 280 nm of 1.75 or greater). The integrity of the total RNA was determined to be acceptable for all samples (RNA Integrity Numbers measured with the Agilent 2100 Bioanalyzer RNA 6000 Nano assay were greater than 9.0). A custom Ziplex TipChip microarray containing oligonucleotide probes of between 35 and 50 bp for 32 genes was used to profile differences in gene expression between the LCL samples. Total RNA (500 ng) from 108 independent samples was amplified and biotin labeled with the Illumina Totalprep RNA amplification kit (Ambion). The concentrations of the labeled aRNAs were determined by measuring the absorbance at 260 nm, and 3 µg was hybridized on the custom TipChip with the Ziplex Automated Workstation protocol (Xceed Molecular). After hybridization, the Ziplex Automated workstation software automatically quantified spot intensities and reported background subtracted expression values. The Ziplex software automatically evaluated attributes of each spot to identify spots that did not conform to quality control criteria and reported the mean value of the duplicate spots of each probe that passed quality control.
Two healthy controls and one severely affected CdLS proband with an NIPBL protein-truncating mutation (G5483A) were used. Cells were crosslinked with 1% formadehyde at 70%–80% confluency for 10 min, chromatin was then prepared after quenching with 125 mM glycine and ChIP was performed as described [57] using anti-hRAD21 polyclonal antibodies (Abcam, ab992). In brief, lysates from crosslinked cells were incubated with the antibodies and preabsorbed protein A Affiprep beads (Bio-Rad) for 14 h at 4°C and for 2 h at 4°C, respectively. After washing, the beads were incubated in the elution buffer (50 mM Tris, 10 mM EDTA, 1% SDS) for 20 min at 65°C. The elutes were treated with proteinase K for 1 h at 37°C and followed by 65°C overnight incubation for crosslink reversal. The samples were then treated with RNase and phenol-chloroform purified for one time, and further purified using PCR purification kit (Qiagen) with 80 µl water used for the final elution.
The eluted chromatin was amplified and labeled with biotin then hybridized to high-density oligonucleotide tiling arrays (Human tilling 2.0R array, Affymetrix) as described [58]. A sample of DNA prepared from whole cell extract (WCE) was prepared in the same way. ChIP and WCE samples were hybridized on arrays according to the manufacturer's instructions, two technique replicates were used for each sample. After scanning and data extraction, enrichment values (ChIP/WCE) were calculated by using the MAT algorithm [59]. MAT is designed for high-density oligonucleotide tiling-array analyses in higher eukaryotes that could reduce probe-specificity biases because of genome complexity or high GC content. The resulting MAT scores are proportional to the logarithm transformed value of the fold-enrichment of the ChIP-chip samples [59]. We mapped MAT scores to positions in human genome assembly Hg 18 (NCBI Build 36). Bandwidth, MaxGap, and MinProbe parameters were set to 250, 1,000, and 12, respectively. The cutoff threshold of p-values was set to 1×10−6, which was equivalent to MAT scores higher than 4.85. FDRs were also calculated with every experiment less than 1% (Figure S4A and S4B). BED files were created, data were visualized in the Integrated Genome Browser (IGB) (http://www.affymetrix.com/support/developer/tools/download_igb.affx) and University of California Santa Cruz (UCSC) genome browser custom track (http://genome.ucsc.edu/).
ChIP was performed as described above using hRAD21 and control antibodies. ChIP-qPCR analysis was performed as previously described [10]. ChIP samples (2 µl) were used for one 25-µl PCR reaction. Analyses by qPCR were performed using a Platinum SYBR Green qPCR SuperMix UDG (Invitrogen) on an ABI 7500 cycler. The results were presented as fold-enrichment over control ChIP.
Gene expression microarray data were processed by DNA-Chip Analyzer (dChip) (http://www.dchip.org) using PM-only background subtraction and invariant set normalization. Differential gene expression between controls and CdLS probands was ranked by the ratio of between- and within-group variance (F statistic). During nearest centroid classification, distance of testing samples to training group centroids was measured as their Pearson's correlation coefficient. Statistical analyses were performed within R software environment (http://www.r-project.org). PCA and heatmap plots were generated by Spotfire DecisionSite version 9.1.1 (Spotfire, Inc.). More details about data analysis are provided in Text S1.
Genomic sequences reported in this manuscript have been submitted to NCBI GEO (http://www.ncbi.nlm.nih.gov/geo): gene expression data are under accession number GSE 12408 and ChIP-chip data are under accession number GSE 12603.
|
10.1371/journal.ppat.1003020 | Genome Structure and Reproductive Behaviour Influence the Evolutionary Potential of a Fungal Phytopathogen | Modern agriculture favours the selection and spread of novel plant diseases. Furthermore, crop genetic resistance against pathogens is often rendered ineffective within a few years of its commercial deployment. Leptosphaeria maculans, the cause of phoma stem canker of oilseed rape, develops gene-for-gene interactions with its host plant, and has a high evolutionary potential to render ineffective novel sources of resistance in crops. Here, we established a four-year field experiment to monitor the evolution of populations confronted with the newly released Rlm7 resistance and to investigate the nature of the mutations responsible for virulence against Rlm7. A total of 2551 fungal isolates were collected from experimental crops of a Rlm7 cultivar or a cultivar without Rlm7. All isolates were phenotyped for virulence and a subset was genotyped with neutral genetic markers. Virulent isolates were investigated for molecular events at the AvrLm4-7 locus. Whilst virulent isolates were not found in neighbouring crops, their frequency had reached 36% in the experimental field after four years. An extreme diversity of independent molecular events leading to virulence was identified in populations, with large-scale Repeat Induced Point mutations or complete deletion of AvrLm4-7 being the most frequent. Our data suggest that increased mutability of fungal genes involved in the interactions with plants is directly related to their genomic environment and reproductive system. Thus, rapid allelic diversification of avirulence genes can be generated in L. maculans populations in a single field provided that large population sizes and sexual reproduction are favoured by agricultural practices.
| Plant disease resistance often relies on simple “gene-for-gene” systems and, in the pathogen, a mutation in a single “avirulence” gene matching the plant resistance gene is sufficient to render the resistance ineffective. In agricultural systems, breeding for resistance is challenged by both the high evolutionary potential of the pathogen and the large scale of crop production; together, these factors encourage “breakdown” of novel sources of resistance soon after their deployment. Here, we established a four-year field experiment to evaluate the mechanisms and speed with which a fungal pathogen of oilseed rape, Leptosphaeria maculans renders ineffective the novel resistance gene Rlm7. The pathogen showed a very high evolutionary potential; the proportion of isolates in the population that were virulent against Rlm7 increased from 0 to 36% in four years. The experiment demonstrated that an extremely diverse range of molecular events leading to virulence, from more or less extensive nucleotide mutations or deletions to complete gene deletion, can occur in a single field. These results suggest that the genomic environment of the avirulence gene and the reproductive regime of the pathogen promote mutability at a single locus to produce virulence. Cropping practices that promote large pathogen populations and encourage sexual reproduction therefore favour rapid adaptation of the pathogen to the novel resistance.
| Fungi are the most important pathogens of cultivated plants, with significant economic, food security and environmental impacts, the latter being due to the large quantities of fungicides used to control plant diseases [1]. In contrast to fungicide use, genetic resistance against pathogens in crops is an environmentally friendly strategy to control diseases. Frequently, effective resistance has been provided by the introduction of major resistance (R) genes into crop genotypes [2]. Unfortunately, fungal pathogens have an incredible plasticity with which they can respond to their environment. Furthermore, their ability to adapt to changes in their environment and to disseminate these adaptations makes them very successful in countering crop defenses and control methods [1]. The rapid emergence of new strains able to render ineffective new R genes is thus a common feature of fungal phytopathogens [3].
Durable disease resistance has always been a goal of plant breeding programs since it is cost-effective and environmentally friendly whilst promoting the conservation of rare genetic resources [4]. Durability of resistance is largely dependent on the biology of the pathogen and the evolutionary potential of the pathogen population [3]. In addition, cropping practices such as crop rotation and stubble management may directly affect the evolutionary potential of the pathogen by reducing its population size or dispersal or by interfering with its reproductive regime [5].
Major gene resistance depends on the “gene-for-gene” concept in which the host R protein interacts with a corresponding pathogen avirulence (Avr) protein to initiate plant disease responses and resistance [6]. Avr proteins are now known to be effectors involved in plant pathogenesis that were recognised by the plant surveillance machinery in the course of plant-pathogen co-evolution [7]. Thus, avoidance of recognition by host R proteins often involves pathogen effector gene inactivation, which may result in a fitness penalty for the pathogen [4]. An additional level of complexity arises because many fungal or oomycete pathogens have their effector genes in non-conventional, adaptable, regions of their genomes [8]. It was recently suggested that the location of effector genes of Leptosphaeria maculans, the cause of “phoma stem canker of oilseed rape”, in AT-rich, transposable element (TE)-rich blocks of the genome has major implications for gene mutability, resulting in either allelic diversification or gene inactivation under R gene selection [9]. For example, it was suggested that Repeat Induced Point mutation (RIP), a fungal-specific mechanism of inactivation of repeated sequences in genomes, was acting on single-copy genes embedded in TE-rich blocks of the genome to favour the allelic diversification of such genes [9], [10], [11].
The cloning and characterization of the molecular determinants of resistance from the plants (R genes) and pathogens (Avr genes) has not only improved understanding of their functions and interactions but also provided the molecular information for detecting mutations in these genes. Fungal Avr genes have been cloned from only seven species, and the mechanisms by which fungal Avr genes evolve to evade host recognition is only documented for few Avr genes, mostly in Cladosporium fulvum, Magnaporthe oryzae, Rhynchosporium secalis and L. maculans [12], [13]. Moreover the link between laboratory and field studies has often been missing, and when field populations of pathogens were investigated for mutations in Avr genes they have often been isolated from uncharacterized crop genotypes or have been collected and analysed a long time after the corresponding R gene had been commercially deployed in crops (e.g. [14]). There is currently no example of identification of the initial events leading to virulence when a fungal pathogen is exposed for the first time to an R gene deployed under field conditions.
L. maculans is a pathogen with a high evolutionary potential combining large population size, mixed reproduction regime and high dispersal ability. Following sexual reproduction taking place on stem debris, leaf infections by ascospores (in autumn in western Europe) cause phoma leaf spots, supporting asexual multiplication. The life cycle of the pathogen is completed by a lengthy symptomless colonisation phase when the pathogen grows from the leaf lesions along the petiole to the stem, where cankers develop to cause lodging and yield losses in crops at the end of the growing season (spring and early summer in western Europe). Sexual mating between the numerous isolates that colonised the stem tissues then takes place [12]. Major gene resistance against L. maculans has been widely used in oilseed rape [15], [16], but was rendered ineffective in only a few growing seasons [17], [18].
Three L. maculans Avr genes, AvrLm1, AvrLm6 and AvrLm4-7 have been cloned [19]–[21]. All encode Small Secreted Proteins (SSPs) embedded in large TE-rich blocks of the genome termed AT-isochores. AvrLm4-7 is located within a 96-kb AT-isochore which only contains two other genes located 26 and 28 kb away. These two genes are also predicted to encode SSPs. AvrLm4-7 is recognised by two distinct R genes, Rlm4 and Rlm7 and escape from recognition by Rlm4 is due to a single-base non-synonymous mutation resulting in a Gly120Arg change in the protein. This change does not alter recognition by Rlm7 [21]. Before 2003, L. maculans has not been exposed to the Rlm7 selection in Europe and a population survey done in 2000–2001 in France identified only one virulent isolate out of 1787 isolates (0.05%) [22]. Rlm7 has been introduced in commercial cultivars in 2003 in France, with only 1% of the hectarages cropped with Rlm7 cultivars until 2005 (X. Pinochet, CETIOM, personal communication), thus providing us with the opportunity to survey emergence of virulent L. maculans isolates at the time of initial selection pressure and to identify the first molecular events responsible for the overwhelming of the resistance gene.
Here, we established a four-year field experiment (Figure S1 in Text S1, Figure S2 in Text S1) and combined molecular genetic and population genetic approaches to evaluate speed and patterns of mutations in the AvrLm4-7 gene responsible for the loss of the AvrLm7 specificity in L. maculans populations exposed to Rlm7 selection. Molecular analysis of events leading to the virulent phenotype in a single field revealed a tremendous diversity of mutation events, and confirmed the importance of the genomic environment in gene mutability.
We established a four-year (2004–2005 to 2007–2008 growing seasons) field experiment at Grignon, France, during which Rlm7 and rlm7 cultivars were grown alongside each other (with relative ca. 2/3 of the area cropped with the Rlm7 cultivar and 1/3 cropped with the rlm7 cultivar) (Figure S1 in Text S1, Figure S2 in Text S1). Exposure of the L. maculans population to Rlm7 was maximised because there was no crop rotation or ploughing in crop debris. Before the start of the experiment at Grignon (i.e. between 2000 and 2004) no Rlm7 cv. had been grown and the frequency of virulent avrLm7 isolates was minimal at both Grignon and another crop located 12 km away at Versailles (Figure S1A in Text S1), with estimated frequencies of avrLm7 isolates ranging from 0.006% to 1.3% at Versailles and from 0 to 1.3% at Grignon (Figure 1, Table 1, Table S1 in Text S1).
In the experimental field, the phoma stem canker was not severe in the summer of 2007 or 2008. The G2 disease indices, a disease severity index summarising the proportions of plants observed within six canker severity classes and ranging between 0 (all plants healthy) to 9 (all plants with severe canker) [4], were 1.53 (rlm7 cv.) and 0.86 (Rlm7 cv.) in 2007. However, the G2 index had increased to 2.34 on the Rlm7 cultivar by 2008, reflecting a localized overwhelming of the Rlm7 resistance during the course of the experiment.
In the course of the four-year experiment, we collected 2551 isolates either from the Rlm7 cv. Exagone or from the rlm7 genotype Campala (Grignon and Versailles; Table S2 in Text S1). Of these, 1987 isolates were characterized for their interactions with Rlm4 and Rlm7 plant genotypes (Table S2 in Text S1). The number of virulent avrLm7 isolates remained low or undetectable in the control plot at Versailles on the susceptible cv. Campala. At Grignon, the frequency of virulent avrLm7 isolates on the susceptible rlm7 cv. grown next to the Rlm7 cultivar steadily increased to 36.2% of the population, whilst avrLm7 isolates were not detected in crops located less than 600 m from the experimental plot (Figure 1, Table 1, Figure S1 in Text S1, Table S2 in Text S1). Consistent with analyses of L. maculans ascospore dispersal indicating that most spores are deposited within 500 meters from the source and especially in the first 100 meters [23], [24], this suggests that the observed increase in avrLm7 isolate frequency in the field experiment was due to the recurrent local use of Rlm7 and not to an increase of its frequency at the regional level.
Nucleotide polymorphisms in AvrLm4-7 were analysed in 169 AvrLm7 isolates from the sample. A very low level of sequence polymorphism was found within the gene with only five polymorphic nucleotides, all corresponding to non-synonymous mutations in the protein (Table 2, Figure 2B). The combination of these polymorphic sites generated five haplotypes (Table 2). In accordance with previous data [21], one haplotype included those of the isolates that showed the AvrLm4 specificity and differed from the other haplotypes by the presence of a guanine at base 358 resulting in a glycine at amino acid 120 (Table 2). All other mutations were found in isolates that had lost the AvrLm4 specificity but maintained the AvrLm7 avirulence, thus defining four distinct alleles for avrLm4-AvrLm7 isolates (Table 2).
Of the 808 virulent avrLm7 isolates obtained from Grignon, 769 were characterized for molecular events responsible for virulence. Numerous mutational events responsible for loss of the avirulence function were identified, single-or di-nucleotide deletions, single nucleotide polymorphisms (SNPs), wide degeneracy due to RIP, complete or partial deletion of the gene, major chromosomal rearrangements and alteration in gene expression. Only a few isolates had SNPs, single-nucleotide deletions, or under-expression of the gene whereas gene deletions and RIP mutations were present in 62.8% and 24.1% of the virulent isolates, respectively (Figure 2A). In most cases the mutational events resulted in lack of protein production or production of a severely truncated protein, while in a few other cases the 3-D structure of the protein was probably modified (e.g. mutation in cysteine residues involved in disulfide bridge formation) (Figure 2B, Table 3). Lastly, in a few cases, the sequence of the gene was unaltered but its expression in planta was impaired (Figure S3 in Text S1) suggesting mutation of regulatory elements outside of the gene sequence.
In 2000, one avrLm7 isolate was collected during a large-scale survey of French L. maculans populations [22]. In this isolate, the virulent phenotype was due to the insertion of a complete LTR of RLC-Pholy at base 6 of the coding sequence, resulting in the production of a 30 amino acid protein mainly corresponding to part of the TE (data not shown). This event was not identified in the 2006–2008 sampling. Major genome rearrangements leading to complete or partial gene deletion and RIP mutations were the two main events responsible for virulence toward Rlm7. However, our survey indicated a contrasting sequence of events (Figure 5). The proportion of the population affected by minor events (SNP, single or di-nucleotide frequency, unaltered gene sequence), or major rearrangements leaving internal part of the gene unaltered remained stable over the three years (Figure 5). In contrast, a significant change in frequency was observed for the two most common events (Pearson's approximate χ2 test, with 20,000 random samplings, P = 0.022). Whereas RIP mutation was the most frequent event leading to virulence in the first year of the survey, at a time when only very few virulent isolates could be found (Table S2 in Text S1), complete gene deletion became more common in years 2 and 3, while the frequency of RIP mutations decreased in the second year and then remained stable at ca. 20% (Figure 5). When analysing the intensity of RIP mutations, there was no significant increase in the mean number of mutated sites per isolate as a function of the year of isolation (Kruskal-Wallis test; P = 0.371, Figure 6) with for example, the same proportion of alleles with the least number of RIP mutations found in autumn 2006 as in summer 2008 (Figure 2, Figure 6), or in contrast, alleles with the greatest number of RIP mutations found in autumn 2006 (isolate G06-436 in Figure 2B).
Effective population size was firstly estimated using Approximate Bayesian Computation (ABC) methods from the minisatellite polymorphism data obtained for the population sampled in Grignon on the susceptible cv. Campala. Effective population size (Ne) estimated at the field scale was 11,500 (CI95% 3,490–29,900).
To investigate the origin of avrLm7 isolates and how their rapid increase in frequency influenced the genetic structure of L. maculans populations, a subset of 161 avrLm7 and 161 AvrLm7 isolates collected from the field experiment were genotyped using seven minisatellite (MS) markers located on different chromosomes. All seven markers were polymorphic and a total of 80 alleles were found in the collection. The average number of alleles over loci ranged between 6.86 and 8.86 (Table 4) and revealed no significant difference in allelic variability between virulent and avirulent isolates sampled (Kruskal-Wallis test; P = 0.95). Over the three years, the average gene diversity was similar amongst avirulent and virulent isolates (two-sided permutation test, 15,000 permutations, P = 0.39). The AvrLm7 and avrLm7 isolates collected between autumn 2006 and summer 2008 showed very considerable genotypic diversity and the number of genotypes was identical or very close to the number of isolates (Table 4). Overall, a total of 313 multilocus genotypes (MLG) were differentiated, of which six were shared by two to four isolates and 307 isolates (95.3%) had unique genotypes. Taking into account mating type alleles, AvrLm4-7 alleles and one additional MS marker located 70 kb away from AvrLm4-7, the isolates with identical MLGs could be differentiated and shown to be unique genotypes (data not shown).
Both mating types were present in all isolate samples and occurred in equal frequencies for most of them, except for the sample comprising avrLm7 isolates collected in autumn 2006, in which a significant deviation (P = 0.012) from a 1∶1 ratio was detected (Table 4). This biased frequency may be attributable to a resampling bias or a bias linked to the scarcity of virulent isolates at the beginning of the experiment. The multilocus linkage disequilibrium values (rd) obtained for all the samples were close to zero and did not deviate significantly from the expectations under the null hypothesis of random mating in all samples (Table 4). These results suggest that there was no genotypic disequilibrium in the samples studied and that recombination regularly generates new AvrLm7 and avrLm7 genotypes in the population.
To estimate differentiation between AvrLm7 and avrLm7 isolate samples, we firstly tested each pair of samples for heterogeneity in allele frequencies using the Fisher exact test (data not shown). These estimates were consistent with the hypothesis that there was no genetic differentiation between the samples. F-statistics also showed no significant population differentiation between them for all loci and overall. Accordingly, the mean FST was not significantly different from zero (P = 0.51) over loci between AvrLm7 and avrLm7 samples. Pair-wise levels of genetic differentiation estimated between all pairs of AvrLm7 and avrLm7 samples gave FST-estimates which were not significantly different from zero (data not shown). Lastly, hierarchical AMOVA on all samples confirmed the absence of genetic differentiation between AvrLm7 and avrLm7 isolates and showed that only 0.04% of the total variance was distributed among populations, with 99.96% within populations (ΦST = 0.0004 ; P = 0.14).
Rapid adaptation of microbes to control methods (drugs such as antibiotics and fungicides or plant disease resistance) is a very common phenomenon driven by mutation and selection, along with reproduction regime and gene flow that amplify and disperse the new character in the pathogen population [3]. Resistance to drugs can be ascribed to various mechanisms (reduced permeability or enhanced efflux, enzymatic inactivation, alteration or over-expression of the target gene) [26]. In contrast, in simpler gene-for-gene systems representative of numerous plant-pathogen interactions, modification of a target (i.e. the avirulence gene product) allows the pathogen to escape the resistance gene-mediated plant defense responses [27, this study]. Since avirulence gene products are pathogen effectors, how easy it is for the pathogen to modify or delete the target gene depends on the fitness deficit linked with loss or attenuation of its effector function [4], [28], [29]. In addition, the ability of a pathogen to render host resistance ineffective is a function of biological traits that contribute to its “evolutionary potential”, including its reproduction regime, size of populations and dispersal ability [1], [3]. Using a fungal pathogen known to have a high evolutionary potential and a dedicated field experiment, we investigated the “breakdown” of the new resistance gene Rlm7, corresponding to the AvrLm4-7 effector which makes an important contribution to fungal fitness [28], [29]. The establishment of this experiment aimed to address two questions for which little or no information is currently available: (i) what are the initial mutations responsible for rendering ineffective a plant resistance gene at the scale of a single field? (ii) how (and how rapidly) are these mutations generated?
Our data suggest that we have captured all of the possible mutational events existing very early in the process of selection and show that adaptation to selection occurs rapidly through numerous diverse mutational events at the AvrLm4-7 locus. Almost all mutations lead to gene inactivation or production of a non-functional effector protein. Unexpectedly, in a single 0.25-hectare field we observed all previously reported molecular events (and more) leading to loss of fungal avirulence in world-wide collections of isolates. Most of these mutational events were even observed during the first year of the experiment. These included complete or partial deletion of the gene [10], [14], [27], [30]–[33], amino acid substitutions [14], [27], [34], point deletions and production of truncated proteins [27], and “insertion” of a transposon [33], [35], [36]. In addition, the RIP mutations that commonly occurred, have not been previously reported as an inactivation mechanism for effector genes for pathogens other than L. maculans [10], [31]. Three other new phenomena observed were common deletion of an AA dinucleotide, alteration of the gene expression and gene duplication possibly favouring (or responsible for?) RIP mutations.
The speed of “generation” and diversity of mutational events and increased ratio of virulent isolates in the population then raises questions about how these events were generated and dispersed. The first postulate was that strong selection in a large local population would have allowed the emergence of numerous mutational variants. The estimate of Ne obtained here using ABC approaches indeed indicated large effective population size in the field and was comparable to what is described for the few sexually reproducing phytopathogenic fungi for which similar analyses were performed [37], [38]. We then investigated the sequence of mutation events found at the AvrLm4-7 locus following selection and showed that two of the many possible types of mutations were favored over others and varied in frequency over time. RIP was the prevalent mutation pattern at the beginning of the sampling at a time when only few mutants could be found, then followed by large-scale deletions. RIP was initially described in the model ascomycete fungus Neurospora crassa as a premeiotic process that efficiently detects and mutates duplicated sequences [39]. In L. maculans, the embedding of effector genes in mosaics of RIP-altered TEs and the presence of RIP signatures in the sequence of effector genes indicated that RIP could act on unduplicated sequences to promote gene diversification and that it could “leak” from the neighbouring RIP-affected sequences to generate mutations in single-copy genes [9]. Consistently, the 3′ part of the gene, directly bordered by TEs, is more affected by RIP than its 5′ part and promoter. This hypothesis, however, has to be reconsidered in view of our finding that part of the isolates with RIPped alleles of AvrLm4-7 probably has two copies of the gene. This might be more consistent with a canonical RIP mechanism, indicating that, at least in part of the cases, gene duplication precedes the action of RIP, thus acting on truly duplicated sequences and may be followed by segregation and deletion of one (or two) copies of the gene. In contrast with this finding, inactivation of the avirulence gene AvrLm6 by RIP mutations [10], [11] was not associated with duplication of the gene in field isolates, thus substantiating the ‘leaking from neighbouring RIPped regions’ hypothesis [11]. Either leaking from neighbouring RIPped TEs or acting on truly duplicated sequences, RIP is an extraordinary efficient mutation mechanism that affects up to 30% of the G:C pairs of duplicated genes in a single sexual cycle of N. crassa [39]. This suggests that RIP mutations can be generated at a very high rate at each sexual cycle of L. maculans in the field, i.e. at the beginning of each growing season. In addition, frequency of RIP mutation is increased by the embedding of effector genes in TE-rich blocks of the genome, allowing action of RIP on single-copy genes. Both these data substantiate the importance of genome environment and sexual reproduction to promote an accelerated mutation rate of effector genes (including AvrLm4-7) due to RIP [9], [11], which is likely to correspond to the most rapid adaptation to selection. In the second and third years of sampling, large-scale deletions became more common than RIP mutations as an inactivation mechanism illustrating a dynamic process in which many possible virulence alleles are generated, but only a small number eventually survive. While some large-scale deletions may in fact correspond to extremely RIPped alleles as found here for isolate NzT-4, this finding is reminiscent of what we observed when analysing the avrLm1 locus in French populations of the fungus many years after the large-scale use of the Rlm1 resistance in the field: more than 90% of the virulent isolates had a 260-kb deletion of the gene and its TE-rich environment while only 0.7% of the isolates had an allele with RIP signatures [31]. In L. maculans, the presence of four widely expanded TE families representing 25.2% of the L. maculans genome [9] provides many targets for mis-pairing between sister chromatids during meiosis and suggests that unequal crossovers lead to production of large deletion/insertion events and production of chromosomes of novel sizes in the progeny [40], [41]. Consistent with clustering of these four families of TEs in AT-isochores containing all currently known avirulence genes of L. maculans, large-size deletions were described as the main event leading to virulence at the AvrLm1 and AvrLm6 loci [10], [31], as for the AvrLm4-7 locus (this study).
Analysis of surrounding populations obtained from susceptible rlm7 cvs. indicated an increase in frequency of virulent avrLm7 isolates next to the resistant Rlm7 plot, but not in plots located less than 600 meters away, reflecting the rapid selection of virulent isolates expected in such experimental conditions. The cropping practices used in the field experiment are likely to have two consequences: (i) a large increase in size of the local population due to the lack of rotation and the close contact of the crop with unburied infected residues from the previous years and (ii) an increased rate of sexual reproduction because infected debris were left on the soil surface. RIP mutations, gene duplications and gene deletions, the most common modes of loss of the avirulence at the AvrLm7 locus directly depend upon the ability of the pathogen to undergo sexual mating. Favouring this part of its life cycle along with increase in population size directly impact the ability to generate a large number of virulent progeny and the probability they will be selected for by the resistant cultivars, eventually improving the opportunities for mating between two virulent parent isolates. The (i) wide diversity of RIPped alleles and the lack of increase in the mean number of mutations per allele during the 3 years of the experiment, (ii) the fact that virulent isolates could not be detected in local populations and (iii) the genetic similarities between the virulent and avirulent populations, indicating that virulent and avirulent isolates are part of the same genetic population in which the virulence allele is independently assorting with respect to all of the other genes, all suggest that at least part of the mutations at the AvrLm4-7 locus selected in the experimental field are generated locally within a short-time period and as a result of the large population size and meiotic recombination.
The field experiment was established at Grignon, France (48° 50′ 28.40″ N latitude; 1° 56′ 13.83″ E longitude) (Figure S1A in Text S1). This location had been used in previous studies to describe L. maculans population race structure [22]. The field was a right-angled triangle with a 50 m long base and a 100 m long side. The experiment was started in autumn 2004 and was cropped for four growing seasons (2004–2005 to 2007–2008) as a monoculture of oilseed rape with minimum tillage (chiselling). In normal agronomic practice (e.g. at Grignon before the start of the field experiment), oilseed rape is grown as part of a farm rotation and rarely returns to the same field more than one year in three (Figure S1B in Text S1). Monoculture of oilseed rape without ploughing to bury infected stubble was chosen to increase the amount of annual sexual reproduction and the local population size of L. maculans, partly mimicking minimum tillage practices in which infected stubble are left at soil surface. No fungicides were used. The plot was sown with cultivars with Rlm7 resistance (Roxet in 2004–2005, and Exagone in 2005 to 2008) and was bordered by a 10-meter wide strip cropped with a cultivar without Rlm7 (Campala) that was used as a trap cultivar [22] (Figure S1B in Text S1). No Rlm7 cv. had been grown in Grignon fields before the start of the experiment. For comparison purposes, control plots were also assessed for disease severity and occurrence of virulent avrLm7 isolates of L. maculans. These control plots included a series of plots at Grignon cropped with a susceptible cultivar between 2000 and the start of sampling of the experiment (autumn 2006) (Figure S1B in Text S1, Table S1 in Text S1). Other control plots were cropped at Versailles (48° 48′ 27.59″ N latitude; 2° 5′ 12.30″ E longitude), ca. 12 km away from Grignon (Figure S1A in Text S1, Tables S1 and S2 in Text S1) with susceptible cultivars (2000–2007) or with the Rlm7 cultivar Exagone in autumn 2006 and 2007. These control fields were cropped with the usual agronomical practices of French farmers with ploughing and rotation.
Data collected included severity of stem canker, evaluated at crop maturity using the G2 disease index [42] on 160 randomly chosen plants per cultivar.
Due to the life cycle of L. maculans in which ascospores are the origin of leaf lesions in autumn, which in turn initiate the systemic colonisation of plants eventually causing the stem canker in the following summer, populations collected from stem canker in the summer of year n, following meiosis, were considered similar to those collected from leaf lesions in the autumn of year n. Isolates were sampled either from leaf lesions (single pycnidial isolates) or from stem cankers (single-ascospore isolates) using methods described by Balesdent et al. [22] and West et al. [43], respectively, during three continuous years corresponding to two cultural cycles (2006–2007; 2007–2008), but to three generations of the fungus (Figure S2 in Text S1).
The number of leaves with lesions collected on the Rlm7 cultivar ranged between 20 and 200. Typical leaf spot lesions caused by L. maculans on the Rlm7 genotype were rare at the start of the experiment due to the scarcity or absence of virulent isolates (Table S2 in Text S1). Only 24 leaves with phoma leaf spots were found in the whole Grignon plot of Exagone in autumn 2006 and all of them were sampled, with sometimes more than one isolate per leaf sampled (Table 1, Table S2 in Text S1). When more leaf lesions were present (subsequent growing seasons for Exagone, all years for Campala), infected leaves were randomly collected with only one isolate obtained from each individual plant. As the frequency of virulent (avrLm7) isolates on Exagone was expected to be small in the first sampling year, 500 distinct leaf lesions were collected from cv. Campala in autumn 2006 (Table S2 in Text S1), so that a frequency ≥0.5% of virulent isolates could be detected with a 95% confidence interval. This number was then decreased to 200 leaves for the second year of the experiment (autumn 2007). Similarly, 100 to 200 infected stems were collected each year from each cultivar. Induction of pseudothecial maturation and isolation of ascospores from infected stems was as described by West et al. [43]. Due to the scarcity of mature pseudothecia, more than one (average three) ejected ascospores were used for isolation from a single stem. Of these, only three isolates or less per stem were analysed using molecular markers (see below).
The frequency of virulent avrLm7 isolates before the establishment of the field experiment and in the vicinity of the field experiment during the course of the experiment was estimated using a collection of 1233 isolates recovered from leaf lesions or from ascospores on stems of susceptible cultivars including around 100 isolates collected in Grignon in autumn 2006 from two oilseed rape fields located less than 600 m from the field experiment (Table 1, Tables S1 and S2 in Text S1).
The reference isolates v23.1.3 (AvrLm4-AvrLm7; double avirulent) whose genome sequence is available [9], v23.1.2 (avrLm4-AvrLm7; avirulent towards Rlm7 only), and Nz-T4 (avrLm4-avrLm7; double virulent) [44], [45] were used as controls for inoculation tests, and as sequence reference for the AvrLm4-AvrLm7 and avrLm4-AvrLm7 alleles of the AvrLm4-7 gene [21]. Isolate M3.2, the first avrLm7 isolate collected in France in 2000 was also included to determine the type of mutation it harbours at the AvrLm4-7 allele [22]. All fungal cultures were maintained on V8-juice agar and conidia were collected from 12–15 day-old cultures according to the procedure described by Ansan-Melayah et al. [46]. AvrLm4 and AvrLm7 avirulence/virulence phenotypes were determined following inoculation of 15-days-old B. napus cotyledons with 10 µL of 107 mL−1 conidia suspension as described by Balesdent et al. [47].
Genomic DNA was extracted from conidia suspensions using the DNeasy 96 Plant Kit and the QIAGEN BioRobot 3000 in accordance with the manufacturer's recommendations. PCR primers used for mating type amplification, AvrLm4-7 analyses and minisatellite analyses were designed with PRIMER 3 [48] and are described in Supplementary Table S4. AvrLm4-7 was amplified using (i) “external” primers spanning part of the promoter region and part of the 3′ UTR and generating a 1434 bp fragment and (ii) “internal” primers located within the coding sequence, between the ATG and intron and spanning 478 bp (Figure 3A). Standard PCR were performed in an Eppendorf Mastercycler EP Gradient thermocycler (Eppendorf, Le Pecq, France), with 30 cycles of 94°C for 30 s, 30 s of hybridization with variable hybridization temperatures, 72°C for 80 s, with a final extension at 72°C of variable duration (Table S4 in Text S1). Sequencing was performed on PCR products using a Beckman Coulter CEQ 8000 automated sequencer (Beckman Coulter, Fullerton, CA, USA) according to the manufacturer's instructions. Primers for sequencing (Figure 3A) were chosen so that all bases of the gene, including the 5′ UTR and 128 bp of the promoter region were independently read two or three times. Sequences were compared following sequence alignment using MULTALIN and CLUSTALX [49], [50]. Automated analysis of RIP in AvrLm4-7 alleles was done using RIPCAL (http://www.sourceforge.net/projects/ripcal), a software tool that computes RIP indexes and performs alignment-based analyses [25].
For high-throughput identification of AvrLm4-7 allelic variants before allele sequencing (when relevant), High Resolution Melting PCR (HRM PCR) was used as an alternative to sequencing, using qPCR 7500 Fast Real-Time PCR equipment (Applied Biosystems). A set of control isolates with known AvrLm4-7 sequences was included in each HRM-PCR run. Melting curves were analysed with High Resolution Melting software v2.0 (Applied Biosystems).
To recover DNA sequences flanking AvrLm4-7, Thermal Asymmetric Interlaced PCR (TAIL-PCR) was used, following the design of nested AvrLm4-7 sequence-specific primers (Table S4 in Text S1). Arbitrary Degenerated (AD) primers used in association with AvrLm4-7 primers were AD1, AD2 and AD3 [51] (Table S4 in Text S1), with AD2 used for the two subsequent rounds of PCR amplification. First and second rounds of TAIL PCR were done as described by Liu & Whittier [51]. Secondary TAIL-PCR products were purified using the Nucleospin Extract II purification Kit (Macherey-Nagel, Hoerd, Fr) and were used either for the third round of TAIL PCR, whenever the amount of amplified product was insufficient, or as template for DNA sequencing using the specific tertiary border primer Tail-GD3 as a sequencing primer (Figure 3A). To validate the TAIL-PCR results, primers were designed with PRIMER 3 from the TAIL-PCR sequence product and used to PCR-amplify the corresponding sequence from genomic DNA of the corresponding isolate, with v23.1.3 as a negative control.
For Southern blots, mycelia from the isolates grown in liquid Fries medium for two weeks were harvested by filtration, freeze-dried, ground to a fine powder, and DNA extraction was done as described by Balesdent et al. [52]. Southern blot analysis was performed on XbaI or SpeI or HpaI restricted genomic DNA (10 µg), size-fractionated on 0.8% w/v agarose gels and transferred to positively charged nylon membrane (Qbiogen) according to standard protocols. A 459-bp probe was generated by PCR using the AvrLm4-7Int-F and AvrLm4-7ext-R primers (Table S4 in Text S1) and gel purified after electrophoresis using the NucleoSpin Plasmid QuickPure kit (Machery-Nagel). Preparation of a [α-32P]dCTP probe was performed using the random priming Ready-To-Go DNA labelling beads kit (GE Healthcare). High stringency hybridization (65°C) was done using standard protocols.
For Quantitative RT-PCR, cotyledons of cv. Westar (susceptible control) were inoculated and sampled 7 days after inoculation. Total RNA extraction and single-strand cDNA synthesis were performed as described by Fudal et al. [19]. Inoculation and RNA extraction were repeated twice. Water and RNA from plants inoculated with isolate G07-E441, lacking the AvrLm4-7 gene, were used as negative controls. Primers for AvrLm4-7 amplification were as described by Parlange et al. [21]. qRT-PCR was performed using 7700 real-time PCR equipment (Applied Biosystems, Foster City, CA, USA) and ABsolute SYBR Green ROX dUTP Mix (ABgene, Courtaboeuf, France), as described by Fudal et al. [19]. Actin was used as a constitutive reference gene.
A multiplex PCR was used to characterize the distribution of the Mat1-1 and Mat1-2 alleles in the collection of isolates [53]. Seven genetically independent minisatellite markers (Table S4 in Text S1) were used for population genetic analyses [54]. For each isolate, the allele sizes were determined using quantity one 1-D Analysis software (BioRad, Marnes-la-Coquette, France) by comparison with band sizes of the 1-kb+ ladder (Invitrogen, Cergy Pontoise, France) and internal control with known allele size and known number of repeats of the core motif (i.e., the sequenced reference v23.1.3 isolate). Data were scored as the number of repeat units for each isolate and each minisatellite locus.
ABC methods implemented in the DIYABC program v1.0.4.39 [55] were used to estimate the effective population size (Ne) of the L. maculans populations sampled in 2006, 2007 and 2008 from the susceptible cv. Campala. For the genetic parameters, the Generalized Stepwise Mutation (GSM) model was used to simulate mutations at the minisatellite loci and the prior interval specifications for the mean mutation rate were as described in Dilmaghani et al. [54]. A total of 1,000,000 data sets were simulated to generate a reference table. This reference table comprised summary statistics (e.g. genetic diversity per sample and genetic distance between samples) that enable estimation of the posterior distributions of the demographic parameters, under a given scenario, using comparisons between simulated and observed data sets. We used DIYABC to estimate effective population size under a simple scenario corresponding to one population, from which several samples had been taken over three consecutive generations (three events of sexual reproduction). The summary statistics used were mean number of alleles per locus, mean genetic diversity [56], mean variance in allele size, genetic differentiation between pairwise groups (FST, [57]) and genetic distances (σμ)2 [58]. A local linear regression on the 1% simulated data sets closest to the observed data sets was then used to estimate the posterior distribution of the parameters. A generation time of 1 year was assumed, based on biological and epidemiological studies [24], [59].
The frequency of each molecular event leading to virulence (avrLm7) was calculated in the different isolate samples. Where more than one isolate was recovered from a single plant, as with isolates obtained from stem residues, a sub-sample was obtained by random selection of a single isolate from each individual plant. The molecular event for each isolate sampled was then recorded and the frequency of each molecular event was calculated on the basis of this sub-sampling. This randomized sub-sampling was repeated 10,000 times, and the mean frequency for each molecular event calculated. The resulting frequencies of each type of event were compared over the three years using the Approximate Pearson's Chi squared test with 20,000 randomizations as implemented in XLSTAT v2010.5.01.
To analyse minisatellite variability, the software FSTAT version 2.9 [60] was used to compute allele frequencies, number of alleles per minisatellite (A), number of private alleles and Nei's gene diversity (H) [61] at each locus and over all loci, within and over the samples. Tests for differences between groups of samples comprising AvrLm7 or avrLm7 isolates for polymorphism statistics were based on two-sided permutation tests (15,000 permutations) and performed using FSTAT.
Linkage disequilibrium was evaluated using two different approaches. First, minisatellites were tested pairwise within and across samples using the genotypic disequilibrium test in Genepop [62]. The statistical significance of each pairwise test of linkage disequilibrium was tested by Fisher's exact test. The associations of alleles among different loci were also estimated with the standardized version of the index of association rd, using MULTILOCUS [63]. The significance of rd was established by comparing the observed value to the distribution obtained from 1000 randomizations with alleles at each locus being resampled without replacement to simulate the effect of random mating. The hypothesis of random mating was tested as follows: the distribution of mating types was compared to the 1∶1 ratio expected under random mating for a haploid fungus using χ2 tests. Genetic structure was analysed with standard FST coefficients of population differentiation, which were calculated and tested for significance using 1000 permutations using FSTAT. To further analyse population differentiation, heterogeneity of allele frequencies among samples was tested for each locus using the Fisher exact test in the GENEPOP program [63]. Genetic differentiation amongst samples was examined using an analysis of molecular variance (AMOVA) in Arlequin v3.5 [64].
|
10.1371/journal.pntd.0006355 | Common occurrence of Cryptosporidium hominis in asymptomatic and symptomatic calves in France | Cryptosporidium spp. infections are the most frequent parasitic cause of diarrhea in humans and cattle. However, asymptomatic cases are less often documented than symptomatic cases or cases with experimentally infected animals. Cryptosporidium (C.) hominis infection accounts for the majority of pediatric cases in several countries, while C. parvum is a major cause of diarrhea in neonatal calves. In cattle Cryptosporidium spp. infection can be caused by C. parvum, C. bovis, C.andersoni and C. ryanae, and recently, reports of cattle cases of C. hominis cryptosporidiosis cases suggest that the presence of C. hominis in calves was previously underestimated.
From February to November 2015, Cryptosporidium spp. infected calves were detected in 29/44 randomly included farms from 5 geographic regions of France. C. hominis and C. parvum were found in 12/44 and 26/44 farms, respectively with higher C. hominis prevalence in the western region. In 9 farms, both C. parvum and C. hominis were detected. Eighty-six of 412 (73/342 asymptomatic and 13/70 symptomatic) one to nine-week-old calves shed C. hominis or C. parvum oocysts (15 and 71 calves, respectively), with no mixed infection detected. The predominant C. hominis IbA9G3 genotype was present in all regions, and more frequent in the western region. An incompletely characterized Ib, and the IbA13G3, IbA9G2 and IbA14G2 genotypes were present only in the western region. For C. parvum, the most frequent genotype was IIaA16G3R1 with no geographic clustering. Most C. hominis infected calves were asymptomatic, with some exceptions of IbA9G2 and IbA9G3 isolates, while C. parvum IIaA16G3R1 was associated with symptoms.
Present results indicate for the first time that in several geographic regions of France, C. hominis was present in about one fifth of both asymptomatic and symptomatic infected calves, with isolated genotypes likely associated with human infection. Further investigations are aimed at documenting direct or indirect transmissions between livestock and humans.
| Symptomatic infection by the Apicomplexan Cryptosporidium spp. is presently considered the most frequent parasitic cause of acute diarrhea in both humans (especially severe in immunocompromised individuals and infants in both developed and developing countries) and cattle (calves), while asymptomatic infections are less often documented. Cryptosporidium (C.) hominis once considered to be restricted to humans accounts for the majority of pediatric cases in several countries. C. parvum can also infect cattle as well as C. bovis, C. andersoni, and C. ryanae. Recently, cattle C. hominis cryptosporidiosis has been reported, suggesting that the presence of C. hominis in calves was previously underestimated. The aim of this work was to characterize Cryptosporidium spp. infection in both asymptomatic and symptomatic dairy and beef calves from Metropolitan France. From February to November 2015, C. parvum or C. hominis infected calves were detected in farms from 5 geographic regions of France. Surprisingly, C. hominis was present in about one fifth of Cryptosporidium spp. infected calves, and exhibited genotypes which were previously reported in human and nonhuman primate. Further investigations are aimed at documenting direct or indirect C. hominis transmissions between and among livestock and humans.
| Cryptosporidium spp. are Apicomplexa which include parasite species causing asymptomatic to severe gastrointestinal infections in a wide range of vertebrate hosts, and exhibiting varying degrees of host adaptation [1]. Previously, information on cryptosporidial host restriction of natural cryptosporidial infection was usually obtained from animal and human cases for which clinical symptoms, ages and immune statuses were recorded. However, evidence of asymptomatic sustained or transitory infection, and the role of additional parameters such as parasite detection methods, host’s genetic background, co-infection and environmental factors such as climate, seasons and socioeconomic status were less documented [2]. For some cryptosporidial host specialisations, information is presently limited to experimentally infected immunocompetent or immunosuppressed laboratory animals [3].
In humans, cryptosporidiosis is presently identified as the most frequent zoonotic cause of parasitic diarrhea, especially severe in immunocompromised individuals and infants in both developed and developing countries [4–7]. In addition to 8 others sporadically observed species, C. hominis, once considered to be restricted to humans, and C. parvum, of which some isolate genotypes also infect ruminants, account for more than 90% of reported human cases worldwide [8–10]. There is equal or higher prevalence of C. hominis than C. parvum in humans in many parts of the world except in Europe where C. parvum largely prevails, likely reflecting the ratios of human to animal sources of anthroponotic C. hominis and anthropozoonotic C. parvum contamination, respectively [11–15].
In cattle, the main symptom of cryptosporidiosis is watery and profuse acute diarrhea which can be associated with dehydration, anorexia, and impaired growth [16]. It was previously established that cattle can be infected by at least 4 Cryptosporidium species, i.e. C. parvum, C. bovis, C. andersoni, and C ryanae [17, 18]. In France, C. parvum, C. bovis and C. andersoni predominate in newborn and older calves, respectively, C. parvum infection is identified as a major cause of diarrhea in newborn calves of less than one month old, with economically significant morbidity and mortality. However, detailed epidemiology on the occurrence of viable oocysts from normal feces of asymptomatic calves is unknown [19, 20]. Recently, a limited number of observations reported of cattle cryptosporidiosis due to C. hominis have been reported in Australasia, Asia, Africa and Europe, suggesting that the presence of C. hominis in calves was previously underestimated in studies on diarrheic and adult animals [21–27].
The aim of this work was to document the prevalence of Cryptosporidium spp oocysts in calves from five different geographic regions of Metropolitan France. Farms were randomly included in the study, the clinical status of each animal was recorded, and the presence of calves with Cryptosporidium spp. oocysts in feces was investigated. Isolates were genetically characterized for their synzootic and zoonotic potentials.
From February to November 2015, 412 calves aged from 1 to 9 weeks were selected in 44 farms from a national list of farms under regular veterinarian survey (16 veterinary offices, from 1 to 4 farm(s) per office). Farms were randomly selected, and within farms, calves aged from 1 to 9 weeks were randomly selected. Selected farms were situated in 14 “départements” (an administrative sub-region) distributed in 5 geographic regions of Metropolitan France: western (Côtes d'Armor, Ille-et-Vilaine, Morbihan), central western (Vendée, Deux Sèvres, Mayenne), northeastern (Pas-de- Calais, Moselle), southwestern (Landes, Pyrénées atlantiques, Tarn, Hautes-Pyrénées), and central (Puy-de-Dôme, Allier). For each calf, the clinical status was evaluated and recorded by all veterinarians at the time of sampling as follows: presence or absence of digestive symptoms such as diarrhea and abdominal bloating and/or respiratory symptoms, and evaluation of the general condition as follows: "Normal general condition: shiny hair coat, regular appetite; impaired general condition: delayed growth, dull hair coat, capricious appetite; Poor general condition: dull hair coat, capricious appetite, marked growth retardation."
From each farm (housing from 11 to 50 calves), feces samples were obtained by veterinarians from 5–10 calves by rectal stimulation. Most farms were dairy farms and breeds consisted of Salers, Holstein, Charolais, Montbeliard, Blonde d'Aquitaine, Parthenaise, Limousine and the Belgian Blue Breed (BBB). In all farms, calves were kept in semi-intensive farming systems and separated from their dams.
The presence of Cryptosporidium spp. oocysts was microscopically determined by the same experienced clinical parasitologists using Bailenger type feces concentration method [28] and Heine staining [29]. The presence of other intestinal parasites (Giardia, Strongyloides, and coccidia) detected in some of the calves using various methodologies was not considered in the present study.
All samples were subjected to molecular analysis for speciation and genotyping of speciation positive samples. Before DNA isolation, feces were subjected to a pre-treatment with a mechanical lysis in Lysing Matrix A Tubes (garnet matrix and ¼ ceramic spheres) (Qiagen, CA, USA) with the Fastprep-24 device and transferred into 2 ml Eppendorf tube prior to thermal shock lysis (6 freeze-thaw cycles). Samples were placed in an ultrasonic bath for sonication (3x20 sec bursts). In accordance with the manufacturer's instructions, a modified QIAamp Stool Mini Kit (Qiagen, CA, USA) was used to isolate DNA from the pre-treated samples. All centrifugation steps were performed at RT (20–25°C), at 14.000 rpm. Eight hundred μL/tube of ASL buffer was added, and tubes were heated at 99°C for 15 min.
For speciation, a 18S rRNA gene sequence was amplified using a nested PCR and restriction digestion of the secondary product with SspI (NEB, MA, USA) and VspI (NEB, MA, USA) was performed [30]. Briefly, for the primary PCR step, a PCR product (about 1,325 bp long) was amplified by using primers 5-TTCTAGAGCTAATACATGCG-3 and 5-CCCTAATCCTTCGAAACAGGA-3. For the secondary PCR step, by using 5 μl of the primary PCR product and primers 5-GGAAGGGTTGTATTTATTAGATAAAG-3’ and 5-AAGGAGTAAGGAACAACCTCCA-3 a PCR product (819 to 825 bp long, depending on the species) was amplified.
Each PCR mixture (total volume, 50 μl) contained 5 μl of 10X DreamTaq Buffer, each deoxynucleoside triphosphate at a concentration of 0.2mM, each primer at a concentration of 100 nM, 2.5 U of DreamTaq polymerase, and 5μL of DNA template. Then, 1.25μL of DMSO (100%) was added to the mixture.
A total of 40 cycles, each consisting of 94°C for 45 s, 55°C for 45 s, and 72°C for 1 min, were performed. An initial hot start at 94°C for 3 min and a final extension step at 72°C for 7 min were also included. Each amplification run included a negative control (PCR water) and two positive controls (genomic DNA from C. parvum oocysts purchased from Waterborne Inc., and C. hominis genomic DNA from fecal specimen collected at Rouen University Hospital). Products were visualized in 2% agarose gels using ethidium bromide staining and identification was confirmed by sequencing. Positive samples were further genotyped by DNA sequencing of the gp60 gene amplified by a nested PCR following the protocol described by Sulaiman et al. [31] (2005). All Amplification experiments were repeated at least thrice to check reproducibility.
Purified PCR products were sequenced in both directions on an ABI 3500 sequencer analyzer (Applied Biosystems, CA, USA) by using the secondary PCR primers and the BigDye Terminator v3.1 Cycle Sequencing Kit (Applied Biosystems, CA, USA). The obtained sequences were inspected using the 4 peaks software (https://nucleobytes.com/4peaks/index.html), edited with the BioEdit sequence alignment editor (version 7.2.5), and analyzed for DNA database search and comparisons using the BLAST server (www.ncbi.nlm.nih.gov/BLAST). Genotypes were named using the established gp60 genotype nomenclature [31].
Statistical analyses were performed using the chi-square (χ2) test or Fisher’s exact test as appropriate using the Number Cruncher Statistical System (NCSS), version 2000 to determine the association between the prevalence of Cryptosporidium infection vs regions, and sampling periods. A p value <0.05 was considered statistically significant.
Before carrying out this work, informed written authorization to perform and anonymously publish the present epidemiological study was obtained from all cattle owners and veterinarians. Clinical examination of calves and stool harvest were part of routine breeding and veterinary procedures, without any invasive, traumatic or specific containment method. Such procedures are not qualified as animal experimentation involving vertebrate according to French laws, and no specific ethical clearing was required.
As shown in Table 1, infected calves were detected in 29/44 farms from all geographic regions and “départements” except Puy-de-Dôme and Mayenne, with no inter-regional difference in the ratios of the number of infected farms to the number of included farms. C. hominis and C. parvum were found in calves from 12/44 and 26/44 farms, respectively. In 9/44 farms, both C. parvum and C. hominis infected calves were found with no mixed infection in any animal. No mixed infection of C. hominis and C. parvum was noted in any animal.
Eighty-six of 412 included calves exhibited Cryptosporidium spp. oocysts in feces, of which 15 and 71 had C. hominis and C. parvum infection respectively. There were no inter-regional differences in the ratio of the number of infected calves to the number of included calves (p = 0.839). In the western region, the ratio of C. hominis infected farms (8/13) was higher than in all other geographic regions (p = 0.018), while no interregional difference was found for C. parvum (p = 0.122). Infections were found in calves from 2 weeks to 7 weeks of age (3 to 7 weeks and 2 to 7 weeks for C. hominis and C. parvum cases, respectively).
As shown in Table 2, C. hominis IbA9G3 genotype isolates were predominant and present in all geographic regions. The incompletely characterized Ib and the IbA13G3, Ib A9G2 and IbA14G2 genotypes were only represented in the western region.
For C. parvum, IIaA16G3R1 genotype was the most frequent with no geographic clustering (p = 0.574), and the limited number of isolates exhibiting other genotypes precluded further investigation on their geographic representation (Table 3).
No C. hominis infected calves exhibited diarrhea during the week before or at the time of stool sampling (Table 4). Two infected calves, one with genotype IbA9G2 and one with genotype IbA9G3 presented with general state alteration and non-diarrheal digestive symptoms. Ten calves infected with the most frequent C. parvum IIaA16G3R1 genotype presented with digestive symptoms, respiratory symptoms, or both, and 12 exhibited an impaired or poor general state. The limited number of observations, precluded from investigating further associations between symptoms and genotypes.
The ratios of the number of infected calves to the number of sampled calves observed during the August-September (50/216) and October-November (28/106) periods were higher than those during February to March (3/45) and April to June (5/42) periods (p = 0.010) with no difference in the respective C. hominis and C. parvum representations (the corresponding values for C. hominis were 9/50, 4/28, 2/3, and 0/5, respectively, p>0.05). No seasonal effect on Cryptosporidium infection prevalence, however, could be unambiguously established, taking into account that due to the yearly calving cycle, most calves were sampled during summer and autumn, and that no feces sample was obtained in January, July and December. The age distribution of C.parvum infected calves is similar to that of C. hominis infected calves (Fig 1)
In this work, the presence of C. hominis in calves was addressed for the first time in France by investigating both asymptomatic and symptomatic calves from randomly chosen farms in several geographic regions. Surprisingly, C. hominis was identified in about one fifth of Cryptosporidium spp. infected animals, a figure which to our knowledge has not been reported previously in Europe.
The random selection of farms resulted in a variety of geographic regions corresponding to different geologic and climatic characteristics with oceanic, or humid continental influences. Due to breeding procedures, most calves from all regions were sampled during summer and autumn. For each calf, age, clinical condition, farm and sampling date were recorded by veterinarians. The study was focused on calves, and unbiased sampling was confirmed by similar C. hominis infection rates of calves aged between 2 and 7 weeks. Most animals were clinically asymptomatic, none of them presenting diarrhea in the week before, or at the time of sampling, and a exhibiting digestive or respiratory symptoms and/or impaired or poor general condition consistent with clinical cryptosporidiosis. Sampling and preservation of samples were adapted to oocyst detection in calf feces: Heine-stained oocysts were detected microscopically by experienced parasitologists, and oocyst species and genotypes were determined using previously validated DNA amplification methods. C. hominis was unexpectedly detected in 15 calves. Four out of 15 C. hominis positive samples could not be identified at the genotype level, with unreadable superimposition of electrophoregrams which might result from amplification and sequencing of different genetic fragments from several genotypes present in the same sample. For both C. hominis and C. parvum, several subtypes were not seen in the same sample (32).
Farms from all “départements” except 2 presented both C. hominis and C. parvum-infected animals. C. hominis was present in about half of the “départements” with Cryptosporidium spp. infected calves, and C. parvum was present in all of them. Both the ratios of C. hominis infected farms and animals were higher in the western geographic region compared with other regions. In 9/44 farms, both C. parvum and C. hominis were detected in different calves. Although no data concerning the origin of water given to calves was available, a waterborne transmission of C. hominis is possible as illustrated by the recent occurrence of a C. hominis waterborne outbreak in France. All samples were obtained during the calving period; thus, no seasonal variation of Cryptosporidium species nor genotypes could be ascertained. The present unexpected high proportion of C. hominis infected calves compares with previously reported figures in New-Zealand Australia, Africa, China and Europe [21, 24, 25, 32, 33]. Several studies have established that cattle can be infected with at least 4 species (C. parvum, C. bovis, C. andersoni, and C. ryanae). C. hominis and C. parvum were the only species detected in the present study, likely due to the age of the calves, and possibly to the masking of concurrent infections by C. bovis, C. ryanae or C. andersoni by the predominant shedding of oocysts of other species, or to preferential PCR amplification of predominant species [17, 18, 35].
It was once generally accepted that C hominis primarily infects humans while C. parvum infects both human and non-human hosts, and reports of cattle C. hominis infections are few. Early observations based on 18S rDNA gene analysis revealed mixed cases with C. hominis in heifers and cows from 2001 to 2003, and C. hominis cases in domestic cattle and goats [22, 26]. More recently, limited numbers of C. hominis isolates from cattle have been genotyped in Africa, China, New Zealand, Australia, the UK and Italy [21, 23–25, 34, 36, 37]. Besides cattle, C. hominis has been reported in sheep and goats in the UK and in foals in Brazil [38, 39].
In addition to species identification, gp60 genotyping was performed to provide clues to symptomatic, epidemiologic, and zoonotic characteristics of isolates. Genotypes included IbA13G3, IbA9G2, and IbA9G3 and the newly described IbA14G2, and for 4 isolates, only attribution to the Ib genotype family was obtained, due to unreadable superimposition of electrophoregrams. No association between genotypes and symptoms could be determined. While 1bA9G3 was only detected in one human patient from the western region, no evidence for C. hominis case clustering and no seasonal variation could be established using genotypes. To our knowledge, none of the above gp60 genotype sequences (IbA13G3, IbA9G2, and IbA9G3) has been previously reported in cattle. C. hominis gp60 IbA10G2 genotype has been detected in Australia and New Zealand [21, 34]. The present genotyping data suggest the existence of potentially zoonotic C. hominis isolates in calves, since the 3 genotypes mentioned above have been previously described in human and non-human primates. Genotype IbA13G3 was reported in wastewaters from a densely populated urban region (Shanghai, China) [40], genotype IbA9G2 was reported in humans and genotype IbA9G3 was found in non-human primates in Kenya and China [41–43]. Such data aim at investigating associations between genotypes found in cattle and humans from the same geographic region, as reported in India for C. hominis genotype IdA15G1 found in one calf, and children from the same geographic region [44].
Present results indicate for the first time that in several geographic regions of France, C. hominis was present in about one fifth of both asymptomatic and symptomatic calves, and exhibited genotypes likely linked to human infection. Cattle have been considered to be a primary reservoir for Cryptosporidium spp. and to play a role in transmitting zoonotic C. parvum organisms to humans [44, 45]. Results of the present study suggest that calves in France also frequently harbor C. hominis isolates which might be cause of human infections. Further investigations are aimed at determining whether the source of cattle infections was other livestock or humans, and whether the transmission was direct or indirect.
|
10.1371/journal.pcbi.1000600 | A Multi-Component Model of the Developing Retinocollicular Pathway Incorporating Axonal and Synaptic Growth | During development, neurons extend axons to different brain areas and produce stereotypical patterns of connections. The mechanisms underlying this process have been intensively studied in the visual system, where retinal neurons form retinotopic maps in the thalamus and superior colliculus. The mechanisms active in map formation include molecular guidance cues, trophic factor release, spontaneous neural activity, spike-timing dependent plasticity (STDP), synapse creation and retraction, and axon growth, branching and retraction. To investigate how these mechanisms interact, a multi-component model of the developing retinocollicular pathway was produced based on phenomenological approximations of each of these mechanisms. Core assumptions of the model were that the probabilities of axonal branching and synaptic growth are highest where the combined influences of chemoaffinity and trophic factor cues are highest, and that activity-dependent release of trophic factors acts to stabilize synapses. Based on these behaviors, model axons produced morphologically realistic growth patterns and projected to retinotopically correct locations in the colliculus. Findings of the model include that STDP, gradient detection by axonal growth cones and lateral connectivity among collicular neurons were not necessary for refinement, and that the instructive cues for axonal growth appear to be mediated first by molecular guidance and then by neural activity. Although complex, the model appears to be insensitive to variations in how the component developmental mechanisms are implemented. Activity, molecular guidance and the growth and retraction of axons and synapses are common features of neural development, and the findings of this study may have relevance beyond organization in the retinocollicular pathway.
| Neural development is a process that involves a wide range of behaviors. As a result of these behaviors, neurons are able to extend axons to different brain areas and produce stereotypical patterns of innervation. One of the most commonly studied of these projections is in the visual system, where retinal axons project to multiple brain regions and produce retinotopic maps. This study examines the relative roles and interactions of different neural mechanisms in guiding axon growth and generating retinotopic order. We did this by producing a computational model of retinotopic development that represented many of the neural mechanisms thought to be involved, including axon and synapse growth, molecular guidance and synapse plasticity. Our results suggest that synaptic plasticity is realized by variation in the number of synapses between neurons, not through alteration of individual synaptic weights; that lateral connectivity between collicular neurons is not required for organization; and that axon arbor development does not require the gradient tracking abilities of growth cones. The mechanisms underlying neuronal development in the visual system are also observed in many other brain areas, so the findings here should apply more generally.
| During neural system development, groups of neurons project to various areas of the brain and produce stereotypical patterns of innervation. These organization patterns are an emergent property of the physiological mechanisms regulating neural behavior. In the visual system these mechanisms include molecular guidance [1],[2], spontaneous correlated activity in the form of retinal waves [3]–[5], neurotrophic factor release and uptake [6], spike-timing-dependent plasticity (STDP) [7],[8] as well as the growth and retraction of axons and synapses. Similar phenomena are observed in many other brain areas during development [9]–[13]. An important question is how these underlying phenomena combine to produce the emergent patterns of connections seen throughout the brain. A well studied example of such organization is the retinotopically ordered projection from the retina to the thalamus and superior colliculus.
Many computational models have examined how one or more of these phenomena are able to produce retinotopic organization (e.g., [14]–[21]). So far, however, none of the models has brought together this diverse set of physiological behaviors, and only a few (e.g., [22]) have addressed development from the perspective of individual axons and how they can grow, branch and retract to reach their retinotopically correct termination zones. Framing development from this perspective is important, as neural connection patterns are ultimately the result of axon growth and branching, hence constraining a model by the physical and geometrical constraints of the axon is a prerequisite to understanding how projections form. An axon extending through any neuropil consisting of cells, axons and dendrites is analogous to a rope being pulled through a corn field: once the rope is extended, lateral motion is not possible without knocking over corn stalks [23]. Similarly, an axon has very restricted lateral motion once it has extended and branched throughout the neuropil. To explain neural organization such as retinotopic development, a model needs to describe not only how these physiological behaviors contribute to development, but also how observed patterns of development can be achieved in light of the physical constraints placed on axon movement.
This study presents a model of retinocollicular development that combines phenomenological approximations of the aforementioned physiological behaviors and examines how these can guide the extension, branching and retraction of individual axons in such a way that leads to a refined arborization at the retinotopically correct location in the colliculus. The stages of development follow those previously described for mouse and chick [24]. In summary, retinal axons enter the anterior side of the colliculus and extend in a largely linear manner to the posterior side. Interstitial branches then sprout and extend towards the retinotopically correct area of the colliculus for the given axon, based on chemoaffinity compatibility between each axon and the expression of molecular markers in the colliculus (Fig. 1A). Activity-dependent trophic feedback mediates growth and retraction of individual synapses, with trophic factor stabilizing synapses that contribute to spiking activity in the postsynaptic neurons and synapses that receive insufficient trophic feedback retracting (Fig. 1B). Correlated retinal activity, in the form of retinal waves, provides spatial information allowing synapses from retinal ganglion cells (RGCs) originating from near the same point in the retina to stabilize on the same collicular neurons. Trophic factors enhance axon and synapse growth in the areas of the axon where they are received and STDP modulates the excitatory strength of individual synapses.
This study continues in the spirit of previous theoretical work on hybrid models (e.g., [25]), allowing the relative roles of and interactions between these different physiological behaviors to be studied, and has generated several new findings. Most significantly, retinotopic organization and refinement appears to be a stable emergent property of the core assumptions so long as the functional behaviors of retinal waves, molecular guidance cues and activity-dependent trophic factor release were represented in the model. The characteristic of retinal waves that was important was the overall correlational structure of activity and not the specific spatiotemporal properties of the waves. Alteration in the correlational structure by using simulated retinal activity similar to that observed in the mutant mouse [26] disrupted the ability of axons from neighboring RGCs to produce overlapping arbors in the colliculus. Neither STDP nor any form of plasticity occurring at the level of individual synapses was necessary for refinement, and analysis of the model suggests that Hebbian synaptic plasticity is a slow-acting process that is instead realized by the addition and subtraction of synapses. Gradient detection by axon growth cones was not required to achieve retinotopic organization or refinement once axons had reached the colliculus, as each axon was able to guide growth based on gradient differentials across its arbor.
The model addresses retinocollicular development over five days (120 hours) of simulated time, similar to the one week period of maturation of the retinocollicular (retinotectal) projection in mice and chicks [24],[27]. While molecular guidance cues and retinal waves are both present throughout this age, modeled development occurred in two stages, each lasting 60 hours. During the first stage of development, an axon's propensity for growth was mediated only by its chemoaffinity compatibility with surrounding tissue, while during the second stage trophic factor receipt by synapses on the axon also contributed to guide growth. Fig. 1C shows the development of a representative axon during chemoaffinity regulated growth. Fig. 1D shows the chemoaffinity-mediated axon growth from groups of neighboring retinal ganglion cells (RGCs) at five different retinal locations. A coarse topological organization is apparent.
After 60 hours, trophic factors also contributed to guiding axon growth and synapse creation. Trophic factor was released by postsynaptic terminals where a presynaptic spike preceded a postsynaptic spike within tens of milliseconds, and was taken up by the presynaptic terminal. As described later, to achieve a smooth retinotopic mapping, it was necessary to delay activity-dependent trophic feedback to axon growth until axons had produced diffuse arbors in the retinotopically correct areas of the colliculus. Fig. 1E shows the continuation of development from Fig. 1D after activity-dependent mechanisms became active. An overview of how the mechanisms of the model generate retinotopic organization and refinement is shown in Fig. 2. In summary, molecular guidance cues guide axons to near their retinotopically correct areas of the colliculus. While individual axons arbors are only loosely targeted, nearby RGCs collectively produce arbors with highest density near the retinotopically correct termination zone. Axon density corresponds with synapse density, resulting in a enhanced collicular response in the areas of higher axon density. This increased response results in increased activity dependent feedback from these collicular neurons, increasing local axon and synapse growth and resulting in an increasingly refined arbor. A movie showing RGC axon development over the full 120 hours can be downloaded as supplementary material (Video S1).
Upon the onset of activity-dependent feedback, the diffuse, chemoaffinity guided arborizations quickly refined. Fig. 3A,B shows the development of two axons at six hour intervals after activity-dependent feedback began to influence axon growth. Initial synapse distribution from any particular axon was diffuse with most arbors becoming largely refined after 24 hours simulated time. Because of the relatively long duration of individual simulations (1–6 days, realtime), analysis of the model focused on its qualitative behavior and attempts were not made to tune the model in such a way as to achieve a particular quantitative goal, such as development time or receptive field size. Quantitative measures were made to help assess qualitative behavior. The average receptive field (RF) radius (see Methods) for individual collicular neurons was (n = 7935 collicular neurons) and the average projective field (PF) radius for RGCs was (n = 7914 RGCs). To assess the continuity of the retinal projections, the RF and PF of groups of neighboring neurons were also measured (19 adjacent RGCs or collicular neurons from 7279 and 7278 non-border locations in the retina and colliculus, respectively). The RF for groups of collicular neurons was and the PF for groups of RGCs was , an increase of and over individual RF and PF, respectively. This small increase in size for groups of cells versus individual cells indicates that there was considerable overlap in the RF and PF of groups of neighboring cells. Visual analysis of the projections confirmed this interpretation.
In addition to the shape of the simulated retina and colliculus described above (Fig. 1), simulations were carried out on a reduced form of the model where a slice of the simulated retina (central 30% of D-V axis; 3023 RGCs) projected to a slice of collicular neurons (central 30% of L-M axis; 2695 collicular neurons; see Methods). The smaller model had qualitatively and quantitatively similar behavior as the full model (Fig. 4A–D, compare red and black traces). Simulation of the reduced size model was faster than the full model and analysis of the model's behavior was carried out using this smaller implementation. There was limited variability in the RF and PF radii between multiple simulations runs (n = 6). The individual RF for collicular neurons was () and the group RF was (). The average individual PF was () and the group PF was (). Because of the limited variability, and the duration of individual simulations, all quantitative measures of model behavior (e.g., parameter exploration and STDP analysis) were based on the behavior of the 3023 RGC or 2695 collicular neurons in the reduced size model, measured from a single simulation run, except as otherwise noted.
The model was analyzed to evaluate its stability and to examine the effects of modifying the physiological behaviors on which it was based. To evaluate the stability of the model, each of 16 free parameters (Table 1) were individually altered from 50% to 200% of their baseline values and the retinotopic organization and refinement of the model was analyzed (Fig. 4A–D, grey traces). In all cases, development was qualitatively similar. The PF of individual RGCs was similar to the PF of groups of neighboring RGCs, indicating a large degree of overlap and limited scatter at the local level. The ratio of the PF size between individuals and groups of neighboring RGCs was consistent for all parameter variations (Fig. 5A, grey circles). Global retinotopic order was assessed by comparing the target position of RGCs between simulations. To do this, the average PF center of each RGC was calculated over five control simulations and this average was used to produce a “normal” map. RGC projections in parameter exploration simulations were then compared to the normal map, and the deviation of each RGC projection from normal was measured. The average deviation in these simulations was approximately the distance between adjacent collicular neurons (10 ), demonstrating that global order was maintained (Fig. 5B, grey circles).
The relative contribution of the different physiological behaviors were analyzed by selectively altering or disabling them. First analyzed was the the effect of changing the spatiotemporal characteristics of retinal waves. The baseline (control) spatiotemporal properties of retinal waves were based on those described for young ferrets [28],[29], as retinal wave properties are similar between species (see [30]) and the size, velocity, frequency and RGC firing properties in ferret are well described. To assess the importance of specific spatiotemporal wave properties to retinotopic development and to see how the selection of control values biased the results, the model was run using patterns of retinal wave activity where the size, velocity or frequency of waves was altered. In all cases, the retinal projection and collicular receptive fields were quantitatively and qualitatively similar (Fig. 6A–D; compare orange to black). Baseline waves had a velocity of 180 , average size of 0.161 , and average interwave interval of 94.2 sec. Ranging the velocity from 112 to 466 , while holding other wave properties largely constant, had minimal effect on retinotopic refinement. Similarly, refinement appeared normal for waves having small (0.101 ) and large (0.428 ) average sizes. Increased wave frequency, as measured by decreasing the interwave interval to 45.1 sec, had minimal effects on refinement. Decreasing wave frequency (interwave interval 202 sec) slowed the rate of refinement but did not have a significant effect on the refined projection.
Mice lacking the subunit of the acetylcholine receptor have been reported to have significantly altered retinal activity patterns [26],[31],[32] as well as altered retinocollicular projections, with the projective and receptive fields of groups of nearby neurons larger than observed in wild type [33],[34]. In wild type mice nearby RGCs having stronger correlations in activity than RGCs farther apart, while in knockout mice () retinal activity is either uncorrelated [31] or strongly correlated over long distances, with RGCs from over a large area of the retina bursting almost simultaneously [26],[32]. In either case, the spatial information provided to refining axons is disrupted, as activity in axons from neighboring RGCs is no longer significantly more correlated than in axons from RGCs located farther apart. To explore the result of this change to retinal activity, simulated retinal activity patterns were approximated based on data reported by [26] (Fig. 6E; green). Using these patterns of RGC activity, the individual RF radius increased by 65% (control: (n = 6 simulations) compared to (n = 1 simulation); all subsequent comparisons are reported in this format) and the group RF radius was similarly increased (; from to ). The refined arbor of each individual RGC showed a minor increase in size compared to control (+14%; from to ), but the group PF radius was increased significantly (+49%; from to )(Fig. 6; A–D). These changes are qualitatively consistent with experimental findings [33],[34], but experimentally observed changes are quantitatively much different than observed here (a 2–2.5 fold increase in RF or PF area in the model compared to a 10-fold increase observed experimentally). Several factors might account for this difference. One factor is that, despite strongly enhanced activity correlations between distant neurons, simulated activity still had higher correlations for nearby neurons than for distant ones compared to experimental data [26]. Another factor is that the firing properties of retinas are not fully understood (compare [31],[33] and [26]) and it seems unlikely that activity is accurately represented here. Simulations in which all RGC firing was decorrelated prevented refinement, suggesting that actual activity is unlikely to be fully decorrelated (e.g., [31],[33]) and there likely exists some distance-dependent pattern of correlation, as indicated by [26],[32]. More generally, the results suggest that it is the correlation patterns between RGCs that drives refinement, not the particular characteristics of retinal waves, and that even small amounts of heightened correlation among nearby neurons can result in nearly normal patterns of refinement.
Molecular guidance cues were implemented as providing axons a bias to preferentially grow near the retinotopically correct area of the colliculus. Disabling this form of guidance by eliminating molecular guidance cues in both colliculus and retinal axons completely disrupted retinotopic organization (Fig. 7). Individual axons did refine, and nearby RGCs often projected to similar collicular areas and had overlapping arbors, but there was no global order in the projections. Disabling chemoaffinity after coarse retinotopic organization was established (i.e. at 60 hours), and allowing subsequent refinement to be driven exclusively by activity-dependent mechanisms, improved refinement (group RF −19% from to ; group PF −22% from to ). The reason for this improvement appeared to be that while the coarse guidance provided by molecular guidance cues was necessary to guide axons to near their retinotopically correct termination zones, once the axons had arrived, coarse guidance worked against activity-dependent refinement by broadening the area of the arbor where growth occurred. Activity-dependent mechanisms focused axon and synapse growth to the vicinity of synapses that induced spikes in the postsynaptic neuron. Molecular guidance cues worked to increase axon and synapse growth in a relatively broad region of heightened chemoaffinity compatibility, thus diffusing the focusing effect of activity-dependent refinement. These results suggest that molecular guidance cues are critical for establishing initial retinotopic order, but after this order is established, they are not necessary to refine the connection, consistent with an analysis of experimental results [35],[36]. Moreover, it appears that the influence of molecular guidance cues might actually inhibit refinement after axons arborize in the vicinity of their termination zones.
To examine the contribution of STDP to refinement, synaptic plasticity was disabled and all synapses maintained a unitary strength. Unexpectedly, this did not significantly affect model behavior (group RF +0.5% from to ; group PF +0.5% from to ). An analysis of the synapses in the unaltered model showed a narrow distribution of synapse weights (; n = 301,343 synapses), so locking synapse weights to unity had little quantitative significance. It is possible that a different implementation of STDP than used here could have more strongly contributed to development, but these results show that STDP, or any form of plasticity regulating the weight of individual synapses, is not required for retinotopic organization or refinement. Plasticity was instead realized through the growth and retraction of individual synapses. LTP and LTD are associated with increases and decreases in the number of synapses, respectively, consistent with the results observed here (see [37]).
Activity-dependent release of brain-derived neurotrophic factor (BDNF) has been linked to long-term potentiation and STDP [38]–[40]. In the model, trophic factor release was linked to STDP, such that trophic factor was released by the postsynaptic terminal proportional to STDP potentiation under a simple pair-based STDP protocol (e.g., [20]). In light of these findings about STDP, the importance of this linkage was investigated by decoupling them and varying the time window for activity-dependent trophic factor release. Specifically, the time window for trophic factor release ( in Eq. 10) was increased by 2, 4 and 8 times (from 13.3 ms up to 106 ms) and the magnitude of trophic factor release was proportionally reduced to account for the longer release window. Trophic factor release was also varied by using a square-wave function, such that if a postsynaptic spike followed within 25 ms of a presynaptic spike, a fixed amount of trophic factor was released (0.532 units, a magnitude that made equal the integrals of square wave and exponential release). In all cases, development was quantitatively and qualitatively similar (maximum changes of +14% PF, +13% RF were observed using the longest time window). Eliminating the activity-dependent mechanism underlying trophic factor release and having trophic factor released on every postsynaptic spike completely prevented refinement, with too much release resulting in very little synapse turnover, as most synapses became stabilized by the trophic factor received, and too little release preventing synapse stabilization and causing very high rates of turnover. These results indicate that activity-dependent trophic factor release, or an equivalent mechanism providing performance feedback to individual synapses, guides the removal of inappropriately targeted synapses and refines the retinotopic projection. The time window for this mechanism, here described as trophic factor release, is consistent with the STDP potentiation window, but is appears not to be restricted to that interval.
Development in the model was split into two distinct stages, with axon growth first being mediated by molecular guidance and later, after axons had reached the vicinity of their retinotopically correct termination zones, activity-based feedback began to contribute to guide axon growth. As shown in Fig. 1, this behavior allows nearby RGCs to project to the same areas of the colliculus and to form a refined retinotopic map. While this temporal segregation of roles worked well and is in line with experimental literature [35],[36], initial assumptions of the model were that molecular guidance and activity-based mechanisms both provided instructive guidance from the time when axons first innervated the colliculus. This coincident onset of guidance cues performed poorly, as axon arbors began to refine before they reached their retinotopically correct areas of the colliculus, resulting in numerous ectopic projections (Fig. 8). Delaying activity-dependent instructive cues until after molecular mechanisms had guided axons to the vicinity of their correct termination zones greatly reduced the incidence of ectopic projections and allowed normal organization and refinement to occur. Ectopic projections were sometimes still observed, but these were largely restricted to areas of the collicular boundary. The predominant view in the experimental literature suggests that molecular guidance is required to initially drive axon development, whereafter activity-dependent mechanisms guide refinement [35],[36]. Our findings go beyond this and suggest that, at least during initial development, a separation of mechanisms is necessary.
The effect of early activation of activity-mediated guidance is that it reduces the relative strength of chemoaffinity-mediated growth, as molecular guidance cues become forced to compete with activity-dependent ones. Experimentally, weakening the molecular guidance mechanisms by blocking production of guidance molecules also results in increased ectopic projections (e.g. [1],[2]), consistent with the behavior observed here. Malformed retinotopic projections resulting from early onset of activity-dependent mechanisms were prevented by sufficiently increasing the relative strength of molecularly-driven guidance, allowing both guidance mechanisms to act simultaneously. However, such increases also resulted in accurate axon targeting in the complete absence of activity. This behavior suggests that small animals such as zebrafish, that do not require neural activity for axons to project to their retinotopically correct targets [41] (but see [42]), and whose tecta are a small percentage (2%) of the length of tecta in larger animals, such as chick [24],[43], may not be adversely affected by early onset of activity-dependent guidance. Larger animals, whose tecta (superior colliculi) are much larger and presumably have much shallower molecular gradients, will be impacted more significantly.
Model synapses required trophic feedback for survival. It was assumed that this mechanism was cooperative, such that trophic factor received by one synapse would also help stabilize nearby synapses on the axon. The theoretical value to such a mechanism, in addition to helping to concentrate synapses to particular areas of an axon arbor, is that trophic feedback from one type of target neuron can help stabilize axonal synapses to different types of nearby neurons (e.g., GABAergic interneurons in retinogeniculate projections), thereby providing a mechanism to spatially align the projection to two (or more) disparate types of target neuron. Functionally similar polysynaptic mechanisms have been hypothesized, such as resulting from rapidly diffusible molecules (e.g., nitric oxide [44]). To evaluate the importance of this assumed behavior, development was examined with the stabilizing effect of trophic factor restricted to the synapse where it was received. Trophic factor receipt continued to influence axon growth and the distribution of axon resources normally. This modification did slow retinotopic refinement, but it also improved the degree of refinement realized (Fig. 4E), reducing RF size by 16% (from to ) and reducing PF by 13% (from to ). While the ability of synapses to help stabilize their neighbors can affect the rate of retinotopic refinement, it is not required to achieve a refined retinotopic projection.
In addition to the simple integrate and fire model used to represent collicular neurons in this study, previous versions of this model used non-linear integrate and fire neurons (i.e., [45]) and two-compartment neural models, and these changes did not qualitatively affect model behavior (data not shown). To investigate the possibility that neural growth had an influence on retinotopic refinement, collicular neurons were allowed to grow during development, with growth defined as an increase in the resting conductance of the neuron with time, such as occurs with increased surface area of the neuron and dendrite. The effect of such growth was that individual synapses had larger somatic excitatory post-synaptic potentials (EPSPs) on immature neurons than on mature ones. Collicular neuron growth was found to influence refinement in a non-linear way, with maximal refinement observed in neurons having moderate growth (−11% RF and −11% PF compared to control). Further increasing maximal growth reduced refinement. Preventing neural growth, such that the somatic EPSP of neurons resulting from a single presynaptic vesicle release was identical in immature and mature neurons, reduced refinement (+16% RF and +20% PF). It thus appears that neural growth, as exhibited by the decreasing somatic EPSP of individual synapses with time, has an influence on retinotopic refinement. Despite this influence, refinement still appears to be a tolerant process and was observed across a wide range of growth values, and more generally, that retinotopic development remains largely normal despite changes to the mechanisms underlying organization, so long certain core behaviors remain, which are spatiotemporally correlated retinal activity, molecular guidance cues, and activity-dependent trophic factor release.
This study has demonstrated how spiking activity, molecular guidance cues and activity-dependent trophic factor release can guide growth and retraction of individual axons, axon branches and synapses to produce the emergent property of retinotopic organization. While there are many components to the model, its functional behavior is relatively simple. The chemoaffinity-mediated bias for each axon to grow to the vicinity of its retinotopically correct termination zone (TZ) results in axons from nearby RGCs producing diffuse axonal projections near the TZ. The relative density of these axons is higher near the TZ, as is the relative density of the synapses on these axons, even if the synapses from any particular arbor are not well targeted (Fig. 2). Retinal waves cause nearby RGCs to fire together. Because the relative density of synapses from neighboring RGCs is higher near their TZ, collicular neurons near the TZ are more responsive to the activity of these RGCs than are collicular neurons farther away, and hence release relatively more trophic factor to their innervating synapses, stabilizing these synapses. The relative sparseness of synapses on collicular neurons further from the TZ results in them being less able to induce spikes, thus receiving less trophic feedback and retracting. Selective stabilization, along with trophic feedback enhancing synapse and axon growth in the area it is received, produces a self-reinforcing mechanism that results in refined axonal projections. Importantly, it provides a mechanism that enables neural components to use locally available information to generate global order.
The modeling approach used here rests on the assumption that neural development is a process involving several interacting mechanisms and it differs from existing neural development models in many ways, most notably by the degree that it is functionally constrained by biological behaviors not explicitly represented in network-level models (e.g., [14]–[22]), including the physical limitations governing axon growth, the functional requirements of forming and retracting synapses, the spike-based communication employed by neurons, and phenomenological approximations of many physiological behaviors. These constraints allow the contribution of, and interactions between, the different phenomena to be evaluated. It is difficult to examine the role and contribution of these underlying phenomena in models based on abstract descriptions that are open to multiple interpretations (e.g., energy functions) or that only represent one or a few phenomena. For example, if a model representing chemoaffinity and not retinal activity can produce a refined retinotopic map (e.g., [15],[21]), but experiments show that retinal activity is required [35],[36],[46], it follows that either the modeled chemoaffinity representation is functionally incorrect as the model contradicts experimental data, or that the modeled chemoaffinity representation implicitly includes activity and so does not accurately represent molecule-driven guidance. In either case, it is difficult to realistically evaluate either the role or contribution of chemoaffinity using such a model. In more general terms, it has been argued [47],[48] that for a model to have explanatory status, it must replicate the different causal mechanisms underlying the system being modeled, not only reproduce the output. While abstract models that examine only one or a few underlying mechanisms might be useful at providing conceptual insight into map development, the explanatory status of a model, and the detail of its predictions and conclusions, are limited by both the detail of the model and the relative accuracy with which the underlying mechanisms are represented. A detailed model such as described here, which is constrained through representation of a broad range of the phenomena contributing to retinotopic organization, should be better suited than contemporary modeling approaches for examining the role and interaction of different mechanisms underlying development, and for making predictions about these phenomena.
The design of the model was based on an analysis of the physiological mechanisms active during development, and the practical biological requirements of these mechanisms. From a physiological perspective, a neuronal projection is defined by the pattern of synapses that exist between bodies of pre- and postsynaptic neurons. The location of these synapses is constrained by the presence of axons, which in turn are constrained by patterns of growth, branching and retraction, as lateral axon motion is not realistic. The growth and retraction of axons and synapses must be governed by locally available information. Neurotrophins, such as BDNF, are prime candidates for mediating this process. Neurotrophins enhance axon growth and synapse numbers [49], are hypothesized to play a role in synapse stabilization and maintenance [50]–[52] and are released in an activity-dependent manner [39],[51]. The effects of BDNF are local to the area released [53], may be synapse specific [13],[51] and postsynaptic activity within tens of milliseconds of presynaptic activity results in synaptic enlargement in a process mediated by BDNF [54]. Molecular guidance cues also influence axon growth [55] and more generally, cellular behaviors are influenced by variations in firing rates (e.g. [56]). Based on these points, two core assumptions were derived to guide model behavior:
1) Axon growth and branching, and synapse formation, had increased probabilities in areas of an arbor with greater relative (a) chemoaffinity compatibility with surrounding tissue than other sections of the arbor, and (b) trophic feedback to the presynaptic terminal, which was provided by the postsynaptic terminal when a postsynaptic spike followed shortly after a presynaptic spike.
2) Synapses required trophic feedback for survival, and synapses with insufficient trophic support were eliminated.
To implement these principles, additional considerations were required, such as how to regulate the size of the axon and the synapse population. This led to three further assumptions:
3) To limit total axon arbor size, axons required a regulated substance, referred to here as axon resources, in order to grow and to persist. Axon resources were produced in finite quantities by the soma and were delivered preferentially to regions of the arbor with higher relative chemoaffinity and to near synapses receiving relatively more trophic feedback. A reduced presence of axon resources resulted in an increased likelihood of axon retraction.
4) To control the number of axonal synapses, the probability of new synapse formation was decreased with increased numbers of existing synapses, and each synapse required increasing amounts of trophic factor to survive with increasing numbers of synapses on the axon.
5) The number of dendritic synapses was controlled through direct and indirect means. The more synapses present on a dendrite, the less likely the dendrite was to accept new synapses. When a collicular neuron's average firing rate (integrated over tens of minutes) was above its target level, it both became less likely to accept new innervating synapses, and existing synapses decreased the trophic feedback provided to presynaptic terminals in order to induce some innervating synapses to retract.
The model was based on implementation of these mechanisms. The first two assumptions were core to the model's behavior and so it was not possible to carefully evaluate alternatives. The others assumptions were tolerant to variation so long as the behavior that these assumptions were designed to produce was realized (as assessed through both parameter variation and unpublished versions of this model). Homeostatic mechanisms were found to be important in the model design. The complexity of the model made it a practical impossibility to pre-define numerical quantities for the large range of mechanisms represented, such as exact EPSP magnitude, the number of synapses, total axon length, trophic feedback quantities, etc. Even when it was possible to define specific values for a quantity, minor modifications to the model often made such selections inappropriate, forcing the parameters to be readjusted. Defining quantities loosely and in such a way that they were subject to dynamic regulation (e.g., assumptions 3–5) produced a system that was very tolerant to perturbation. The same issues encountered in producing this model are also observed by nature, as there is a high degree of variability that can arise from genetic and environmental factors, and the biological system is tolerant to perturbation and it preserves its functionality despite changes to the mechanisms underlying this functionality [57].
The finding that STDP was not required for retinotopic refinement was unexpected. On reflection, this finding is consistent with the results of several experimental studies. Synaptic plasticity saturates after 60–100 spike pairings [7],[58], meaning that synapses that are already maximally potentiated for a given interval between pre- and postsynaptic spikes do not further potentiate. The fact that it is possible to observe significant synaptic potentiation and depression in STDP studies therefore suggests that most synapses exist in largely non-potentiated states, for otherwise such potentiation would not be observable in them. The notion that synapses are not significantly potentiated or depressed in their normal state is reinforced by findings that artificially induced STDP is lost if cells are allowed to resume their normal firing patterns [59] and that the distribution of individual synapse strengths is unimodal [60]. Further, experimental studies have indicated that it is either the timing of bursts between pre- and postsynaptic neurons, or the coincidences of individual spikes, that underlies plasticity, not STDP [61].
Cross-correlograms (CCGs) between pairs of monosynaptically connected cells often show a number of uncorrelated spike pairs and a peak a few milliseconds offset from time zero (Fig. 9; e.g., [62]), indicating that the postsynaptic cell has a higher than average probability of firing immediately after the presynaptic neuron, a behavior observed in modeled neurons. This CCG pattern should result from any system where there are several innervating neurons for each target neuron and when a spike in the presynaptic neuron is followed by a spike in the postsynaptic neuron only infrequently. While the peak in the CCG between monosynaptically connected cells is typically in the optimal location for inducing maximum potentiation in the synapse, the relatively large number of non-correlated firings would be expected to have a counteracting and depressing effect on plasticity. Further, when observing saturation of plasticity constraints, where maximal plasticity appears to be approached asymptotically [7],[58], the more a synapse is potentiated the less potentiating force there is after every pre-post spike pair. Depressing spike pairs (i.e., post-pre) are far from their saturation level and so are likely to maintain full potency, further inhibiting strong potentiation.
Manipulating the strength of individual synapses is not the only way to vary the effective strength of a monosynaptic connection. As indicated in Fig. 3C, variation in the strengths of synaptic coupling between two neurons can be achieved by variable numbers of synapses between the cells. Increasing or decreasing this number alters the effective strength of the “monosynaptic” connection. Based on the interpretation that “synapse” plasticity is realized by altering the number of synapses between neurons, the model's use of trophic factor release to regulate synapse stabilization, retraction and growth is consistent with the single-spike coincidence plasticity rule of Butts et al. [61], if plasticity is considered to represent the formation and retraction of synapses rather than the modification of the weight of individual synapses. The timing of individual spikes in pre- and postsynaptic neurons is important for plasticity, and the plasticity realized is Hebbian in nature, but it appears to not be realized by the changing of the efficacy of individual synapses.
It is possible that the implementation of STDP used here is too restrictive and that a different implementation could more strongly contribute to retinotopic organization and refinement. However, as argued above, STDP may not play a significant role in development. If that is the case, why is it a seemingly ubiquitous phenomenon? The neurotrophin BDNF has been associated with STDP and LTP [38]–[40]. Further, BDNF in the presence of glutamate mediates enlargement of synaptic spines in hippocampal slices [54], while LTP is associated with an increase in the number and size of synaptic spines, and LTD is associated with spine shrinkage and retraction [37]. It is entirely possible that what is observed as STDP experimentally is actually the byproduct of another mechanism, such as synapse stabilization. Using the retinocollicular projection as an example, numerous synapses are created during development but the only ones that persist are those that produce the refined retinotopic projection. There must exist a mechanism to remove inappropriately targeted synapses. One mechanism to accomplish this, as demonstrated here, is the activity-dependent release of trophic factors, where synapses contributing to a spike in the postsynaptic neuron receive trophic support and stabilize, while synapses receiving insufficient trophic factor retract. The timing of trophic factor release in the model is consistent with the time window for STDP potentiation. What is observed experimentally as STDP, at least in retinotectal synapses, might be an experimental artifact of a process relating to synapse stabilization and retraction, with what is observed as potentiation reflecting a mechanism that stabilizes synapses and depression reflecting a mechanism that makes the synapse more likely to retract. It is also possible that STDP is a redundant or complementary to another mechanism, or that it plays a functional role that was not examined in this study (e.g., [63]).
The extent and direction of axon growth in the model was mediated by probabilistic growth and retraction. Areas of an arbor with higher chemoaffinity compatibility with their surroundings, and/or increased trophic feedback, were more likely to extend and branch, and areas with relatively lower amounts were more likely to retract. During chemoaffinity-mediated growth, this mechanism was sufficient to produce a coarse retinotopic projection (Fig. 1C, D). After activity-dependent feedback began to influence axon growth, this same principle was able to generate refined arbors in the retinotopically correct termination zone. What is notable about this finding is that both chemoaffinity and activity-based axon guidance can be mediated by the same functional mechanism, and that the gradient detection and tracking of extracellular molecules by growth cones was not required during arborization and refinement. We note that growth-cone mediated guidance is still required for an axon to reach the colliculus and extend to its posterior pole.
Axon growth cones can detect gradients with remarkable sensitivity [64],[65], however it is not clear that the expression of guidance molecules is sufficiently smooth at the cellular and sub-cellular level to support such accurate guidance during retinotopic organization and refinement, especially considering that similar guidance molecules are expressed not only on collicular neurons but also on innervating axons [24], that measured mRNA levels for guidance molecules may not be locally smooth [1],[66] and that there may be non-uniformities in the density of axons and dendrites. If an axon is able to sense the relative difference in chemoaffinity compatibility in different parts of the arbor, and use this to influence the relative likelihood of local growth in the arbor, the arbor is effectively able to act as a very large gradient detecting growth cone. This behavior has been previously postulated for chick tectal development [43]. Such a mechanism could guide axon growth in the presence of shallower gradients, or in a noisier environment, than possible by gradient detection in individual growth cones.
Axon growth in the model does phenomenologically approximate experimentally observed patterns of axon growth, with initially coarse arbors refining into retinotopically ordered projections in the presence of normal retinal activity patterns (e.g., [33],[49]), and that the number of synapses and axon branches increase with exposure to trophic factor [49]. However, the implementation is very simplified compared to biology. Physiologically, there are interactions between the molecular machinery underlying chemoaffinity and trophic factor influence on axon growth (e.g., [67],[68]) and it is possible that activity-dependent influences are present throughout axon arbor development, and also that trophic factors help regulate the influence of molecular guidance cues [68]. On the other hand, molecular guidance cues could simply be sharing the same signaling pathway as trophic factors, and despite this molecular overlap between mechanisms, both could remain functionally distinct. Similarly, only the positive effects of trophic release were represented, not the opposing behaviors of trophic factors, where mature forms of the molecules promote growth, and the immature uncleaved molecules, such as proBDNF, promote synapse and axon retraction [37]. While a more mechanistically accurate model of axon growth will provide better insight into the molecular interactions involved in signaling axon and synapse growth and retraction, we found nothing to indicate that our phenomenological approximation of axon growth would be significantly different with a more mechanistic representation, nor that a more mechanistic representation would alter our findings on the overall behavior of growing axons.
Retinotopic development has been the subject of many computational models [69], and these models have been used to help identify the functional mechanisms necessary for development. In order to produce an ordered projection, the majority of these models (but not all, e.g., [18]) assume lateral connectivity between target neurons, where typically activity in one neuron results in excitation of nearby neurons and inhibition of neurons farther away (see [69]). This excitation/inhibition mechanism imposes architectural requirements on what is necessary for organization, and the high reversal potential for chloride early in development [70] suggests that lateral inhibition is not realistic, as synapses traditionally considered inhibitory (e.g., GABAergic) would be excitatory during the period of retinotopic organization and refinement.
In this study we have found that lateral synaptic connectivity was not required for producing an ordered retinotopic map, simplifying the theoretical functional requirements of the developing network. Simulated axons from neighboring RGCs were able to target the same collicular neurons based on their correlated firing properties and on the stabilizing effects of trophic factor. Synapses from RGCs stabilized on collicular neurons that were responsive to their activity by means of increased trophic factor receipt (Fig. 2B). Correlated activity between nearby RGCs resulted in axons from nearby RGCs targeting the similar collicular neurons. Over the course of hours of simulated time, this mechanism results in increased axon and synapse growth in the area where more trophic feedback was received and these new synapses targeted nearby collicular neurons, focusing the axon projection. Collicular neurons sought to maintain a target firing rate, producing a normalizing force that limited the number of synapses present. Because of these factors, the resulting projection was ordered at the global level (Fig. 1F) though was subject to scatter at the local level (Fig. 1G).
The focus of this study was to examine the behavior and interaction of the mechanisms underlying neural development, and the approach here follows that used in the modeling of other complex phenomena, most notably climate [71]. Both climate and neural development are examples of complex systems, and predictive and descriptively accurate models of such complex systems may themselves be complex and not necessarily capable of being simplified to a simple or mathematically analyzable form. Climate models represent approximations of many of the causal mechanisms underlying weather, such as radiation, cloud cover, humidity, momentum, sea surface temperature and pressure gradients [71],[72]. The model described here addresses retinotopic organization and refinement as being causally produced from phenomenological approximations of many mechanisms known to be active during development of the retinocollicular projection. Two very important mechanism are the growth, branching and retraction of individual axons, and the durability of individual synapses. Axon growth is a process underlying the formation of all neural projections, and axons have extremely restricted movement once extended through neuropil. Synapses must retract based on information available to each individual synapse. A descriptively accurate model of retinocollicular development requires consideration of the physical constraints posed on development by these and other mechanisms.
The model represents many physiological phenomena active during development in as simple a form as practical while still approximating the functional behaviors of the phenomena. The lack of detailed representation of these mechanisms can be justified, we would argue, because the details of the mechanisms can vary between species though the developmental outcomes are similar. For example, similar patterns of retinal waves are observed in many species yet their statistical and molecular details vary (see [5],[30],[73]). Likewise, chemoaffinity gradients are a common phenomenon but they are mediated by different molecules in different species [24]. Despite these differences, similar patterns of retinotopic organization persist. It stands to reason that it is the commonalities of behavior observed between species that are important for producing the common patterns of development, not what are essentially biological implementational details. The results of this study support such a conclusion, as qualitatively similar development was observed despite variations and perturbations to the model. It was only when key functional mechanisms were disabled that the model failed to produce retinotopic organization or refinement.
Predictions of the model include that:
Although the model is restricted to the retinotectal/retinocollicular system, the phenomena represented in it are found in neurons throughout the brain, and the findings here may apply more broadly. With minor modifications, the model is potentially applicable to the description of development in different brain areas. Explicit representation of many physiological mechanisms allows the model to be more easily compared to and constrained by physiology than most contemporary modeling approaches. It may be that the most predictive and descriptively accurate models of retinocollicular development, and of neural development in general, will resemble the approach described here, incorporating phenomenological approximations of many physiological mechanisms, in particular explicit representation of the growth and retraction of individual axons and synapses.
The structure of the model is shown in Fig. 10A. A circular retina composed of 7915 RGCs projected to an octagonal colliculus having 7934 neurons. Neurons in both retina and colliculus were distributed on a hexagonal matrix. The model retina was circular (diameter 1.6 mm) and the colliculus had 110 rows of neurons with each row having 80 neurons (0.8 mm×0.94 mm), with the corners of this rectangle truncated. In a reduced size version of the model that was used for model analysis, only the central 30% of the simulated retina and colliculus were modeled (Fig. 10A, white rectangular areas). The smaller model had 3023 RGCs projecting to 2694 collicular neurons. The model was not sensitive to small changes in the ratio of retinal to collicular neurons, but this was not systematically explored. Map compression and expansion was examined in a previous version of this model [74]. The dendritic radius for each collicular neuron was . The soma of collicular neurons was considered to reside at the center of the dendritic arbor.
Axons in the model were represented as a connected series of segments, each in length, a size selected to be sufficiently small to allow for realistic patterns of growth but large enough to make the model computationally tractable. Each axon segment was able to extend and branch, and retraction occurred at axon tips (Fig. 10D). Axon segments were considered to have an “affinity” for their surroundings, which determined their propensity to grow, sprout synapses and retract. Axon segments required resources to grow, and the availability of these resources was managed by the soma. Segments received an amount of growth resource that was a function of the segment's affinity. Axons with higher amounts of growth resources were more likely to extend, branch and generate new synapses, while segments with lesser amounts were more likely to retract. To achieve self-limiting axon growth, each soma was assumed to have a finite amount of growth resources that was distributed throughout the arbor.
Simulations began with each RGC axon extending along the A-P axis of the colliculus, corresponding to development as seen in P1 mouse [33]. Initial axon placement had each RGC axon entering the colliculus at the anterior side and extending along the length of the anterior-posterior (A-P) axis, in a position along the lateral-medial (L-M) axis that corresponded the the RGCs location along the retinal dorsal-ventral (D-V) axis. The exact L-M position varied by a random amount (a Gaussian distribution with mean zero and standard deviation of 20% of the width of the colliculus). This design was based on descriptions of mouse and chick retinocollicular development [24],[34],[43]. The colliculus had flat sides both so axons could linearly project along the collicular boundary and so the model did not rely on an isotropic projection from retina to colliculus. The orientation of axon segments in the initial projection was parallel to the A-P axis except for a small random variation. Specifically, the orientation of each segment was described by 2 vectors, one of unit length and parallel to the A-P axis, and a second perpendicular vector whose magnitude was a uniform random variable selected on the interval [−0.2, 0.2]. Subsequent branching and growth occurred as described below.
Development occurred in two stages, each lasting 60 hours of simulated time. During the first 60 hours, development was mediated by chemoaffinity, and interstitial branching and subsequent growth was guided only by the differential in chemoaffinity compatibility across the arbor. During the second 60 hours, trophic feedback and chemoaffinity both contributed to growth. While synapses may be present throughout axon development in the colliculus, synapse creation in the model was inhibited until the onset of trophic feedback influence on axon behavior (i.e., 60 hours development time) as synapses had no influence on axon growth before this time. This allowed the first 60 hours of axon growth to be pre-computed and used as a starting point for simulations of the second development stage, reducing the computational requirements of the model. Quantitative analysis as reported in Results was performed at 119 hours as synapse generation was turned off during final hour of the simulation to assess the stability of synaptic projections and to passively allow poorly targeted synapses to retract (e.g., note removal of mistargeted synapses at 120 hours in Fig. 3C). PF and RF sizes were reduced as a result of passive pruning, but the projections were qualitatively similar (data not shown).
The excitation level of model neurons were updated on every simulation clock cycle (1 ms) while synapses were updated only on the occurrence of a pre- or postsynaptic spike. When a neuron fired, it cycled through all its axonal synapses, “pushing” excitation onto the target cell of each, and updating synaptic potentiation based on STDP learning rules for a presynaptic spike. The neuron then cycled through its dendritic synapses, updating synaptic potentiation based on the occurrence of a postsynaptic spike. To improve simulation performance, many cellular behaviors were updated less frequently. Equations relating to axon growth, branching and retraction, and to synapse growth were recalculated every 5 sec simulated time. Equations relating to synapse resources, synapse retraction, axon resources, homeostatic controls and intra-axon diffusion were recalculated every 0.5 sec. With the exception of millisecond calculations (e.g., EPSP summation, STDP and trophic factor release), the model was not dependent on the interval between updates. Different intervals were used in some simulations and no change to model behavior was observed.
Previous versions of this model ([74] and unpublished) used mathematically different but functionally similar representations of these mechanisms and produced qualitatively similar results.
|
10.1371/journal.ppat.1003606 | Methionine Biosynthesis in Staphylococcus aureus Is Tightly Controlled by a Hierarchical Network Involving an Initiator tRNA-Specific T-box Riboswitch | In line with the key role of methionine in protein biosynthesis initiation and many cellular processes most microorganisms have evolved mechanisms to synthesize methionine de novo. Here we demonstrate that, in the bacterial pathogen Staphylococcus aureus, a rare combination of stringent response-controlled CodY activity, T-box riboswitch and mRNA decay mechanisms regulate the synthesis and stability of methionine biosynthesis metICFE-mdh mRNA. In contrast to other Bacillales which employ S-box riboswitches to control methionine biosynthesis, the S. aureus metICFE-mdh mRNA is preceded by a 5′-untranslated met leader RNA harboring a T-box riboswitch. Interestingly, this T-box riboswitch is revealed to specifically interact with uncharged initiator formylmethionyl-tRNA (tRNAifMet) while binding of elongator tRNAMet proved to be weak, suggesting a putative additional function of the system in translation initiation control. met leader RNA/metICFE-mdh operon expression is under the control of the repressor CodY which binds upstream of the met leader RNA promoter. As part of the metabolic emergency circuit of the stringent response, methionine depletion activates RelA-dependent (p)ppGpp alarmone synthesis, releasing CodY from its binding site and thereby activating the met leader promoter. Our data further suggest that subsequent steps in metICFE-mdh transcription are tightly controlled by the 5′ met leader-associated T-box riboswitch which mediates premature transcription termination when methionine is present. If methionine supply is limited, and hence tRNAifMet becomes uncharged, full-length met leader/metICFE-mdh mRNA is transcribed which is rapidly degraded by nucleases involving RNase J2. Together, the data demonstrate that staphylococci have evolved special mechanisms to prevent the accumulation of excess methionine. We hypothesize that this strict control might reflect the limited metabolic capacities of staphylococci to reuse methionine as, other than Bacillus, staphylococci lack both the methionine salvage and polyamine synthesis pathways. Thus, methionine metabolism might represent a metabolic Achilles' heel making the pathway an interesting target for future anti-staphylococcal drug development.
| Prokaryote metabolism is key for our understanding of bacterial virulence and pathogenesis and it is also an area with huge opportunity to identify novel targets for antibiotic drugs. Here, we have addressed the so far poorly characterized regulation of methionine biosynthesis in S. aureus. We demonstrate that methionine biosynthesis control in staphylococci significantly differs from that predicted for other Bacillales. Notably, involvement of a T-box instead of an S-box riboswitch separates staphylococci from other bacteria in the order. We provide, for the first time, direct experimental proof for an interaction of a methionyl-tRNA-specific T-box with its cognate tRNA, and the identification of initiator tRNAifMet as the specific binding partner is an unexpected finding whose exact function in Staphylococcus metabolism remains to be established. The data further suggest that in staphylococci a range of regulatory elements are integrated to form a hierarchical network that elegantly limits costly (excess) methionine biosynthesis and, at the same time, reliably ensures production of the amino acid in a highly selective manner. Our findings open a perspective to exploit methionine biosynthesis and especially its T-box-mediated control as putative target(s) for the development of future anti-staphylococcal therapeutics.
| Staphylococci are important skin and mucosa commensals but also major human pathogens. The most pathogenic species Staphylococcus aureus causes a wide range of diseases and, together with coagulase-negative staphylococci (CoNS), accounts for approximately 30 per cent of all hospital-acquired infections [1]. The development of antibiotic resistance in staphylococci increasingly limits therapeutic options and is a matter of major concern [2]. In recent years, studies into staphylococcal metabolism and its possible links to bacterial virulence have become a major focus of research but basic metabolic pathways remained largely unexploited in the development of new antibiotic drugs [3]. In this study, we investigate the regulation of methionine biosynthesis in staphylococci. Methionine and its chemical derivatives have important functions in the cell. For example, (formyl-)methionine is the universal N-terminal amino acid of nearly all proteins and therefore plays an eminent role in the initiation of protein biosynthesis. Moreover, the methionine derivative S-adenosylmethionine (SAM) serves as a methyl group donor in a variety of cellular processes and is the precursor molecule in polyamine synthesis [4]. Many microorganisms are able to synthesize methionine de novo and staphylococci employ the trans-sulfuration pathway to generate methionine [5]. Most bacteria from the order Bacillales are thought to control this pathway by SAM-binding S-box riboswitches [5], [6], [7]. Interestingly, in silico analysis predicts the presence of a T-box riboswitch in the 5′-untranslated region of the methionine biosynthesis operon (metICFE-mdh operon) in staphylococci [5], [6], [8], suggesting the use of alternative mechanisms to regulate methionine synthesis. T-box riboswitches are transcriptional control systems which have been extensively studied in Bacillus subtilis and other Firmicutes (reviewed in [6]). Their function is controlled by specific interactions and differential binding to charged and uncharged cognate tRNA, respectively, thus providing a means to “sense” the amino acid concentration in the cell [9]. T-box leader RNA/tRNA interaction essentially occurs at two sites: (i) the tRNA anticodon basepairs with the specifier-loop domain of the T-box leader RNA ensuring specific binding of the respective T-box element with its cognate tRNA; (ii) the free 3′-CCA end of an uncharged tRNA binds to the T-box motif, thereby triggering the formation and stabilization of an antiterminator which enables transcription of downstream genes [9]. In this study, we characterized interactions of the metICFE-mdh leader RNA with methionyl-tRNAs and demonstrate that they represent a functional T-box riboswitch that preferentially binds to initiator formylmethionyl-tRNA (tRNAifMet). We further show that, in staphylococci, T-box control of methionine biosynthesis has a key role in a complex regulatory network that also involves stringent response-mediated CodY regulation and RNA decay to tightly control this pathway.
The methionine biosynthesis genes (metI, metC, metF, metE, mdh (metal-dependent hydrolase)) are organized in an operon-like structure and are annotated as SACOL0431 - SACOL0427 in S. aureus COL and as NWMN_0351 - NWM_0347 in S. aureus Newman, respectively, with metF being named metH in the latter strain (Figure 1A). Northern blot analysis using a double-stranded DNA probe confirmed the expression of a stable transcript of approximately 400 nucleotides (nt) from the intergenic region (IGR) upstream of metI (Figure 1B, left panel). Hybridization employing in vitro-transcribed RNA probes revealed that the orientation of the transcript was identical to that of the metICFE-mdh operon (Figure 1B, middle and right panels). 5′- and 3′-RACE experiments identified a single transcription start site and a transcript length of 439 nt (Figure 1C). Sequence analysis of various S. aureus and S. epidermidis strains demonstrated that the region is highly conserved (SI Figure S1) but lacks ribosomal binding sites and open reading frames, suggesting specific (non-coding) functions of the transcript, for example, as a 5′-untranslated region (5′-UTR) of the metICFE-mdh RNA. Interestingly, a putative binding site for the repressor protein CodY [10], [11] could be identified next to the 5′-UTR promoter region (Figure 1C). Most striking was the presence of a highly conserved canonical T-box sequence motif (5′-AAGGUGGUACCGCG-3′) which partially overlapped with a strong Rho-independent transcription termination signal in the 3′-portion of the transcript (Figure 1C). Overall, the transcript, which was named met leader RNA, harbored all the characteristics of previously characterized T-box riboswitches from Bacillus subtilis, and sequence alignments with Bacillus T-box systems led to a putative structural model of the Staphylococcus met leader RNA (SI Figure S2).
First, we tested if the Staphylococcus met leader RNA interacts specifically with methionyl-tRNAs (Figure 2A). met leader RNA was in vitro transcribed in the presence of the appropriate radioactively labeled tRNA species. Binding between radioactively labeled tRNA and in vitro-transcribed met leader RNA was determined by non-denaturing polyacrylamide gel electrophoresis and autoradiography. Staphylococci genomes harbor four methionyl-tRNA gene loci, two of which (tRNA-Met-1 and -2) being identical and representing the initiator tRNAifMet. Binding studies using tRNAifMet, tRNAMet3 and tRNAMet4, respectively, with free 3′-CCA ends revealed that the met leader RNA interacted strongly with tRNAifMet while interactions with tRNAMet3 and tRNAMet4 proved to be weak (Figure 2B). tRNAifMet binding to met leader RNA increased linearly within a 5-fold molar range (Figure 2C). In contrast, binding was abolished when the 3′-end of tRNAifMet included one additional cytosine, mimicking a charged tRNA molecule (3′-CCAC; AdC in Figure 2D). Also, no interaction was detectable in the presence of cysteinyl-tRNA, regardless of whether 3′-CCA or 3′-CCAC was present at the 3′ end (Figure 2C, D).
In classical T-box riboswitches, tRNA/leader RNA interaction is mediated by the T-box motif which forms a bulge that facilitates basepairing interactions with the tRNA 3′-CCA and supports antiterminator formation [9] (Figure 2A). We therefore studied whether the predicted T-box motif participates in tRNAifMet binding by generating a series of met leader RNAs carrying T-box mutations (SI Table S2, Figure 3A). tRNAifMet binding was clearly diminished in mutants SC2 and SC5, which are both likely to lack the putative T-box bulge for tRNA 3′-CCA interaction (Figure 3A). Also, in SC8, a single U to A exchange at position 363 was sufficient to reduce tRNAifMet binding, whereas other mutations within the putative T-box bulge, i.e. SC3, SC4, SC6 and SC7, enhanced tRNAifMet interactions with the met leader RNA (Figure 3B). Finally, alteration of a methionine-specific codon AUG to cysteine UGC in the putative specifier box of the met leader RNA in SC1 did not affect tRNAifMet binding efficiency (Figure 3B), nor did it confer tRNACys binding activity (SI Figure S3) suggesting that other components of this T-box system confer specificity to initiator tRNAifMet binding. Taken together, these data suggest that the met leader RNA upstream of the staphylococcal metICFE-mdh operon harbors a canonical T-box riboswitch that specifically binds uncharged initiator tRNAifMet.
Through interaction with either uncharged tRNAs (antiterminator formation) or charged tRNAs (terminator formation), T-box riboswitches indirectly sense amino acid levels in the bacterial cell [6], [9]. To determine whether or not met leader RNA/metICFE-mdh transcription is sensitive to methionine availability, S. aureus strain Newman was grown in chemically defined medium (CDM) in the presence or absence of methionine. RNA was isolated in the early exponential (E1), exponential (E2) and early stationary (S) growth phase and analyzed by Northern hybridization using met leader RNA- and metI-specific DNA probes, respectively (Figure 4A, B). In the presence of methionine, basal met leader RNA transcription could be detected. Methionine starvation induced the transcription of this RNA, especially during exponential growth (Figure 4A). In contrast, the metI mRNA signal was not detectable in the presence of methionine (Figure 4B). Upon methionine deprivation, however, metI mRNA transcription was activated with the strongest expression detected during exponential growth (Figure 4B). Interestingly, metI-specific Northern probing did not reveal a distinct fragment representing the full-length metICFE-mdh mRNA (Figure 4B). Instead, an RNA smear was detected in repeated experiments. As the RNA quality and integrity had been confirmed prior to the experiments, the Northern blot data suggest rapid degradation of the methionine starvation-induced metICFE-mdh transcript.
CodY is a global transcription repressor that controls the expression of a variety of genes in S. aureus, many of which being involved in amino acid biosynthesis and transport [10], [11]. Detection of a consensus sequence for CodY binding upstream of the met leader RNA and the recent identification of the methionine biosynthesis genes as direct CodY targets in S. aureus [10] prompted us to study the role of this factor for met leader/metICFE-mdh transcription in more detail. A S. aureus codY deletion mutant was grown in CDM with and without methionine and analyzed for met leader and metICFE-mdh transcription by Northern blot hybridization. In the absence of methionine, a stronger met leader RNA signal was detected in both the wildtype and the codY mutant (Figure 4A). However, compared to the wildtype, the codY mutant showed a generally enhanced met leader transcription, suggesting that met leader RNA expression was de-repressed in the absence of CodY irrespective of whether methionine was present or not (Figure 4A). In contrast, downstream metICFE-mdh operon expression remained sensitive to varying concentrations of methionine and was only activated both in the wildtype and the codY mutant in the absence of methionine (Figure 4B).
In many bacteria, nutrient limitation triggers the so-called stringent response to appropriately adjust gene expression patterns. Stringent response is characterized by the rapid synthesis of the alarmone (p)ppGpp involving bifunctional RelA/SpoT synthetases/hydrolases (RSHs) and affecting/modulating many cellular functions. Recently, a link between CodY and the stringent response of S. aureus has been demonstrated [12]. The CodY repressor function depends on its two effector molecules GTP and branched-chain amino acids (BCAA, valine, leucine, isoleucine) which enhance synergistically the affinity of CodY for its DNA targets [13], [14]. RSH-mediated (p)ppGpp synthesis lowers the GTP levels in the cell and eventually facilitates release of CodY from its DNA targets [12]. In a next set of experiments, we sought to identify possible regulatory links between methionine deficiency, stringent response and CodY. For this purpose, met leader RNA/metICFE-mdh transcription upon methionine depletion was studied in a rsh mutant carrying a deletion of the (p)ppGpp synthetase domain in strain S. aureus Newman [12]. As a marker for stringent response-controlled genes, brnQ-1, which encodes a CodY-repressed BCAA permease, was included in the analysis. In the S. aureus wildtype, methionine depletion led to brnQ-1 induction along with met leader RNA and metICFE-mdh expression (Figure 5A). In contrast, in the rsh mutant, induction of both brnQ-1 and met leader/metICFE-mdh transcription were significantly reduced upon methionine starvation in comparison to the wildtype, suggesting that RSH-mediated (p)ppGpp synthesis may be required for efficient activation of the system. Deletion of codY resulted in a generally higher basal transcription of both brnQ-1 and met leader RNA in the presence of methionine, whereas metICFE-mdh transcription remained tightly controlled and switched off under these conditions (Figure 5A). In the codY mutant, brnQ-1 expression was de-repressed and not further inducible by methionine deprivation, suggesting that CodY is responsible for the methionine-dependent brnQ-1 induction observed in the wildtype (Figure 5A).
The experiments described in Figure 4 suggest that metICFE-mdh mRNA is subject to rapid degradation. To investigate the possible involvement of specific RNases in this process, the stability of met leader RNA and metICFE-mdh was analyzed in S. aureus mutants that were deficient in RNase J2 and RNase III activity, respectively. For this purpose, de novo RNA synthesis was interrupted by the addition of the RNA polymerase inhibitor rifampicin to the cultures. Total RNA was isolated at different time points and subjected to Northern blot analysis (Figure 5B). Comparison of the wildtype and the RNase-deficient mutants revealed that the metICFE-mdh transcript was more stable in the RNase J2 mutant, suggesting that RNase J2 may be involved in metICFE-mdh degradation. In contrast, no significant effect on met leader RNA stability was detectable in the RNase J2 mutant (Figure 5B, upper panel). The RNase III mutant exhibited a slightly enhanced stability both of the met leader and metICFE-mdh RNA indicating a possible function of RNase III in met leader RNA and metICFE-mdh decay (Figure 5B).
In this study, we show that methionine biosynthesis control in S. aureus involves a T-box riboswitch. While the conservation of this T-box in staphylococci was predicted previously using bioinformatic tools [5], [6], [8], we now provide, for the first time, direct experimental proof for a specific interaction of the predicted T-box with initiator tRNAifMet. While methionyl-tRNA-specific T-box riboswitches (met-T-box) are rare among Bacillales they are more common in Lactobacillales. In both orders they are associated with methionine metabolism or transport (Table 1). Methionyl-tRNAs (tRNAMet) are encoded by four distinct gene loci in the genomes of S. aureus and S. epidermidis. Two of them are identical and represent the initiator tRNAifMet, while the other two tRNAMet loci differ in their nucleotide sequence from each other and from the tRNAifMet. Surprisingly, we found a clear preference for interaction of the met leader RNA T-box with the initiator tRNAifMet (Figure 2B). In prokaryotes, the first N-terminal methionine of newly synthesized proteins is N-formylated and, hence, N-formylmethionine (fMet) is indispensable for protein translation initiation and bacterial growth. fMet is carried to the ribosomal translation initiation complex by tRNAifMet which differs structurally from the elongator tRNAMet used for the incorporation of methionine residues into the growing polypeptide chain. Although all tRNAMet are charged with methionine by (the same) methionyl-tRNA synthetase, it is only tRNAifMet that is specifically recognized by the methionyl-tRNA-formyltransferase which then mediates N-formylation of methionine to produce fMet. At present, it is not clear how the observed specificity of the met leader RNA T-box for tRNAifMet is accomplished. An involvement of the putative specifier box in the 5′-region of the met leader RNA seems unlikely because all methionyl-tRNAs use the same anticodon, suggesting that other regions of the met leader RNA interact with structures that are unique to the initiator tRNAifMet. In line with this hypothesis, our data showed that tRNAifMet binding by the met leader RNA was not affected when the specifier box in the met leader was substituted with a cysteine-specific codon (Figure 3B). Also, these nucleotide replacements were insufficient to confer tRNACys binding activity (SI Figure S3). Our observation that the T-box riboswitch, shown in this study to be a key regulator of methionine biosynthesis in S. aureus, preferentially binds tRNAifMet points to an elegant mechanism by which protein translation initiation efficiency could both be sensed and, if necessary, adjusted by modulating fMet supply. It will be interesting to investigate if this tRNAifMet preference also applies to other met-T-box riboswitches that control the expression of genes not directly involved in methionine biosynthesis. Also, potential metabolic implications of the use of T-box-controlled fMet supply in staphylococci versus S-box-controlled methionine biosynthesis in other bacteria remain to be studied.
The data obtained in this study lead us to propose that a hierarchical regulatory network controls methionine biosynthesis in S. aureus, most likely, to minimize unnecessary de novo methionine biosynthesis. Centerpiece of this regulation turns out to be the tRNAifMet-specific T-box riboswitch located in the 5′-met leader that precedes the coding regions of the metICFE-mdh mRNA. Another important player is the global repressor CodY which drives met leader RNA transcription and links the system to the metabolic emergency circuit of the bacterial stringent response. Finally, staphylococcal RNases were implicated in this network by degrading both metICFE-mdh mRNA and met leader RNA, which may be considered a form of posttranscriptional control of metICFE-mdh gene expression. Figure 6 summarizes our major findings and suggests a model for the control of methionine biosynthesis in staphylococci. While stringent response-mediated CodY release and subsequent met leader RNA transcription are sensitive to general amino acid availability and the energy status of the cell, the T-box riboswitch is highly selective and ensures that downstream metICFE-mdh mRNA transcription only occurs if methionine concentration is low. The experiments also indicate that lack of methionine alone is sufficient to trigger the stringent response and, as a consequence, the release of CodY, thus securing efficient met leader RNA/metICFE-mdh transcription when needed (Figure 6). Interestingly, the regulatory cascade identified in this study seems to represent an exception rather than common rule. Thus, database searches of B. subtilis and S. aureus genomes revealed that combinations of CodY with T-box riboswitches are restricted to methionine and tryptophan biosynthesis in S. aureus and branched-chain amino acid (BCAA) biosynthesis in B. subtilis, respectively [15], [16] (Table 1).
Methionine biosynthesis is a costly process consuming ATP and other resources. Riboswitches are generally regarded as fast and tight regulatory systems that do not depend on protein factors which, in many cases, react more slowly and/or are subject to complex regulation of protein expression, activation and degradation [5], [17]. Our study suggests that, in addition to regulation at the transcriptional level, S. aureus employs RNA decay mechanisms to quickly remove newly transcribed metICFE-mdh mRNA from the system, thus further limiting the risk of sustained (over)expression of genes involved in methionine synthesis. Although the complete mechanism and enzymes involved in the process still need to be established, the data give a first hint that RNase J2 participates in metICFE-mdh degradation (Figure 5B). This is in contrast to the situation in B. subtilis where mRNAs of the methionine biosynthesis genes, polyamine synthesis as well as the methionine salvage pathway (see below) were found to be not degraded by RNase J1/J2 [18]. The data further suggest that RNase III might be involved in met leader RNA degradation (Figure 5B). This observation is consistent with previously published data showing that RNase III co-immunoprecipitates with met leader RNA and targets also other S. aureus riboswitches for degradation [19]. Taken together, the data lead us to suggest that RNA decay is another mechanism involved in the control of methionine synthesis in staphylococci that merits further future investigation.
T-box control of methionine biosynthesis genes in staphylococci is an exception among Bacillales which usually regulate this pathway by S-adenosylmethionine (SAM)-binding S-box riboswitches [5]. Apart from protein synthesis, most microorganisms use methionine to produce SAM which plays a central role in many cellular functions [4]. First, SAM serves as a methyl group donor for nucleic acid and protein methylation. Products of the methylation reaction are detoxified and recycled to homocysteine which is then reused for methionine/SAM synthesis. Second, SAM is used, following decarboxylation, to form polyamines. The remaining 5′-methylthioadenosine moiety is again metabolized to methionine by enzymes of the methionine salvage pathway. Comparative genomics using the KEGG database (http://www.genome.jp/kegg/) and experimental research [20], [21], [22] suggest that, unlike Bacillus, staphylococci may have only limited capacity to reuse or redirect methionine to other pathways because they lack both the methionine salvage and polyamine synthesis pathways (Table 1). Therefore, synthesis and recycling of SAM may be the only possibility to make use of excess methionine, implying that a more stringent control of de novo biosynthesis may be required in staphylococci. Interestingly, the Lactobacillales, which preferentially control their methionine biosynthesis genes by T-box riboswitches [5], also appear to lack polyamine synthesis and methionine salvage genes (Table 1). Based on these observations, we hypothesize that the lack of both polyamine synthesis and methionine salvage might favor control by a T-box rather than a S-box riboswitch, the major advantage being that the T-box riboswitch is able to sense methionine supply directly and then react immediately by switching off transcription of methionine biosynthesis genes, whereas S-box regulated systems would require an additional step (i.e. SAM synthesis) to produce the effector molecule required to stop methionine production. Alternatively, microorganisms that produce polyamines may need a larger SAM pool as precursor for the synthesis of these important compounds. Therefore, S-box control of methionine biosynthesis might be more effective in these organisms to ensure a constant SAM supply. More experimental work is needed to further substantiate these hypotheses.
The general life style of staphylococci provides easy access to methionine sources from the respective host they colonize or infect and, therefore, suggests that methionine supply may not be a limiting factor under normal conditions. The data presented in this paper provide first insight into the regulation of methionine synthesis gene expression in staphylococci and, more importantly, show that staphylococci have evolved special mechanisms to tightly restrict de novo methionine biosynthesis. It is tempting to speculate that overproduction (rather than lack) of methionine may be critical to staphylococci and, thus, this strict control of methionine de novo synthesis would not only save resources and energy but also meet the requirement to prevent methionine accumulation. While excess methionine biosynthesis might be critical for staphylococci under most conditions, de novo methionine biosynthesis becomes crucial for growth and survival when methionine supply is limited, for example, when entering the host during infection [23] or under specific external stress conditions like antibiotic and antimicrobial peptide exposure [24], [25]. It is therefore conceivable that both the efficient activation and shut-off of methionine biosynthesis might represent a metabolic “Achilles' heel” for staphylococci. Interestingly, successful inhibition of the S. aureus methionyl-tRNA synthetase by an experimental compound has already provided evidence that methionyl-tRNA metabolism is a suitable anti-staphylococcal target [26]. Also, structure-based drug design recently resulted in the identification of lead compounds that specifically interact with T-box structures in vitro, indicating that RNA-based drug targeting is a promising new avenue in medicinal chemistry [27], [28], [29]. The data presented in this paper may open perspectives for specifically targeting methionine metabolism and protein translation initiation in future efforts to develop novel Staphylococcus-specific antibiotics.
Bacterial strains, construction of mutants, growth conditions as well as RNA isolation and Northern blot procedures can be found in the SI Materials and Methods (Text S1, Table S1–S5).
Total RNA was treated with recombinant DNase I (Ambion) and RNA quality was checked with the Agilent 2100 Bioanalyzer (Agilent Technologies). The 5′/3′ RACE Kit, 2nd Generation (Roche) was used for determination of transcript ends by synthesis of first-strand cDNA with reverse transcriptase (Fermentas) and the oligonucleotides listed in Table S4 (Text S1).
The met leader RNA sequence of S. aureus COL was amplified by PCR with oligonucleotides T7-F_met-sRNA and R_met-sRNA (Tables S5, Text S1). The product was inserted into the pGEM-T Easy vector (Promega) yielding plasmid pGEMmetCOL. For site directed mutagenesis, the oligonucleotides listed in Table S5 were used in PCR reactions with pGEMmetCOL as a template followed by DpnI treatment prior to transformation of the PCR products into Escherichia coli DH5α cells. The rescued plasmids were sequenced and constructs SC1 to SC8 (Table S2) with the appropriate mutations were used as met leader RNA templates in tRNA binding assays.
tRNA templates were generated by PCR from genomic DNA of S. aureus COL and oligonucleotides listed in Table S4 (Text S1). Four picomol DNA template were subsequently used for IVT in a 20 µl reaction volume consisting of T7 transcription buffer, 20 U RNase inhibitor, 20 U T7 RNA polymerase (Fermentas), 0.5 mM ATP, UTP, GTP, 12 µM CTP and 9 mM GMP as described [30]. [α32P]-CTP was added and the reaction was incubated at 37°C for six hours. For met leader RNA in vitro transcription, PCR templates were either generated from plasmid pGEMmetCOL or constructs SC1 to SC8 (Table S2) with oligonucleotides T7-F_met-sRNA and R_met-sRNA (Table S5). The products were used in IVT/tRNA binding assays in the presence of pre-formed tRNAs: 10 µl reaction volume consisted of T7 transcription buffer, 10 U RNase inhibitor, 6 U T7 RNA polymerase (Fermentas) and 0.5 mM NTPs. DNA templates of the met leader RNA and of pre-formed tRNA were adjusted to end concentrations of 8 nM and 50 nM, respectively, and the mix was incubated at 37°C for two hours. Samples were immediately separated on a non-denaturing 6% (w/v) polyacrylamide gel by electrophoresis at 4°C. Visualization was attained with the PhosphoImager (Fujifilm FLA-7000) and quantification of bands was achieved with the software Multi Gauge V2.2.
Bacteria from overnight cultures were diluted in 100-ml flasks in 40 ml CDM medium with methionine to an initial optical density at 600 nm (OD600) of 0.05 and grown with shaking at 220 rpm at 37°C to an OD of 0.5. The cultures were filtered over a 0.22 mm filter applying vacuum, washed twice with sterile phosphate buffered saline (PBS) and bacteria were resuspended in 15 ml CDM medium without methionine and grown for another 60 minutes in a 30-ml tube. Then rifampicin (500 µg ml−1) was added to the cultures. Before (0) and after 0.5, 2, 5, 10 and 60 minutes of rifampicin exposure, RNA was isolated and Northern blot analyses were performed as described in the Supporting Information (Text S1).
|
10.1371/journal.pcbi.1005311 | Genome-Wide Association between Transcription Factor Expression and Chromatin Accessibility Reveals Regulators of Chromatin Accessibility | To better understand genome regulation, it is important to uncover the role of transcription factors in the process of chromatin structure establishment and maintenance. Here we present a data-driven approach to systematically characterise transcription factors that are relevant for this process. Our method uses a linear mixed modelling approach to combine datasets of transcription factor binding motif enrichments in open chromatin and gene expression across the same set of cell lines. Applying this approach to the ENCODE dataset, we confirm already known and imply numerous novel transcription factors that play a role in the establishment or maintenance of open chromatin. In particular, our approach rediscovers many factors that have been annotated as pioneer factors.
| Transcription factor binding occurs mainly in regions of open chromatin. For many transcription factors, it is unclear whether binding is the cause or the consequence of open chromatin. Here, we used datasets on open chromatin and gene expression provided by the ENCODE project to predict which transcription factors drive transitions between open and closed states. A signature of such a factor is that its expression values are correlated to chromatin accessibility at its motif across the same set of cell lines. Our method assesses this correlation while accounting for the fact that some tested cell lines are more related than others. We find many transcription factors showing evidence of driving transitions and a high proportion of these transcription factors are known pioneer factors, i.e., they play a role in opening up closed chromatin.
| In higher eukaryotes, certain sequence-specific transcription factors (TFs), which we will call chromatin accessibility regulators (CARs), are responsible for establishing and maintaining open chromatin configurations [1,2]. CARs therefore play a fundamental role in transcriptional regulation, because open chromatin configurations are necessary for additional TFs to bind and transcriptionally activate target genes.
CARs that can bind closed chromatin and open up chromatin are called pioneer TFs [3]. The comprehensive identification of pioneer TFs with high confidence still needs further research. While some pioneer TFs are well studied, others have only preliminary evidence, or are only computationally predicted. Some well studied examples include FOXA1, whose winged helix domains disrupt DNA–histone contacts, and POU5F1, SOX2 and KLF4, which are used in production of induced pluripotent stem cells (iPSC) [4,5]. Further pioneer TFs such as ASCL1, SPI1 and the GATA factors are used in transdifferentiation, and PAX7 plays a role in pituitary melanotrope development [5–7]. However, not all pioneer TFs are involved in development and cell type conversions: the CLOCK-BMAL1 heterodimer is part of the circadian clock and the tumour suppressor TP53 is involved in the cell cycle, while its close homolog TP63 is involved in skin development [8–10].
Recent studies suggest that maintaining open chromatin is a dynamic process with pioneer and other TFs binding and unbinding rapidly and continually recruiting additional chromatin remodelling factors that are not sequence specific [2,11,12]. TFs vary in their ability to recruit particular remodelling factors, for example the TFs STAT5A/B and MYOG motifs enrich in binding sites of the SWI/SNF remodelling complex but not in ISWI remodelling complex binding sites, whereas YY1 motifs were found exclusively in ISWI complex binding sites [2]. A natural question then is which TFs are relevant to maintain open chromatin and can therefore be called CARs.
One approach to test whether a given TF is a CAR is to perform a knock-down of this TF followed by an open chromatin assay to see whether chromatin regions containing the respective motif preferentially change from open to closed [13]. However, this approach is very time consuming because it requires a separate knock-down experiment for each TF. To define pioneer TFs specifically, one can check if the TF has the ability to bind nucleosomal DNA in vitro and validate the results in vivo [14]. Recently, a computational method called Protein interaction Quantification (PIQ) has been published that aims to recover pioneer TFs by estimating both TF binding and ensuing chromatin changes from the same Dnase1 hypersensitivity (DHS) experiments [15]. However, PIQ did not predict some well known pioneer TFs such as FOXA1, SOX2 and POU5F1 showing that further improvements are possible [3].
Here we introduce a data driven approach to predict CARs. Our approach relies on the joint analysis of a large collection of DHS and coordinated gene expression data to estimate TF activity independently of DHS data. We first define the motif accessibility score for a given TF for each cell line based on the enrichment of its binding motif in regions with open chromatin. We then associate these scores with gene expression values across all available cell lines. This should allow us to predict which factors have a role either in establishment or maintenance of open chromatin, although it will not reveal which mode predominates (to determine this, further experiments will be necessary).
We used our approach on data generated as part of the ENCODE project [16,17]. This uncovered numerous TFs whose motif accessibility is robustly associated with mRNA expression across 109 cell lines suggesting either a role in the establishment or maintenance of open chromatin. Also, we see that our uncovered TFs are strongly enriched for known pioneer TFs. This suggests that the TFs we identified are good candidates for CARs.
Our approach rests on the assumption that the activity of a CAR is correlated with the amount of open chromatin in the vicinity of its potential binding sites. Both quantities can be estimated from genomic data: For the CAR activity we use its gene expression level as a proxy for the active protein concentration. The effect of this activity is approximated by the open chromatin fraction of the genome around its binding motif instances (Fig 1). Specifically, we count the number of instances of the binding motif of a given TF in the open chromatin fraction of the genome to define a motif accessibility score. A naive approach would be to use standard linear regression between the motif accessibility score and the expression level of a given TF to identify CAR candidates. Yet, this method has an elevated type I error rate, as it does not account for confounding due to cell line relatedness or batch effects. To overcome this limitation, we use here a linear mixed model (LMM) framework, where a random effect accounts for such confounding factors (which has been shown to work well in genetic association studies [18–20]). For a given motif, we use the linear mixed model framework to find the association p-value between its accessibility score and the measured expression of the TF gene. We then compare this p-value to the p-values calculated using the measured expression of each of the other genes as regressors. If confounding is controlled for, most association p-values should follow a uniform [0,1] distribution. Furthermore, if the TF is a CAR, its p-value should be low compared to other genes. We thus define the CAR rank of a TF as the rank of its association p-value among all genes (see example in Fig 1). Low CAR ranks indicate strong association between motif accessibility and TF expression, suggesting that the TF is a CAR.
Specifically, we used DHS data as well as mRNA expression data across 109 cell lines. To calculate motif accessibility scores we used 325 TF binding motifs from the HOCOMOCO database [21]. As expected, we observed severe confounding when using standard linear regression, which was controlled using linear mixed effect model regression (Fig 2).
Our method relies on TF motif accessibility and expression data to predict CARs. However, evolutionarily related TFs have similar binding motifs [23]. Motif accessibility may therefore associate not only with the expression of the annotated TF, but also with the expression of a homologous TF with a similar motif. Therefore, we mapped TFs into subfamilies using the homology-based clustering TFClass [24]. The 1,557 TFs were grouped into 397 subfamilies. Using a collection of 329 ChIP-seq profiles from ENCODE, we saw strong enrichment of TF motifs in ChIP-seq peaks of the TF as well as its subfamily members (Fig 3). We therefore consider any strong association between a motif and a member of the subfamily of its TF as a signal for a CAR.
Next, we used the linear mixed model strategy to predict CARs among TFs. We used 325 motifs from HOCOMOCO (after filtering motifs showing low overlap with DHS signal, see Methods). For each motif, we used a linear mixed effect model to compute its association with mRNA expression for 1,188 known TFs. Due to the redundancy of motifs within the same TF subfamily (see preceding section), we also computed CAR ranks at the level of TF subfamilies. To this end, we retained the most significant association p-value within each subfamily corrected for subfamily size (see Methods and S2 Fig). Under the null model (when TFs are not CARs), CAR ranks should be uniformly distributed across all subfamilies, so that deviation from uniformity indicates presence of CARs.
We found strong enrichment of low CAR ranks at the subfamily level (Fig 4, S1 Table). The enrichment was stronger when using mixed modelling instead of standard linear regression, underlining again the importance of proper control for confounding factors. When looking at the threshold that leads to 10-fold enrichment of low CAR ranks compared to uniformity (i.e., 10% false discovery rate), we found that 25% of all subfamilies have a CAR rank that falls below that threshold. These results show that many TFs do have an impact on the open chromatin fraction and can be defined as CARs.
To validate our results based on the ENCODE dataset, we applied our CAR calling strategy to data from another large scale effort, the ROADMAP Epigenomics consortium [26]. Coordinated open chromatin and expression data have been released for 56 samples. For 29 of these samples, open chromatin was assayed directly. For the other samples, open chromatin information was imputed from other available epigenetic measurements. The ROADMAP collection is derived mainly from human tissue samples and primary cell lines (whereas ENCODE is biased towards immortalized cell lines). Further differences are that expression was measured using RNA sequencing. We applied our method to these datasets and compared results to the results derived in ENCODE. Most subfamilies predicted to be CARs in ROADMAP were recovered in ENCODE (see S3 Fig). Furthermore, while subfamilies predicted to be CARs in ENCODE showed enrichment for low CAR ranks in ROADMAP, subfamilies not predicted to be CARs in ENCODE did not show enrichment for low CAR ranks in ROADMAP (see S4 Fig). These results are concordant with both datasets, pointing toward the same factors being CARs and the higher power of the ENCODE data to detect CARs, potentially due to higher sample size, reliance on direct measurements of DHS and lower fraction of complex tissue samples.
To evaluate the impact of the motif search strategy, we investigated the robustness of the pipeline with respect to the motif search. Results were stable and power was only affected by varying motif cutoffs (S5 Fig, S6 Fig). Additionally, we investigated whether choosing the cutoff based on ChIP-seq data changed results. For each TF with available ChIP-seq data, we used an individual cutoff such that all called binding sites have fixed true positive rate (using the ChIP-seq data as the ground truth). Again, results were stable no matter how the cutoff was assigned (S7 Fig and S8 Fig).
As mentioned above, one well-defined class of CARs are pioneer TFs that can bind and open closed chromatin. Therefore, subfamilies annotated to known pioneer TFs should have low CAR ranks. To test enrichment formally, we used a recently published list of established pioneer TF subfamilies (Methods) [3]. We asked whether these subfamilies were predicted as CARs using our methodology. For eight subfamilies in the list for which we had the motif, six showed at least ten-fold enrichment (i.e. having a CAR rank at the subfamily level below ten) (Fig 5). To assess significance, we used the Wilcoxon ranksum test leading to a p-value of 0.0087. When using the hypergeometric test with 10-fold enrichment cutoff (Fig 4), the p-value was even lower (P = 0.0016). Because our approach to uncover CARs is biased towards TFs with large mRNA expression variability (S9 Fig), we sought to control for potential confounding introduced by the fact that the tested pioneer factors might also have large expression variability. Controlling for expression variability only slightly increased the p-values from 0.0087 to 0.024 and from 0.0016 to 0.0027, respectively.
It is known that the activity of some TFs is mainly regulated by the level of their cofactors rather than their own protein concentration [27]. These TFs are often present in their inactive form in the cell, which can then be quickly activated upon binding of the cofactor. This allows the cell to rapidly respond to environmental cues. An example of this phenomenon are steroid receptor TFs, which initiate transcriptional changes upon steroid hormone binding [28]. In such cases, one would not expect a strong association between the mRNA expression level of a receptor TF and its motif accessibility because mRNA expression would rather be correlated to the amounts of inactive TF protein in the cell, while TF activity should depend on the strength of the environmental stimulus. However, if the TF strongly activates mRNA expression of other genes, it might be possible to predict whether the TF is a chromatin accessibility regulator by looking at associations between the motif accessibility of the TF and the expression of its downstream genes.
To explore this strategy, we looked at associations across all genes and motifs that were below the overall Bonferroni threshold (9.6 x 10−9). For five out of 13 such motifs, members of the corresponding subfamily had top scores. In four further cases, a gene from a TF subfamily was ranked close to the top that was highly related (i.e. part of the same family [24]) to the motifs’ corresponding subfamily but not identical with it. This suggests that the TF subfamily clustering was too fine-grained in these cases. Surprisingly, for one motif, the significant association had a negative effect size (the negative association was observed between NUDT11 and the motif for RARG), which might reflect an indirect effect. The remaining three motifs were all annotated to the GR-like receptors, which encompass four TFs (AR, NR3C1, NR3C2, PGR). The accessibilities of these three motifs all associated strongly with the expression of three genes (FKBP5, ZBTB1, TSC22D3). When using the STRING database to check for functional links between these genes, all genes had links to a GR-like receptor (Fig 6) [29]. In fact, all three genes are known to be glucocorticoid response genes. These results suggest that some GR-like receptors might act as a CAR. For strongly activating factors, the power of the analysis can therefore be strengthened by incorporating results from downstream genes.
It is well known that TF binding correlates with open chromatin [17]. However, for many TFs, it is not clear whether their binding is the cause or the consequence of open chromatin. Here, we used datasets provided by ENCODE to predict chromatin accessibility regulator candidates, i.e., TFs that are able to establish or maintain open chromatin configurations. We devised an approach using linear mixed models to deal with the extensive confounding that one encounters in genome-wide data from heterogeneous sources. Our method uncovers a set of TFs whose expression is associated with their motif accessibility, suggesting a role in maintenance of an open chromatin configuration.
Potentially our methodology could be extended to histone modification data instead of DHS data. We applied our method to H3K4me3 data for cell-lines but did not see strong enrichment (S10 Fig).
Because pioneer TFs are by definition CARs, our predictions should be enriched for known pioneer TFs. We tested this formally for a list of pioneer TF subfamilies recently published by Iwafuchi-Doi et al. [3]. Six out of eight pioneer subfamilies were indeed predicted by our method to be CARs: FOXA1, GATA6, KLF4, SOX2, SPI1 and TP63 were the pioneer TFs driving these signals. The two subfamilies not predicted to be CARs were POU5 and CLOCK. SOX2 was the gene most strongly associated with POU5F1 motif accessibility with a low p-value of 5 x 10−6 (S11 Fig). POU5F1 acts together with SOX2 to maintain undifferentiated states [30]. The two TFs also physically interact and a recent study proposed a model where SOX2 guides POU5F1 to target sites [31]. The CLOCK subfamily members have a role in the cell cycle, acting as TFs for the circadian pacemakers [32]. It is possible that average mRNA expression of these TFs in unsynchronized cell lines is not a meaningful measure for their activity. In addition to the eight aforementioned factors we found further factors discussed in the pioneer TF literature such as TFAP2C, EBF1, CEBPD/B, OTX2, NFKB and STAT5 (Table 1) [22,33–37]. In addition, when combining our predictions with those from the PIQ method[15], we observed substantial performance improvement compared to either method alone (S12 Fig).
One limitation of our approach is that it cannot discern between open chromatin establishing TFs and open chromatin maintaining TFs. A way to discern the relative roles could be to perform overexpression and knock-down experiments followed by an open chromatin assay for the TFs found by our approach. While this is out of the scope for the current study, we hope that our method can help in prioritizing such experimental efforts.
Further, by its very nature, our methodology cannot with certainty resolve between TFs that belong to the same sub-family. It shares this weakness with almost any method relying on TF motifs. The procedure associates the expression values of each TF separately to the motif accessibilities and one strong association is enough to lead to low CAR ranks for the subfamily. The TF in the subfamily whose expression is the most strongly associated to one of the subfamily motif is naturally also the strongest candidate for CAR activity. (This information is given in Table 1 as well as in S1 Table). However, if the expression values of the subfamily members are also strongly correlated, we cannot be sure which ones are driving the association.
It is also clear that multiple conditions have to be met for the approach to work. First and foremost, mRNA expression has to be correlated sufficiently with protein concentration of the CAR. Typically, only a fraction of the variation in protein concentration can be explained by variation in mRNA abundances [39]. Nevertheless, better power of our approach can always be achieved by increasing sample size, as long as there is at least some correlation. Further, it is reasonable to assume that our approach will perform better on TFs with a large dynamic range across cell types. This seems indeed to be the case, since most TFs predicted to be CARs tend to have large mRNA expression variance (S9 Fig). Sampling more and diverse cell lines could address this issue, because it should increase the dynamic range.
This restriction would also suggest that our approach is biased against cell type specific TFs. However, when looking at tissue expression patterns (www.gtexportal.org [40]) of the predicted CARs, we found both: TFs that showed expression in a large proportion of cell lines such as EBF1 and STAT5B as well as quite specific TFs. Examples of specific CARs are SPI1, which only showed expression in whole blood, and OTX2, which only showed expression in some brain regions. It is possible that the use of immortalized cell lines leads to larger gene expression variability in the sample facilitating the detection of such tissue-specific CARs.
For some TFs, activity mainly depends on cofactors. For example, for steroid hormone receptors, hormone molecules activate a pool of inactive TF already present in the cell. In such cases measuring TF activity with gene expression measures can be misleading and one would not expect an association between the expression of a TF and the accessibility of its motif. For example, for the accessibility score of NR3C1, we saw much stronger associations with the expression levels of a small set of glucocorticoid response genes (ZBTB16, FKBP5, TSC22D3) than that of NRC1 itself [41–43]. This difference in signal strength is in line with the activity of NR3C1 being mainly regulated by glucocorticoid binding and not NR3C1 gene expression levels. Of note, NR3C1 was reported to have pioneer activity [1].
In summary, we exploited the rich data source of ENCODE to find TFs whose mRNA expression levels are directly linked to the open chromatin fraction of the genome. Although our approach in its current form is able to find TFs with strong associations, it is also clear that increasing power by adding more cell lines would find more TFs with an association. From the current data, we would estimate that at least 25% of TF subfamilies show a low CAR rank at the subfamily level, suggesting that the regulation of chromatin accessibility is a pervasive phenomenon amongst TFs.
Annotated open chromatin (FDR <0.01) peaks were downloaded from the EBI website (see URL section) and trimmed to the top 90,000 peaks for each cell line. 426 motifs were downloaded from the HOCOMOCO website and aligned to the reference genome with FIMO [21,44]. Motif occurrences with a p-value below 10−5 were kept for processing. For each motif, we counted the number of DHS peaks overlapping a motif instance in a given cell line using bedops [45]. Results were filtered to motifs that were present in at least 150 DHS peaks on average, leaving 344 motifs. For a given motif, we quantile-normalized the values to follow a normal distribution yielding the raw motif-activity matrix with rows corresponding to motifs and columns corresponding to cell lines. The resulting matrix was iteratively scaled to zero mean and unit standard deviation, first row-wise (across cell lines) then column-wise, until convergence [46,47]. Next, we saw that the cell-line wise covariance matrix had a very large first eigenvalue, with a corresponding eigenvector that did not track well the different tissue origins of the various cell lines. Assuming that this leading principal component largely captured batch effects, we chose to regress out the first eigenvector from each row of the matrix, leading to better agreement between expression and motif accessibility correlation matrices (S13 Fig). After this step, we quantile-normalized the data per motif to follow a normal distribution to ensure that the assumptions of the applied statistical model were met. To map motifs to TFs and TF subfamilies, we used the TfClass hierarchy [24]. Of the 344 tested motifs, we mapped 330 to a TF and its subfamily. Of these, 325 had expression data available for a subfamily member.
We downloaded raw expression microarray data from the GEO repository (GSE1909 and GSE15805). (ENCODE micro-array data was used instead of RNA-seq because to-date more cell lines with DHS information have also RNA expression measured by micro-array than RNA-seq). We background corrected and normalized using the RMA-algorithm implemented in the oligo package to process all arrays for which DHS data was also available [48,49]. Only the core set data was used. The data were summarized to gene level [50]. Only results that had a one-to-one mapping between genes and gene probesets were kept. 15,119 genes could be annotated in this fashion. Because for many cell lines more than one experiment was conducted, we summarized multiple plates by averaging gene results across experiments. The resulting matrix was iteratively scaled to zero mean and unit standard deviation, first row-wise (across cell lines) then column-wise, until convergence [46,47].
The model proposed is
y=xiβi+δi+εi.
Where y is a vector of motif accessibility scores across n cell lines, xi is the expression vector of gene, i, βi is the effect size of gene i:
εi∼Nn(0,σr2In)
and
δi∼Nn(0,σe2Ce).
Ce is the covariance matrix of the n x p expression matrix:
Ce=1p∑i=1pxixiT.
For each gene i, βi, σr, and σe are estimated via maximum likelihood and the null hypothesis βi = 0 is tested via a likelihood ratio test [18,20]. More details on this procedure are given in S1 Appendix.
For each motif in HOCOMOCO, we used the mixed model association results across all 1,188 known TFs for which we had mRNA expression data [21]. This yielded a matrix of association p-values for all pairs of 325 motifs (belonging to 147 TFClass subfamilies) and 1188 TFs (belonging to 368 TFClass subfamilies). Due to the fact that homologous TFs have similar binding motifs, we sought to aggregate results into CAR ranks at the subfamily level (S1 Fig). To achieve this, we reduced the 325 x 1188 motif-TF association matrix to a 147 x 368 matrix of associations between motif subfamilies and TF subfamilies. In practice, for each motif subfamily-TF subfamily pair we collected the most significant p-value among all motif-TF pairs in these subfamilies and multiplied it with the total number of such motif-TF pairs to correct for subfamily size. Finally, for each motif subfamily, we ranked the adjusted p-values across all TF subfamilies and defined its CAR rank as the rank of its corresponding TF subfamily.
To get an external annotation of pioneer factors, we used a recently published list of established and predicted pioneer factors (Table 1 in Iwafuchi-Doi et al. [3]). We used a hypergeometric test at the 10-fold enrichment cut-off (Fig 4), as well as a ranksum enrichment test. To derive a ranksum statistic, we summed the CAR ranks of the eight subfamilies annotated as pioneers. To assess significance of this statistic, we used permutation tests: For each of the 50,000 permutation samples, we picked eight CAR ranks from the set of subfamilies not annotated as pioneers and summed them to derive 50,000 permutation sample statistics. The p-value was approximated as the fraction of permutation sample statistics of greater or equal size as the statistic derived for the annotated pioneers. To control pioneer enrichment for mRNA expression variation, we first calculated the expression variance of each TF across all cell lines. The distribution of variance values was transformed to follow a standard normal distribution. We then used the maximal expression variance observed for any TF in each subfamily. To assess significance, we used permutation tests: we sampled eight non-pioneer subfamily level CAR ranks 50,000 times. However, subfamilies were not sampled uniformly: We sampled four non-pioneer subfamilies with maximal expression variance between the 0th and the 50th quantile of the eight pioneer subfamilies, and four non-pioneer subfamilies with maximal expression variance between the 50th and the 100th quantile of the eight pioneer subfamilies.
RNA-seq data were downloaded from the ROADMAP website (see section ‘URLS’) for 56 cell lines. We used only genes with average read count above 50, which removed 12% of genes. The number of reads plus a pseudo-count of one to were log-transformed. Samples were then quantile normalized to the average mean distribution [51]. The resulting matrix was iteratively scaled to zero mean and unit standard deviation, first row-wise (across cell lines) then column-wise, until convergence [46,47].
To derive motif accessibility scores, imputed DHS data were downloaded for 56 cell lines from the ROADMAP website (see section ‘URLs’). From these datasets motif accessibility scores were derived in the same fashion as for the ENCODE DHS data. To derive CAR ranks, the same strategy was employed as for the ENCODE dataset.
To compare fixed motif cutoffs to a variable motif cutoff guided by ChIP-seq, the following procedure was used. ChIP-seq data from the Myers and Snyder lab in the ENCODE collection for which dnase1 and expression data were available were downloaded and each ChiP-seq experiment was mapped to a dnase1 experiment based on cell line and to the motif of the TF, yielding mappings to 75 motifs (belonging to 50 subfamilies). For a given motif and cell line pair for which ChIP-seq data (as well as DHS data) was available, each DHS region was annotated with the p-value of its most significant motif instance (given that they contained a motif with p-value below 5x10-5) as well as whether it overlapped with a ChIP-seq peak. The motif p-value cutoff was defined such that a fixed fraction of peaks with motifs below that cutoff would validate in the ChIP-seq experiment. Three true positive rates were chosen for this comparison 0.3, 0.5 and 0.7 (see S7 Fig, S8 Fig). Only experiments were used for which it was possible to choose a motif cutoff such that the highest validation rate (i.e. 0.7) could be reached. If multiple ChIP-seq experiments were available per motif, the median p-value cutoff was chosen for each validation rate. We compared these strategies using a fixed cutoff for all motifs of 10−5, which was used throughout the rest of the paper. Results obtained are similar when using ChIP-seq guided cutoffs or fixed cutoffs.
Code for reproduction (including scripts for data download) is available at: https://github.com/dlampart/csrproject
ENCODE DHS peaks were downloaded from: http://ftp.ebi.ac.uk/pub/databases/ensembl/encode/integration_data_jan2011/byDataType/openchrom/jan2011/fdrPeaks/
ROADMAP expression data were downloaded from: http://egg2.wustl.edu/roadmap/data/byDataType/rna/expression/57epigenomes.N.pc.gz
ROADMAP imputed DHS peaks were downloaded from: http://egg2.wustl.edu/roadmap/data/byFileType/peaks/consolidatedImputed/narrowPeak/
ENCODE histone files were downloaded from: ftp://hgdownload.cse.ucsc.edu/goldenPath/hg19/encodeDCC/wgEncodeUwHistone/
|
10.1371/journal.pntd.0001399 | Dynamics of Th17 Cells and Their Role in Schistosoma japonicum Infection in C57BL/6 Mice | The current knowledge of immunological responses to schistosomiasis, a major tropical helminthic disease, is insufficient, and a better understanding of these responses would support vaccine development or therapies to control granuloma-associated immunopathology. CD4+ T cells play critical roles in both host immune responses against parasitic infection and immunopathology in schistosomiasis. The induction of T helper (Th)1, Th2 and T regulatory (Treg) cells and their roles in schistosome infections are well-illustrated. However, little in vivo data are available on the dynamics of Th17 cells, another important CD4+ T cell subset, after Schistosoma japonicum infection or whether these cells and their defining IL-17 cytokine mediate host protective responses early in infection.
Levels of Th17 and the other three CD4+ T cell subpopulations and the cytokines related to induction or repression of Th17 cell generation in different stages of S. japonicum infection were observed. Contrary to reported in vitro studies, our results showed that the Th17 cells were induced along with the Th1, Th2, Treg cells and the IFN-γ and IL-4 cytokines in S. japonicum infected mice. The results also suggested that S. japonicum egg antigens but not adult worm antigens preferentially induced Th17 cell generation. Furthermore, decreasing IL-17 with a neutralizing anti-IL-17 monoclonal antibody (mAb) increased schistosome-specific antibody levels and partial protection against S. japonicum infection in mice.
Our study is the first to report the dynamics of Th17 cells during S. japonicum infection and indicate that Th17 cell differentiation results from the integrated impact of inducing and suppressive factors promoted by the parasite. Importantly, our findings suggest that lower IL-17 levels may result in favorable host protective responses. This study significantly contributes to the understanding of immunity to schistosomiasis and may aid in developing interventions to protect hosts from infection or restrain immunopathology.
| Th17 immune cells secrete the IL-17 cytokine and contribute to host defenses against certain infections. Recent studies linked IL-17 with the severity of liver inflammation and suggested that Th17 cells contribute to the pathology in schistosomiasis, a serious disease caused by parasitic worms such as Schistosoma japonicum widespread in vertebrates including humans. However, the role of Th17 cells in protection against S. japonicum infection is still unclear. For the first time, we describe here the changes in Th17 cell levels during S. japonicum infection and suggest that the schistosome egg antigens are primarily responsible for stimulating the generation of host Th17 cells after S. japonicum infection. We further show that the level of Th17 cells in the host is determined by a combination of factors, namely exposure to complex parasitic antigens that either induce or suppress their generation. We also suggest that lowering IL-17 levels may favor the host's protective responses against S. japonicum infection. Our findings help to better understand the relationship between the host and parasite in terms of immune protection and pathology in schistosomiasis and may contribute to the future development of vaccination and therapeutic strategies.
| CD4+ T cells play an important role in the initiation of immune responses against an infection by providing help to other cells and by taking on a variety of effector functions during immune reactions. Upon antigenic stimulation, naive CD4+ T cells activate, expand and differentiate into different effector subsets termed T helper (Th) 1 and Th2 cells. The appropriate induction and balance between Th1 and Th2 cellular responses to an infectious agent can influence both pathogen growth and immunopathology [1]. Th17 cells recently emerged as a third independent effector cell subset differentiated from CD4+ T cells upon antigenic stimulation [2]–[5]. Although the functions of these cell subtypes are not completely understood, emerging data suggest that by producing their defining cytokine IL-17, Th17 cells play an important role in host defenses against extracellular pathogens, such as Klebsiella pneumoniae [6], Pseudomonas aeruginosa [7], Porphyromonas gingivalis [8] and Bacteroides fragilis [9], which are not efficiently cleared by Th1-type and Th2-type immunity. Meanwhile, several studies have shown that Th17 cells and IL-17 also play important roles in immunopathology in some infectious diseases, such as pulmonary tuberculosis [10], toxoplasmosis [11] and schistosomiasis [12]–[17].
CD4+ T cells can also be induced to differentiate into CD4+CD25+ T regulatory (Treg) cells with immunosuppressive activities that down-regulate immune responses, thereby inhibiting immunopathology while promoting parasite survival via direct repression of the induction and responses of the other CD4+ subsets, Th1, Th2 and Th17 cells [18]–[22].
Since the functional analysis of IL-17 produced by Th17 cells has suggested an important and unique role for this cytokine in both host protection against specific pathogens and immunopathologic damage to the host, much of the research focus has been placed on the factors that either positively or negatively regulate differentiation of Th17 cells. To date, several studies have shown that Th17 cells require specific cytokines for their differentiation, different from those for Th1 and Th2 cells. A combination of TGF-β plus IL-6 was recently described to be essential for initial differentiation [23]–[27], IL-21 for the amplification [28], [29] and IL-23 for the subsequent stabilization [25], [30], [31] of the Th17 cell subset. On the other hand, both high levels of Th1 and Th2 cells and their respective cytokines, IFN-γ and IL-4, antagonize the development of Th17 cells [2], [4], [5]. Additionally, in the absence of IL-6, TGF-β alone is clearly favored as the cytokine for differentiation of Treg cells while suppressing the differentiation of Th17 cells [4], [32]. These findings suggest an intimate link between the Treg and Th17 cell programs of differentiation. However, thus far the notion that CD4+ T cell subsets represent distinct terminally differentiated lineages has been favored on the basis of a series of in vitro experiments, and the suppression of Th17 differentiation by Th1, Th2 and Treg cells and/or their cytokines has been demonstrated in numerous in vitro studies or under certain simplified or defined conditions [25], [27], [32]–[34]. However, there is very little in vivo data available to support such a cross-regulation between Th17 cell differentiation and Th1, Th2 and Treg cells during multicellular pathogenic infection.
Schistosomiasis, a major neglected tropical helminthic disease infecting 200 million people with an estimated 600 million at risk worldwide, is an excellent model for studying the induction and regulation of differentiation of the various CD4+ T cell subsets in response to infection. Infection of Schistosoma japonicum, a multicellular parasite which has an extremely diverse repertoire of antigens, induces the production of bulk cytokines to induce Th1, Th2 and Treg cells that play important roles in the immune response to infection. In particular, a recently growing number of studies have indicated that IL-17, a CD4+ T cell-derived cytokine, is most directly associated with the severity of hepatic granulomatous inflammation [12]–[16], [35]–[37], suggesting that IL-17-producing T cells are a major force behind severe pathology in schistosomiasis. During schistosome infection, the immune response progresses through at least three phases. (1) During the first three weeks of the infection, when the host is exposed to migrating immature and mature parasites, the dominant response is Th1-like. The response is induced by non-egg antigens, such as the schistosomula and soluble worm antigen (SWA) [38], [39]. (2) As the parasites begin to produce eggs (beginning 4–5 weeks post-infection), the response alters, with the emergence of a stronger Th2 response which is primarily induced by egg antigens [38], [40]. The granulomas that form around the eggs in the liver, which are reported to be positively regulated by Th17 cells and the secreted IL17 cytokine [12]–[16], [36], develop to their maximum size around 8–9 weeks post-infection. (3) During the chronic phase of infection (beginning 11–13 weeks post-infection), the Th2 response is predominant and modulated. The granulomas are also smaller than at earlier times. At this stage, CD4+CD25+Foxp3+ Treg cells are believed to be induced mainly by egg antigens and play an important repressor role in down-regulation of pathologic immune responses [20]. In addition, a recent study reported elevated Th17 levels in response in vaccination against S. mansoni infection in C57BL/6 mice [41]. However, there is very little data available showing the dynamics of Th17 cells after S. japonicum infection as well as whether Th17 cells/IL-17 mediate the host protective responses at the early stage of S. japonicum infection. In the present study, we observed the changes in Th17 cell levels at different stages of S. japonicum infection and investigated the role of IL-17 in the host protective responses.
Animal experiments were performed in strict accordance with the Regulations for the Administration of Affairs Concerning Experimental Animals (1988.11.1), and all efforts were made to minimize suffering. All animal procedures were approved by the Institutional Animal Care and Use Committee (IACUC) of Nanjing Medical University for the use of laboratory animals (Permit Number: NJMU 07-0137).
Female 8-week old C57BL/6 mice were purchased from SLAC Laboratory (Shanghai, China) and bred in university facilities. All animal experiments were performed in accordance with the Chinese laws for animal protection and in adherence to experimental guidelines and procedures approved by the Institutional Animal Care and Use Committee (IACUC), the ethical review committee of Nanjing Medical University, for the use of laboratory animals. Oncomelania hupensis harboring S. japonicum cercariae (Chinese mainland strain) were purchased from the Jiangxi Institute of Parasitic Diseases (Nanchang, China).
For kinetic analysis of T cell populations and cytokines, each mouse was infected with 12 cercariae of S. japonicum through the abdominal skin. At 3, 5, 8 and 13 weeks post-infection, four mice were randomly chosen from the infected and normal control groups and sacrificed for further study. For challenge experiments, each mouse was infected with 40 cercariae of S. japonicum by abdominal skin exposure.
S. japonicum SWA were prepared by harvesting the soluble fraction obtained from sonicated S. japonicum adult worms as previously described [42]. S. japonicum eggs were extracted from the livers of infected rabbits and enriched. The S. japonicum soluble egg antigens (SEA) were then prepared from the homogenized eggs as previously described [43]. SWA and SEA were diluted with PBS to a final concentration of 10 mg/ml for immunization.
Three independent experiments were carried out in the same manner. In each experiment, C57BL/6 mice were divided randomly into three groups (two test and one control) consisting of eight mice per group. Each mouse was injected subcutaneously in the back with 100 µl of a solution containing 50 µg of SEA, 50 µg of SWA or PBS emulsified in incomplete Freund's adjuvant (IFA, Sigma-Aldrich, ST. Louis, MO) [44]. Each mouse was immunized two times with a 14-day interval. Two weeks after the last immunization, serum samples were collected, and mice were sacrificed for further study.
Single cell suspensions of splenocytes and lymphocytes were prepared by mincing the mouse spleens and mesenteric lymph nodes in PBS containing 1% FBS (Gibco, Grand Island, NY) and 1% EDTA. Red blood cells were lysed using ACK lysis buffer.
For preparation of single cell suspensions of hepatic lymphocytes, mouse livers were perfused via the portal vein with a PBS/heparin mixture (75 U/ml, Sigma Chemical Co., St. Louis, MO). The excised liver was cut into small pieces and incubated in 10 ml of digestion buffer (collagenase IV/dispase mix, Invitrogen Life Technologies, Carlsbad, CA) for 30 min at 37°C. The digested liver tissue was then homogenized using a MediMachine with 50 µm Medicons (Becton Dickinson, San Jose, CA) for 3 min at low speed [45]. The liver suspension was then centrifuged at low speed to sediment the hepatocytes. The remaining cells were separated on a 35% Percoll gradient by centrifuging at 600×g. The lymphocyte fraction was resuspended in 2 ml of red cell lysis buffer and then washed in 10 ml of complete RPMI 1640 with 0.1 M EDTA.
The cells were cultured in triplicate in complete RPMI 1640 medium (Gibco) containing 10% FBS, 2 mM pyruvate, 0.05 mM 2-mercaptoethanol, 2 mM L-glutamine, 100 U of penicillin/ml and 0.1 mg/ml streptomycin. Subsequently, 2×105 cells per well in 200 µl of complete media were cultured in 96 well plates (Nunc, Roskilde, Denmark) for 72 h at 37°C in the presence of 25 ng/ml phorbol 12-myristate 13-acetate (PMA) and 1 µg/ml ionomycin (Sigma-Aldrich) [46]–[48]. Alternatively, in some experiments, the cells from S. japonicum infected mice were stimulated with or without 50 µg/ml of SEA for 48 h. Culture supernatants were collected for ELISA after incubation.
Cytokines in the culture supernatant were analyzed using mouse cytokine multiplex assay kits for detecting IL-6 and IL-21 (R&D Systems, Inc. Minneapolis, MN) and for detecting TGF-β and IL-23 (Bender MedSystems, Novato, CA). IL-17A, IFN-γ and IL-4 levels in the supernatant were measured by ELISA using the eBioscience ELISA Ready-SET-Go kit (eBioscience, San Diego, CA) according to the manufacturer's protocol.
The SWA and SEA specific IgG, IgG1 and IgG2a antibodies in mouse serum samples were detected by standard ELISA using the SWA and SEA as the coated antigen [42], [43]. HRP-conjugated rat anti-mouse IgG (Calbiochem, Darmstadt, Germany), IgG1 and IgG2a monoclonal antibodies (mAbs) (BD Pharmingen) were used. In brief, ELISA plates (Titertek Immuno Assay-Plate, ICN Biomedicals Inc., Costa Mesa, CA) were coated with 0.1 mg/ml of SEA or SWA in 50 mM carbonate buffer (pH 9.6) and incubated overnight at 4°C. Plates were washed three times with PBS (pH 7.6) containing 0.05% Tween-20 (PBS-T) and blocked with 0.3% (w/v) bovine serum albumin (BSA) in PBS for 1 h at 37°C. The plates were further washed three times with PBS-T and then incubated with the sera diluted with 0.3% BSA (1∶100) at 37°C for 1 h. The plates were washed four times with PBS-T, followed by incubation with HRP-conjugated rat anti-mouse IgG, IgG1 and IgG2a (1∶1000) for 1 h at 37°C. The plates were then washed five times with PBS-T and developed with tetramethylbenzidine (TMB) substrate (BD Pharmigen) for 30 min. The optical density (OD) of the color developed in the plate was read at 450 nm using a BioRad (Hercules, CA) ELISA reader.
For detection of Th17, Th1 or Th2 cells, single cell suspensions of splenocytes, lymphocytes or liver cells from each mouse were prepared, and 1×106 cells from each sample were stimulated with 25 ng/ml PMA and 1 µg/ml ionomycin (Sigma-Aldrich) in complete RPMI 1640 medium in the presence of 0.66 µl/ml Golgistop (BD Biosciences PharMingen) for 6 h at 37°C in 5% CO2. After 6 h, the cells were collected and surface stained with anti-CD3-APC (eBioscience) and anti-CD4-FITC (eBioscience). Subsequently, the cells were washed, fixed, permeabilized with Cytofix/Cytoperm buffer (BD PharMingen) and intracellularly stained with PE conjugated antibodies against IL-17A, IFN-γ or IL-4 (or isotype IgG2a control antibody) (eBioscience) for detection of Th17, Th1 or Th2 cells, respectively, according to the manufacturer's protocol and analyzed with a FACS Calibur flow cytometer. Cells were gated on the CD3+CD8− population for analysis of Th17, Th1 or Th2 cells.
For detection of Treg cells, the Mouse Regulatory T Cell Staining Kit (eBioscience) was used. A single cell suspension of splenocytes from each mouse was prepared, and 1×106 cells were surface stained with anti-CD3-PerCP mAbs (eBioscience), anti-CD4-FITC mAbs and anti-CD25-APC mAbs, followed by fixation and permeabilization with Cytofix/Cytoperm and intracellular staining with anti-Foxp3-PE or IgG2a-PE rat immunoglobulin control antibody, according to the manufacturer's protocol. Cells were gated on the CD3+CD4+ population for analysis of Treg cells.
The recombinant mouse IL-17A (rmIL-17A), the neutralizing rat anti-mouse IL-17A mAb (clone 50104) and its control IgG2a mAb were purchased from R&D Systems, Inc. Two independent experiments were carried out in the same manner. In each experiment, 16 mice were randomly assigned in four groups (four mice per group). Each mouse was challenged with 40 cercariae of S. japonicum as described above. For two groups of mice, 70 µg of mAb or its control IgG2a mAb per mouse were administered intraperitoneally (i.p.) four days before S. japonicum infection, and the administration of the same dose of mAb was repeated every four days during the infection until two days before the mice were sacrificed [12]. Simultaneously, for the other two groups of mice, 500 ng/mice of rmIL-17A or PBS were administered i.p. two days before S. japonicum infection and repeated every 48 h during the infection until two days before the mice were sacrificed [49]. Forty-two days after the challenge infection (two days after the last injection of rmIL-17A or anti-IL-17A), all four mice in each group were sacrificed. Serum samples were collected for ELISA detection of the levels of SEA or SWA specific antibodies, and the livers were isolated for histopathological examination. The splenocytes were prepared for incubation as previously mentioned for detection of cytokines in the culture supernatant or intracellular staining for detection of Th17, Th1, Th2 and Treg cells.
Forty-two days after the challenge, all mice injected with neutralizing mAb or rmIL-17A and their controls were sacrificed, and perfusion was performed with saline containing heparin to recover the adult worms. Two grams of each liver were digested with 5% KOH at 37°C overnight, and the numbers of eggs were determined by microscopic examination. The remaining parts of the livers were dissected and immediately fixed in 10% buffered formalin. Liver sections were embedded in paraffin and stained with hematoxylin and eosin (H&E) for microscopic examination. The lesions were assessed on coded slides by an observer unaware of the experimental setting. The sizes of the granulomas were measured by computer-assisted morphometric analysis as previously described [44], and 50 visual fields in the liver section of each mouse (ten sections for each mouse and five random microscope fields for each section) were measured under a microscope (magnification: 100×) (Olympus, Tokyo, Japan). Granuloma sizes are expressed as means of areas measured in µm2 ± SD. The percentages of neutrophils, eosinophils, lymphocytes and macrophages in the same granulomas were determined by microscopic examination (1000× magnification) of 200 randomly selected cells (not including hepatocytes) in each granuloma. Ten sections for each mouse and five microscope fields for each section were counted. Percentages of cells were calculated from microscopic analysis of the same granulomas analyzed for lesion size [14], [50], [51].
The worm/egg reduction rate (percentage of protection) was calculated according to the following formula: (1 - mean of worms/eggs in injected mice/mean of worms/eggs in control mice)×100% [52].
Statistical analysis was performed using the SPSS version 10.1 (Statistical Package for Social Sciences, Chicago, IL) software. Statistical significance was determined by Student's t-test and P<0.05 was considered significant.
Consistent with previous studies, the granulomas began to form from five weeks after infection in the mouse liver after egg deposition and continued to develop (Fig. 1A). As shown in Fig. 1B and 1C, in parallel with the development of the granulomas, the proportion of Th17 cells in splenic CD4+ T cells increased very slowly during the first five weeks post-infection compared to that before infection (week 0) and increased rapidly thereafter. Meanwhile, the proportion of the Treg cells in the total splenic CD4+ T cell population showed a continuous increase after infection. Additionally, the proportions of both Th1 and Th2 cells in CD4+ T cells also increased. During the first three weeks post-infection, the proportion of Th1 cells rose much more quickly than that of the Th2 cells. However, after egg deposition, the number of Th2 cells kept increasing rapidly, while the number of Th1 cells reached a plateau by eight weeks post-infection (Fig. 1B and 1C). Compared to the CD4+T cells responses in the spleen, that of the mesenteric lymphocytes showed a more rapid increase of Th17 cells during the first three weeks post-infection and a weaker Th1 response throughout infection (Fig. 1D and 1E). Meanwhile, stronger Th2 but weaker Treg responses were observed in the liver (Fig. 1F and 1G). These results indicated that all of the CD4+ T cell subsets (Th17, Th1, Th2 and Treg cells) increased as over the course of infection.
To further investigate the kinetics of cytokines which affect the differentiation, development and proliferation of Th17 cells during S. japonicum infection in C57BL/6 mice, the splenocytes of infected mice were cultured, and the cytokine levels in the supernatants were detected by ELISA. The results in Figure 2 show that, consistent with the generation of Th17 cells, the level of IL-17 increased very slowly in the first five weeks post-infection compared to that before infection (week 0). However, IL-17 increased rapidly after five weeks post-infection. Meanwhile, both the inducing cytokines (TGF-β, IL-6, IL-23 and IL-21) and the inhibitory cytokines (IFN-γ and IL-4) of Th17 cell generation all increased after infection. Taken together, these results suggest that the generation of Th17 cells during infection with S. japonicum may occur as a net effect of the inducing and inhibitory factors.
Schistosome parasitic worms are multicellular pathogens which have three different life cycle stages (schistosomula, adult worm and egg) in definitive hosts including humans. Among the multitude of schistosome antigens that stimulate host immune responses, the adult worm and egg are two important sources of antigens that are involved induction of different types of Th cell responses or Tregs at different infection stages. Studies show that Schistosoma mansoni eggs induce egg antigen-specific Th17 responses and contribute to the severe immunopathology in murine schistosomiasis [12]–[17]. Consistent with these studies, our data also suggest that the S. japonicum egg antigens have the ability to significantly induce egg antigen-specific Th17 responses (Figure S1A, S1B and S1C). To further investigate the roles of these two major types of S. japonicum antigens on Th17 cell generation, SWA and SEA were used to immunize C57BL/6 mice or to induce CD4+T cells to differentiate in vitro. As shown in Figure 3A and 3B, a significantly higher percentage of Th17 cells was only observed in the SEA immunized group by FACS analysis, suggesting that repeated vaccinations of mice with SEA, instead of SWA, preferentially induced Th17 cells in vivo. Furthermore, the data also suggests that the eggs produced by adult worms in hosts, compared to the adult worms themselves, may more rapidly induce Th17 cells during S. japonicum infection. Meanwhile, additionally, SEA preferentially induced a significant increase of Th2 cells and Tregs, while SWA preferentially induced a significant increase of Th1 cells and caused only a slight increase of Treg and not of Th17 or Th2 cells (Figure 3A and 3B). The profiles of CD4+ T cells differentiation induced by SWA or SEA stimulation in vitro also confirmed the above findings (Figure S2A and S2B).
To further investigate the cytokines that were reported to affect the differentiation, development and proliferation of Th17 cells, splenocytes from mice after vaccination with SWA or SEA were isolated and cultured as described in Materials and Methods, and the levels of cytokines in the supernatants were detected by ELISA. As shown in Figure 4, compared to the SWA and PBS control groups, a significantly higher level of IL-17 was observed in the SEA group, suggesting that repeated vaccination with SEA preferentially induced the production of IL-17. Compared to the PBS control group, the increase of IL-4 was mainly observed in the SEA group, while the increase of IFN-γ was only observed in the SWA group. Compared to the PBS control group, the levels of TGF-β, IL-6, IL-23 and IL-21, which are associated with the generation of Th17 cells, were significantly increased in both the SEA and SWA groups. However, when comparing the SWA group with the SEA group, the data showed that repeated vaccination with SEA preferentially induced higher levels of IL-23 and IL-21, and further supports that egg antigens are possibly more important in the increase of Th17 cells during S. japonicum infection.
To evaluate the role of IL-17 in the host protective responses against S. japonicum infection, C57BL/6 mice were injected with rmIL-17A or anti-IL-17A neutralizing mAb to increase or decrease the level of IL-17 in vivo, respectively, and then challenged with S. japonicum. The protection was measured by the reduction in the worm and egg burden [52] compared between groups injected with rmIL-17A or anti-IL-17A neutralizing mAb and their respective controls. Compared to the control IgG2a mAb group, injection of mice with neutralizing anti-IL-17A mAb led to a 26.61% reduction (P<0.01) in worm burden (Fig. 5A). On the other hand, compared to the PBS control group, no reduction of worm (P>0.05) or egg (P>0.05) burden was observed in the rmIL-17A group.
Consistent with other reports [12]–[16], [36], our results in Figure 5B and 5C also show that elevating IL-17 in vivo by injecting mice with rmIL-17A led to slightly enhanced hepatic granulomatous inflammation (without statistical significance), while decreasing the level of IL-17 by use of an anti-IL-17A neutralizing mAb led to decreased hepatic immunopathology. The percentages of neutrophils and eosinophils in the granulomas, which are thought to be the important populations [14], were increased when S. japonicum infected mice were injected with rmIL-17A. However, injection with an anti-IL-17A neutralizing mAb led to decreases in the percentages of neutrophils and eosinophils in the granulomas (Fig. 6). In addition, the results in Figure 6 also show that rmIL-17A decreased while anti-IL-17A neutralizing mAb increased the percentages of lymphocytes and macrophages in S. japonicum infected mice.
To further investigate the possible mechanism underlying the effects of IL-17 on the response to anti-schistosome infection and the hepatic immunopathology, we detected the levels of CD4+T cells, cytokines and antibody responses in mice after administration of rmIL-17A or anti-IL-17A neutralizing mAb. The proportions of Th1/Th2/Th17/Treg cells did not change after injection with either rmIL-17A to increase the level of IL-17 or anti-IL-17A neutralizing mAb to decrease the level of IL-17 in vivo (Fig. 7A and 7B). The production of IFN-γ, IL-4, IL-6, IL-23, IL-21, TGF-β and IL-17 from splenocytes of S. japonicum infected mice increased after injection with rmIL-17A (Fig. 7C). However, injection of S. japonicum infected mice with anti-IL-17A neutralizing mAb resulted in the increase of IFN-γ, IL-4, IL-6 and IL-21, but the decrease of TGF-β and IL-17 produced by mouse splenocytes. When compared to the PBS control group, administration of rmIL-17A statistically significantly decreased the level of SEA specific IgG1 antibody but increased the level of SWA specific IgG2a antibody in S. japonicum infected mice (Fig. 7D and 7E). However, when compared to the isotype antibody control group, administration of anti-IL-17A neutralizing mAb statistically significantly increased the levels of SEA specific IgG1, IgG2a and total IgG antibodies, as well as SWA specific IgG2a antibody in S. japonicum infected mice.
During the past several years, the Th1/Th2 paradigm has been updated to include a third helper subset called Th17. Through the induction of chemokines and the recruitment of other effector T cell populations [53]–[55], the responses of Th17 cells dominate in response to certain defined pathogens and play important roles in both host defense against pathogens and immunopathogenesis [11], [56]. Schistosomiasis is a typical chronic infectious disease. Infection induces the generation of Th1, Th2 and Treg cells, as well as Th17 cells that are involved in the formation of hepatointestinal perioval granulomas. In this study, we investigated the kinetics of the generation of Th17 cells induced by parasite antigens from different stages of S. japonicum infection in mice as well as the role of Th17 cells in the host protective responses.
In our study, as the parasites began to produce eggs, the granulomas formed in the mouse liver and developed continuously. After the eggs were deposited into the liver and the granulomas were beginning to form, the proportion of Th17 cells in the spleen, mesenteric lymph nodes and liver CD4+ T cells increased slowly up to five weeks post-infection but then increased more rapidly between five and eight weeks post-infection while accompanied by the development of the granulomas. Meanwhile, the proportions of Th1, Th2 and Treg cells in the CD4+ T cells also increased. These findings suggested that the schistosomal antigens induced the simultaneous generation of both Th17 cells and the other CD4+ subsets that are thought to suppress the generation of Th17 cells during infection as reported in many previous studies [2], [5], [32]; however, these factors seem to have failed to suppress the generation of Th17 cells in our S. japonicum infection experiments. In addition, the results also suggested it may be the egg antigens of S. japonicum that were responsible for the more rapid increase in the proportion of Th17 cells in total CD4+ T cells from five weeks onward after infection.
Schistosome eggs and adult worms are two important sources of antigens exposed to the host during S. japonicum infection [57]. They both have the potential to induce Th1, Th2, Th17 and Treg cells and the corresponding cytokines. Therefore, we confirmed the above hypothesis by using to SEA or SWA to immunize mice as well as to stimulate CD4+ cells in vitro, and our data showed that it was SEA that preferentially induced the generation of Th17 cells and production of IL-17.
The generation and suppression of Th17 cells by Th1, Th2 and Treg cells and/or their cytokines have been demonstrated in numerous studies of in vivo and/or in vitro induction of T cells under defined polarizing conditions [25], [27], [32]–[34]. Based on these reports, the current widely accepted differentiation mode of the Th17 cell subset is as follows. In the presence of TGF-β and IL-6, CD4+ T cells are induced to express the transcription factor RORγt by stimulating the STAT3 and Smad signaling pathway, which leads to the differentiation of Th17 cells. IL-21 together with TGF-β also stimulate the alternative pathway for Th17 differentiation. Meanwhile, the mature Th17 cells amplify themselves by autocrine IL-21. IL-23 also contributes to the maintenance of Th17 cell stability through engagement of the IL-23R, which is expressed by memory or activated T cells [58]. Simultaneously, IFN-γ and IL-4 can effectively inhibit the generation of Th17 cells [2], [5], [59].
Our study clearly showed that during the course of S. japonicum infection, in parallel with the increase of the proportion of Th17 cells, both the inducing (TGF-β, IL-6, IL-21 and IL-23) and inhibitory (IFN-γ, IL-4, Th1, Th2 and Treg cells) factors of Th17 cell generation increased as the infection progressed. These results suggested that a multicellular pathogen such as S. japonicum introduced complex sets of antigens at different stages of infection into the host that could promote both the inducible and the inhibitory factors of Th17 cell generation, but the overall net result was an increase in Th17 cells. In another words, the observed increase in Th17 cells during S. japonicum infection was probably due to the ability of the S. japonicum antigens to more strongly upregulate Th17 inducing factors than the Th17 suppressive factors. In addition, immunization of mice with SEA also preferentially induced Th17 cell generation and the production of higher levels of known factors involved in the generation of Th17 cells (TGF-β, IL-23 and IL-21), while accompanied by the increase in the reported inhibitory factors of Th17 cell generation (Treg and Th2 cells, as well as IL-4).
Many studies have shown that Th17 cells play important roles both in host defenses against extracellular pathogens [6], which are not efficiently cleared by Th1-type and Th2-type immunity and in immunopathogenesis caused by infection. In S. japonicum infection, it has been reported that Th17 may play an important role in the liver immunopathogenesis and in the formation and growth of granulomas around the eggs produced by the adult worm. The findings indicate that the development of severe murine schistosomiasis correlates with high levels of IL-17 and suggest that the exacerbated egg-induced immunopathology is largely mediated by the subset of SEA-induced Th17 cells which produces IL-17 [12], [13]. Our study also suggested that the IL-17 level was positively related to the severity of liver pathogenesis, which was possibly due to the enrichment of inflammatory cells including neutrophils and eosinophils in the granulomas. However, there has been no evidence reported yet to indicate whether Th17 and its product IL-17 either improve or impair the anti-schistosome immune response. Considering that the protection against S. japonicum infection is mainly based on the clearance of the schistosomulum at the early stage of infection, we investigated the potential protective effect of IL-17 levels during that period. Our results showed that administration of rmIL-17A to mice failed to reduce the worm and egg burdens, indicating that high levels of IL-17 did not contribute to the protective responses. Instead, decreasing the level of IL-17 in mice by injecting anti-IL-17A neutralizing mAb led to reductions of worm and egg burdens, suggesting that the decrease of IL-17 levels contributed to effective protective responses against S. japonicum infection. Our study further showed that administration of the anti-IL-17A neutralizing mAb increased the levels schistosome specific IgG1, IgG2a and/or IgG, suggesting that increased antibody-dependent-cell-cytoxicity (ADCC), one of the well accepted mechanisms of killing extra-cellular residing pathogens including schistosome [60], may at least partially contributed to the more effective protective responses against S. japonicum infection. However, the mechanism underlying the role of IL-17 in the protection against S. japonicum infection needs to be further investigated.
In summary, for the first time our study reported on the kinetics of the generation of Th17 cells, which were likely preferentially induced by egg antigens, during S. japonicum infection. We also determined that the proportion of Th17 cells increased together with other CD4+ subsets reported to inhibit them, including Th1, Th2 and Treg cells, as well as their suppressive cytokines in a S. japonicum infection mouse model. These findings suggest that the generation of Th17 cell is determined by the integrated impact of the inducing and suppressive factors promoted by parasitic antigens. More importantly, our study for the first time indicates that a decrease in the level of IL-17 in the early stage of S. japonicum infection may contribute to the host protective responses.
|
10.1371/journal.pntd.0004727 | Targeted Echocardiographic Screening for Latent Rheumatic Heart Disease in Northern Uganda: Evaluating Familial Risk Following Identification of an Index Case | Echocardiographic screening for detection of latent RHD has shown potential as a strategy to decrease the burden of disease. However, further research is needed to determine optimal implementation strategies. RHD results from a complex interplay between environment and host susceptibility. Family members share both and relatives of children with latent RHD may represent a high-risk group. The objective of this study was to use echocardiographic family screening to determine the relative risk of RHD among first-degree relatives of children with latent RHD compared to the risk in first-degree relatives of healthy peers.
Previous school-based screening data were used to identify RHD positive children and RHD negative peers. All first-degree relatives ≥ 5 years were invited for echocardiography screening (2012 World Heart Federation Criteria). Sixty RHD positive cases (30 borderline/30 definite RHD) and 67 RHD negative cases were recruited. A total of 455/667 (68%) family members were screened. Definite RHD was more common in childhood siblings of RHD positive compared to RHD negative (p = 0.05). Children with any RHD were 4.5 times as likely to have a sibling with definite RHD, a risk that increased to 5.6 times when considering only cases with definite RHD. Mothers of RHD positive and RHD negative cases had an unexpectedly high rate of latent RHD (9.3%).
Siblings of RHD positive cases with RHD are more likely to have definite RHD and the relative risk is highest if the index case has definite RHD. Future screening programs should consider implementation of sibling screening following detection of an RHD positive child. Larger screening studies of adults are needed, as data on prevalence of latent RHD outside of childhood are sparse. Future studies should prioritize implementation research to answer questions of how RHD screening can best be integrated into existing healthcare structures, ensuring practical and sustainable screening programs.
| Rheumatic heart disease (RHD) affects at least 33 million people, most of who live in low-resource environments. RHD is a cumulative process and there exists a latent period between early valve damage and presentation with symptoms. Echocardiographic screening (ultrasound of the heart) has proven highly sensitive for latent RHD detection, but implementation research is needed to effectively develop sustainable public health strategies. Critical to this research is determining whom to screen. As family members have both a shared environment and shared genetic susceptibility, they may represent a high-risk group that could be targeted once a case of RHD is identified. We conducted an echocardiographic family screening study to determine the risk of RHD in families with and without an RHD positive child and found that siblings of children with latent RHD are more likely to have latent RHD themselves. Our data suggest that siblings may represent a particularly high-risk group that could be targeted for echocardiographic screening. Future studies are needed to answer questions of how RHD screening can best be integrated into existing healthcare structures, ensuring practical and sustainable RHD screening programs.
| Rheumatic heart disease (RHD), the long-term consequence of acute rheumatic fever (ARF), is the result of a complex interplay between host and environment. Endemic areas are consistently marked by poverty, poor sanitation, and limited access to primary healthcare [1]. These factors increase the incidence of group A β-hemolytic streptococcal (GAS) carriage, infection, and transmission. Repeated, untreated GAS infections create the substrate for development of ARF, a systemic immune system over-reaction that results, for many, in RHD [2].
However, environmental exposure is only one component of RHD susceptibility. Even in the presence of endemic GAS and poor primary prevention (penicillin for acute streptococcal pharyngitis) not all children are equally at risk. ARF follows only 3–6% of cases of GAS and only 40% of children with ARF develop chronic RHD [3]. Historically, RHD was noted to cluster in families, and a meta-analysis of twin studies showed a pooled concordance risk for ARF of 44% in monozygotic twins and 12% in dizygotic twins, giving an estimated heritability of 60%[4]. The majority of these data were captured from observational studies of ARF, and pre-dated routine echocardiography [5].
In many low-resource settings today, presentation with ARF has become rare even as echocardiographic screening of school-aged children has revealed a large burden of latent RHD (RHD apparent on echocardiography that has not previously come to clinical attention). Given what is known about genetic susceptibility and a shared environment, it is reasonable to assume that family members of children with latent RHD may themselves be at greater risk of latent RHD. However contemporary echocardiographic screening of families living in RHD endemic areas has not been reported. The objective of this study was to use echocardiographic family screening to determine the relative risk of RHD among first-degree relatives of children with latent RHD compared to the risk in first-degree relatives of healthy peers.
We utilized a cross-sectional family design to compare the risk of RHD among first-degree family members of primary school children previously identified with latent RHD compared to the first-degree family members of age/gender matched children with normal echocardiograms. The study occurred over a 3-month period from February-April, 2015. Informed consent was obtained from all participants at least 18 years of age, and informed assent and parental permission was obtained for those between 5–17 years. Approval for this study was granted from the Institutional Review Boards at Children’s National Health System, Washington DC, Makerere University School of Medicine, Kampala, Uganda, and the Ugandan National Council of Science and Technology.
RHD positive index cases included children with borderline or definite RHD (2012 WHF criteria), identified through previous echocardiographic school screening programs in the Gulu District of Northern Uganda in 2014.[6] These children are followed clinically at the Gulu Regional Referral Hospital, Gulu, Uganda. RHD negative index cases were recruited from screen-negative peers who were similar in age and gender and attending the same schools (reflecting the same general socioeconomic status). All RHD positive and RHD negative index children underwent repeated echocardiographic evaluation at time of study enrollment to ensure their RHD status had not changed since first screen in 2014.
The parents/guardians of RHD positive and RHD negative index children were approached to invite them, and all first-degree relatives (≥5 years of age) in the family, to undergo echocardiographic screening to evaluate for the presence of latent RHD. Children without at least one parent alive/available were excluded.
Following family recruitment, a list of all first-degree family members (at least 5 years of age)–living or deceased was captured. For living family members, age, gender, and known history of ARF/RHD were recorded. For family members who were deceased, attempts were made to understand the cause of death. For those family members who were alive but unavailable for screening, the reason for absence was recorded.
Each participating family member underwent a focused transthoracic echocardiogram performed by a pediatric cardiologist (TA) with expertise in RHD. A standard acquisition protocol in the parasternal long, parasternal short, and apical 4- and 5-chamber views focused on assessment of the mitral and aortic valves. Additional views were obtained when needed. All images were obtained using fully functional standard portable echocardiographic equipment (GE, VIVID Q, Milwaukee, WI) (Fig 1). Studies were transferred through a secure telemedicine system to PACS (Philips Xcelera, Best, Netherlands) for offline review. Three reviewers (AB, CS, AT), blinded to the RHD/- status of the index child reviewed studies and classified them according to the 2012 WHF criteria (S1 Table) (normal, borderline RHD, definite RHD, or other for subjects ≤20 years of age and normal, definite RHD, or other for subjects >20 years of age)[7]. All positive studies were confirmed by a second reviewer and, in cases of disagreement, a third reviewer determined the final classification.
Demographic information is presented by number and percentage, and where applicable with standard deviation. Continuous variables were compared using Student’s t-test. Fisher’s exact tests were used to evaluate the exact probability under the null hypothesis of observing results as or more extreme based on comparisons of differences in categorical variables between study groups. Poisson Regression was used to estimate the average risk and relative risk of RHD positivity in first-degree family members of RHD positive and RHD negative index children. The method of Poisson Regression was chosen because it appropriately handles outcomes based on counts, including low frequency counts, that generally do not meet criteria for tests requiring the data to meet the parametric assumption and appropriately accounts for the natural clustering of family data. Prevalence rates of RHD among mothers vs. fathers vs. siblings are presented as percentages and compared using model z-statistics. Relative risk was also presented according to the presence of borderline vs. definite RHD in index children. Agreement between reviewers was calculated with the Kappa statistic. Greatest emphasis was placed on results that achieved statistical significance at the p<0.05 level, but substantive differences that achieved borderline significance were also described.
The index group consisted of 61 RHD positive children (30 with definite RHD and 31 with borderline RHD) and 67 RHD negative children (Table 1), generating a complete list of 320 (5.3/child) and 347 (5.2/child) first-degree family members of RHD positive and RHD negative index subjects, respectively, p = 0.22. During enrollment, 1 child (borderline RHD) was excluded from participation when no biological parent could attend screening–leaving 60 RHD positive index cases and 67 RHD negative index cases.
Of the 667 identified first-degree relatives, 455 (68.2%) attended screening including 107 mothers (83.5%), 48 fathers (37.8%), and 300 siblings (72.6%) (Table 2). Fathers (24/127, 19%) were more likely to be deceased than mothers (3/127, 2.4%, p<0.01), with the reasons for paternal death including HIV/AIDS (6), accidental trauma (5), other illness (5), the LRA conflict (3), and other/unknown (5). Fathers who were alive were also less likely to be available for screening (55/103, 53% unavailable) than mothers (17/124, 14% unavailable, p<0.01). Siblings of RHD negative cases were less available to participate in screening (p = 0.03). A breakdown of reasons for all family members who were alive but unavailable and those who are deceased are listed (Table 3). No absent family members were reported to have cardiovascular symptoms and no causes of death were known to be attributable to cardiac disease.
The prevalence of all latent RHD was similar in first-degree relatives of RHD positive and RHD negative cases 9.8% vs. 9.0% (23/235 screened vs. 20/220 screened, p = 0.87). Similarly there was no difference between prevalence of definite latent RHD between groups (Cases: 11/235, 4.3% vs. Controls: 9/220, 4.1%, p = 1.00). Definite RHD was more likely to be found in mothers, with 9.3% (10/107 screened) having echocardiographic evidence of definite RHD, compared to fathers 0% (0/48 screened, p = 0.03), and siblings 3.3% (10/300 screened, p = 0.02). Borderline RHD, a category reserved only for those ≤20 years of age, was similar prevalence between siblings of RHD positive vs. RHD negative (7.7% vs. 8.1%, p = 1.00). However, definite RHD was more common among siblings of RHD positive cases (5.2% vs. 1.4%, p = 0.11), but only reached borderline significance. There were 7 families (4 cases, 3 controls) where 2 or more first-degree relatives were found to be RHD positive (Fig 2).
There was no increased familial, or sibling risk of RHD in the first-degree relatives of RHD positive cases (borderline & definite RHD) vs. RHD negative cases. However, RHD positive cases had a 4.5 times greater chance of having a sibling with definite RHD (p = 0.05) and this risk increased to 5.6 times greater chance if you limited the comparison to RHD positive cases with definite RHD (n = 30, p = 0.03) (Table 4).
There was 97% agreement between reviewers 1 and 2 (κ = 0.86, 95% CI 0.78–0.93), with 13 cases of non-agreement adjudicated by the third reviewer. All cases of non-agreement were between the diagnoses of “borderline RHD” or “normal” with 100% agreement on the diagnosis of definite RHD.
This is the first study to assess the utility of echocardiography screening of first-degree relatives of children with latent RHD. Siblings of RHD positive cases with any RHD are more likely to have definite RHD and the relative risk goes from 4.5 to 5.6 if the index case has definite RHD. Additionally, we found that nearly 10% of mothers had latent RHD by echocardiography while no fathers were positive. Unlike our sibling results, the likelihood of a mother being positive for RHD has no association with a positive index case.
Echocardiographic screening has shown potential as a public health strategy to decrease the global burden of RHD. Population studies, mostly involving schoolchildren, have revealed a weighted pooled prevalence of 1.3% of children living in endemic areas show evidence of latent RHD [8]. Early detection of these children provides the opportunity for secondary prophylaxis, monthly penicillin injections that prevent recurrent streptococcal infections, rheumatic fever, and further valve damage. This is of particular importance in areas such as sub-Saharan Africa where RHD remains endemic, ARF rarely comes to clinical attention [9], and RHD patients most commonly present late, with advanced disease and resulting complications [10].
Optimal implementation strategies, the who, when, in what setting, and how often to screen, have received little study to date, yet these details are critical to developing cost-effective and sustainable screening programs. Our study suggests that siblings of children identified with latent RHD are a high-risk group, and should be prioritized for screening. Siblings of index controls showed a 1.4% prevalence of definite RHD, which is comparable to previously published data from the pediatric population in Gulu, Uganda [6]. In contrast, siblings of RHD positive cases were found to have a 5.2% prevalence of latent, but definite RHD–and to be at 4.5–5.6 times risk, depending if you included index cases with borderline and definite RHD or only those with definite RHD respectively. In a resource-constrained setting, identification of this increased risk could translate into strategic targeting of siblings for single or repeated echocardiographic screening and for education on primary prevention. Strategies such as this would need formal evaluation, but hold promise to save financial and human resources.
There is a similar precedent from the World Health Organization for prioritizing household contacts when an index case of tuberculosis is identified [11]. In tuberculosis, household screening has been shown to dramatically improve prevention, early diagnosis, and outcomes [12, 13]. Similar to tuberculosis, RHD has a strong environmental component. RHD originates from group A Streptococcus (GAS), which is endemic in areas of poverty and overcrowding. While the risk of invasive GAS infection among household contacts is only mildly increased when an index case is identified [14], it is likely that family members are exposed to the same streptococcal strains at a similar frequency over time. Thus, while familial chemoprophylaxis is not recommended for individual invasive streptococcal infections [15], our data suggest familial echocardiographic screening, once RHD is identified, may be worthwhile.
In addition to a shared environment, first-degree relatives also share genetic susceptibility, which is thought to play a crucial role in RHD development [2, 16]. Engel et al. reported a meta-analysis of twin-studies that included 435 twin pairs between 1933 and 1964[4]. The pooled concordance risk for ARF was 44% in monozygotic twins and 12% in dizygotic twins (OR 6.39, p<0.001), with an estimated heritability of 60%. To date, targeted genomic investigations have examined select genes involved in immune regulation including human leukocyte antigens (HLA), transforming growth factor-beta1, toll-like receptor 5, angiotensin I-converting enzyme gene, PTPN22, and signal transducers and activators of transcription (STATs) gene polymorphisms [17–23]. The most robust data comes from studies of the major histocompatibility complex human leukocyte antigens [3, 24], with many finding polymorphisms within the HLA-DR locus, including a study from Uganda [25]. However, most of these investigations have been small and no single or combination haplotype has consistently emerged [3, 24]. In this study, we cannot separate the influence of host susceptibility from that of the shared environment; our phenotypic data supports the concept of genetic predisposition in RHD. Future genome-wide association studies are needed to link phenotype to genotype and elucidate the drivers of familial susceptibility.
While not the primary objective of this investigation, our study design captured some of the first data on the community burden of RHD in adults. We found an unanticipated high prevalence of latent RHD among adult females (10/107 or 9.3% of those screened), while no cases were identified in adult males (0/43 screened). These findings are similar to those by Paar et al, who reported a preponderance of female adults both available for screening and having RHD in Nicaragua (91% of all adult cases of definite RHD)[26].
The high prevalence of definite RHD also did not vary between the mothers of cases and those of controls. We hypothesize that these women may have been at particularly high risk due to historical events in the region. Gulu, Uganda, the site of our study, was subject to a major humanitarian crisis between 1996 and 2007, when most of the included mothers would have been children or adolescents. This crisis resulted in families being displaced to refugee camps, where poor sanitation, overcrowding, and constrained access to primary care–all drivers of streptococcal disease–were commonplace. Studies examining the impact of these camps have found a disproportionate impact on woman’s health during this time period [27,28]. While more extensive survey of adult populations in sub-Saharan Africa are needed to confirm, we speculate that the extremely high prevalence of RHD among women in Gulu may not be replicated to this extreme in the wider population. Further echocardiographic studies of latent RHD prevalence in adult populations are needed, in particular as RHD in pregnancy carries high risks of maternal and fetal morbidity and mortality [29,30].
Our study has several limitations. While we were able to capture most mothers and over three-quarters of siblings for screening, we were able to screen less than half of the fathers. This was due to both higher levels of paternal mortality and for those fathers who were alive, high rates of living and working outside of Gulu. Absentee rates for fathers were comparable between cases and controls, but we cannot accurately determine prevalence of latent RHD among fathers. Additionally, a greater number of siblings were captured from RHD positive compared to RHD negative cases. While there was no difference in death rates between siblings and no evidence of poorer cardiovascular health captured among reasons for sibling non-attendance, the impact of this difference cannot be known. We included as RHD positive index cases both children identified with borderline and with definite RHD. While our most significant findings were associated with a sub-analysis of siblings of definite RHD cases, it is possible that the inclusion of borderline cases weakened our overall phenotype, and study power, which could result in underestimation of risk. It is also important to remember that the total number of RHD cases in our families was small, leading to wide confidence intervals and less certainty in our findings, making them more hypothesis-generating than definitive. Finally, we did not specially control for variations in socio-economic status level between families. However, index controls were recruited from the same schools as cases to indirectly equilibrate socio-economic conditions between groups, and the number of people per household did not differ between cases and controls.
It is also necessary to point out that many important practical and logistical questions remain before a public health strategy that includes echocardiographic screening can be broadly recommended. Our study demonstrates that, given the right resources, large-scale echocardiographic screening is feasible in low-income impoverished areas of the world. However, healthcare resources in most endemic areas are highly constrained, and lack of human and financial resources commonplace. Studies examining optimal training strategies and use of less-expensive handheld echocardiography in the hands of non-experts are beginning to address these barriers [31–34]. The natural history of latent RHD, in particular the category of borderline RHD, is unknown. Natural history studies of children with latent RHD are ongoing and may provide answers [32, 35, 36, 37].
In conclusion, siblings of RHD positive cases with any RHD are more likely to have definite RHD and the relative risk is highest if the index case has definite RHD. Future screening programs should consider implementation of sibling screening following detection of an RHD positive index case. Follow-up of this cohort is needed to determine if latent RHD that exists in more than one family member is more likely to persist and progress. Future studies should prioritize implementation research to answer questions of how RHD screening can best be integrated into existing healthcare structures, ensuring practical and sustainable screening programs.
|
10.1371/journal.ppat.1000703 | Th1-Th17 Cells Mediate Protective Adaptive Immunity against Staphylococcus aureus and Candida albicans Infection in Mice | We sought to define protective mechanisms of immunity to Staphylococcus aureus and Candida albicans bloodstream infections in mice immunized with the recombinant N-terminus of Als3p (rAls3p-N) vaccine plus aluminum hydroxide (Al(OH3) adjuvant, or adjuvant controls. Deficiency of IFN-γ but not IL-17A enhanced susceptibility of control mice to both infections. However, vaccine-induced protective immunity against both infections required CD4+ T-cell-derived IFN-γ and IL-17A, and functional phagocytic effectors. Vaccination primed Th1, Th17, and Th1/17 lymphocytes, which produced pro-inflammatory cytokines that enhanced phagocytic killing of both organisms. Vaccinated, infected mice had increased IFN-γ, IL-17, and KC, increased neutrophil influx, and decreased organism burden in tissues. In summary, rAls3p-N vaccination induced a Th1/Th17 response, resulting in recruitment and activation of phagocytes at sites of infection, and more effective clearance of S. aureus and C. albicans from tissues. Thus, vaccine-mediated adaptive immunity can protect against both infections by targeting microbes for destruction by innate effectors.
| The bacterium Staphylococcus aureus and the fungus Candida are the second and third leading cause of bloodstream infections in hospitalized patients. A vaccine to prevent such infections would be of enormous public health benefit. The leading hypothesis to explain why vaccines have not been successfully developed against these infections is that the microbes causing the infections are highly complex, and use multiple weapons (so-called “virulence factors”) to cause disease in humans. Therefore, a vaccine targeting either infection would have to neutralize many of these virulence factors at the same time. We have been developing a vaccine that simultaneously targets both types of infections. Our vaccine is based on a single virulence factor used by Candida, which has a similar shape to virulence factors used by S. aureus. In the current study, we report that our vaccine induces specialized cells in the immune system to more effectively call in reinforcements to kill the organisms. These data demonstrate that vaccines against both organisms can be developed even if they do not work by neutralizing multiple virulence factors, and therefore open the door to a far wider array of vaccine types against both infections.
| Staphylococcus aureus and Candida spp. are the second and third leading causes of bloodstream infections in hospitalized patients [1]. These organisms jointly cause at least 150,000 clinical bloodstream infections resulting billions of dollars of health-care expenditures and ∼40,000 deaths per year in the US alone [1]–[4]. Identification of immune mechanisms of protective adaptive immunity against these organisms is critical to lay the groundwork for development of active vaccine strategies against both organisms.
We previously reported that vaccination with the recombinant N terminus of the candidal Als3p adhesin (rAls3p-N) with aluminum hydroxide (Al(OH)3) adjuvant improved the survival of mice subsequently infected intravenously with lethal inocula of Candida albicans or methicillin resistant Staphylococcus aureus (MRSA) [5]–[7]. The vaccine retained efficacy against both infections in B cell deficient animals but not T cell deficient animals [6],[7]. Furthermore, adoptive transfer of CD4+ T cells but not B220+ B cells or immune serum improved the survival of recipient mice infected with both organisms [6],[7].
Although T cells are necessary for rAls3p-N vaccine efficacy, lymphocytes are not capable of directly killing C. albicans or S. aureus in culture [8],[9]. Therefore, the downstream effectors of vaccination against both organisms have remained unclear. In contrast to lymphocytes, phagocytes kill C. albicans and S. aureus in vitro [8],[10],[11] and in vivo [12]–[16], especially when primed with pro-inflammatory cytokines such as interferon (IFN)-γ, which is produced by CD4+ lymphocytes. Therefore, we hypothesized that the end effectors of rAls3p-N vaccine-mediated protection against bloodstream infection caused by S. aureus and C. albicans were phagocytes primed by pro-inflammatory cytokines produced by vaccine-responsive lymphocytes. We sought to elucidate fundamental requirements of protective host immunity to bloodstream infection caused by S. aureus and C. albicans.
We previously established that the rAls3p-N vaccine was not effective against C. albicans iv infection in IFN-γ-deficient mice [6]. We sought to determine if IFN-γ was similarly required for vaccine-mediated protection against S. aureus, and also to determine if CD4+ T cells were the required source of IFN-γ production to mediate vaccine efficacy against both organisms. IFN-γ-deficient mice or their wild-type, congenic controls were vaccinated with rAls3p-N plus Al(OH)3 (vaccinated) or Al(OH)3 alone (control), and boosted at three weeks. Two weeks following the boost, CD4+ splenic and lymph node lymphocytes from vaccinated or control donor mice were purified and cross-adoptively transferred to recipient mice (IFN-γ deficient donor cells were transferred to wild type recipient mice, and visa versa). As a negative control, vaccinated or control IFN-γ knockout mice were infected without undergoing adoptive transfer. Mice were infected via the tail-vein with C. albicans or MRSA the day following adoptive transfer.
IFN-γ-deficient mice receiving immune CD4+ lymphocytes from vaccinated, wild type donor mice had improved survival after either infection, whereas wild type mice receiving immune CD4+ lymphocytes from IFN-γ-deficient, vaccinated donor mice did not have improved survival (Fig. 1). Cells from control donor mice were not effective at improving survival of recipient mice. Hence, IFN-γ produced by vaccine-primed CD4+ T cells was required for mediating adaptive immunity against both infections.
Because lymphocyte derived pro-inflammatory cytokines, including IFN-γ, can activate phagocytes to mediate superior killing of C. albicans or S. aureus in culture [8], [11], [17]–[22], we sought to define the role of downstream phagocytes in adaptive immune-mediated protection. First we vaccinated mice as above and administered cyclophosphamide to induce neutropenia two weeks after the boost. Two days later we infected the mice with one of two clinical isolates of C. albicans, or with MRSA. The second isolate of C. albicans (15563) was used because it results in a less rapidly lethal infection than SC5314, and the diminished severity of infection would afford the opportunity to unmask any subtle, residual protection afforded by vaccination in the neutropenic mice. Cyclophosphamide-induced neutropenia disrupted the improvement in survival mediated by the vaccine against infections caused by either strains of C. albicans and S. aureus (Fig. 2 and Fig. S1 for the 15563 strain).
We also tested vaccine efficacy in gp91phox−/− deficient mice, the phagocytes of which are unable to generate superoxide and have marked defects in microbial killing. Such mice have been previously shown to have enhanced susceptibility to pulmonary and intraperitoneal infection by C. albicans [23],[24], but have not been studied in the intravenous model. We performed pilot studies to determine how susceptible to iv infection with C. albicans or S. aureus the gp91phox−/− mice were. Remarkably, we found that the 100% lethal dose (LD100) of C. albicans SC5314 was >150-fold lower in gp91phox−/− mice vs. wild-type controls (<103 vs. 1.5×105). The LD100 of S. aureus LAC was 2-fold lower in gp91phox−/− mice vs. wild type controls (5×106 vs. 107).
Subsequently, gp91phox−/− and wild type mice were vaccinated with rAls3p-N plus Al(OH)3 or Al(OH)3 alone. CD4+ T cells from vaccinated or control wild type donor mice were adoptively transferred into gp91phox−/− recipient mice, and visa versa. As well, some mice were vaccinated and infected without undergoing adoptive transfer as positive (wild type) and negative (gp91phox−/−) controls for vaccine efficacy. The vaccine did not improve the survival of gp91phox−/− mice infected with either organism (Fig. 3A). While CD4+ T lymphocytes from vaccinated, gp91phox−/− donor mice improved the survival of wild type recipient mice, CD4+ T cells from vaccinated, wild-type donor mice failed to improve the survival of gp91phox−/− recipient mice (Fig. 3B).
The need for downstream functional phagocytes to mediate vaccine efficacy suggested that Th17 cells, which are known to act by recruiting phagocytes to the sites of infection [25],[26], might play a role. To determine the requirement for IL-17 and Th17 cells in mediating vaccine efficacy, we vaccinated mice deficient in IL-17A, or their wild type congenic control mice. IL-17A-deficiency abrogated vaccine-mediated efficacy (Fig. 4A). Of note, in contrast to IFN-γ deficiency, IL-17A deficiency did not exacerbate the severity of infection in unvaccinated mice (comparing survival of unvaccinated, deficient vs. wild type mice).
To determine if CD4+ T cells were the primary source of IL-17A in mediating vaccine efficacy, CD4+ T cells from vaccinated or control mice were cross-adoptively transferred into recipient mice (IL-17A-deficient donor cells transferred to wild type recipient mice; wild type donor cells transferred to IL-17A-deficient recipient mice). We also repeated the survival study in wild type and IL-17A deficient mice that did not undergo adoptive transfer to serve as positive and negative controls for the adoptive transfer study. Mice were infected the day after adoptive transfer. Once again, the vaccine improved the survival of the positive control wild type mice but not the negative control IL-17A deficient mice (Fig. 4B). Adoptive transfer of CD4+ cells from vaccinated wild type donor mice improved the survival of IL-17A-deficient recipient mice (Fig. 4B). In contrast, transfer of CD4+ T cells from vaccinated IL-17A-deficient donor mice to wild type recipient mice failed to improve survival (Fig. 4C), confirming that CD4+ T cell derived IL-17A was necessary to mediate vaccine efficacy.
To define the populations of cells induced by vaccination, spleens and lymph nodes were harvested from vaccinated and control mice two weeks following the boost. The cells were stimulated ex vivo for 5 days with rAls3p-N. Analysis of supernatants confirmed that cells from vaccinated mice produced significantly more IFN-γ and IL-17, as well as the neutrophil-acting chemokines, KC and MIP-1α, than did cells from control mice (Fig. 5A). IL-4 levels were not detectable in any supernatant from control cells; levels were detectable at low levels (< 2 pg/ml) in supernatants from 4 of the 8 mice in the vaccinated group. However, IL-10 and IL-13 levels were higher in supernatants from vaccinated than control mice. Levels of TGF-β and IL-6 were low and not significantly different in supernatants from vaccinated or control mice. Supernatants from stimulated, immune cells markedly enhanced phagocytic killing of C. albicans and S. aureus ex vivo, compared to supernatant from control cells (Fig. 5B).
Intracellular cytokine analysis of the stimulated cells demonstrated that vaccination resulted in increased frequencies of Th1 (CD4+IFN-γ+), Th17 (CD4+IL-17+), and Th1/17 (CD4+IFN-γ+IL-17+) cells in draining lymph nodes lymphocytes (Fig. 6 and Figs. S2 and S3) compared to the frequencies in unvaccinated mice. Murine CD4+CCR6- cells were enriched for the Th1 phenotype, and CD4+CCR6+ cells were enriched for the Th17 phenotype. However, a substantial proportion of CCR6+ splenocytes, and particularly CD4+CCR6+ lymphocytes, were Th1/17 (IFNγ+IL-17+) cells. The Th1/17 phenotype was predominantly found in CD4+CCR6+ cells, not in the CD4+CCR6- cells.
To confirm the in vivo biological relevance of the ex vivo lymphocyte phenotypes, vaccinated or control mice were infected via the tail vein with C. albicans or S. aureus 2 weeks following the boost. At day 4 post-infection (the day before control mice were anticipated to begin dying), burden of infection and cytokine levels in homogenates of individually marked kidneys (primary target organ) were determined. Levels of myeloperoxidase (MPO), which is constitutively expressed at the protein level in neutrophils and has been extensively used in previous studies to quantify neutrophil influx into tissues during infection and inflammation [27]–[31], were also measured.
Vaccination resulted in a ∼10-fold reduction in kidney fungal burden and ∼5-fold reduction in kidney bacterial burden (Fig. 7A). MPO levels were increased in vaccinated mice relative to control mice infected with either organism (Fig. 7B). A recent study reported a 95% correlation between organ fungal burden and neutrophil influx in mice infected with different strains of C. albicans or C. dubliniensis [32]. Therefore, any enhanced neutrophil influx resulting from vaccination could be offset by the diminished stimulus for neutrophil influx caused by reduced fungal burden in the vaccinated mice. To isolate the impact on MPO levels of vaccination, and not severity of infection, we adjusted absolute MPO levels in individually marked organs for the fungal or bacterial burden in those individual organs. Vaccination resulted in a marked increase in neutrophil influx relative to the infectious burden of organism in the tissues (Fig. 7B). By histopathology, the inflammatory infiltrate was predominantly neutrophilic, with scattered foci of macrophages. Semi-quantitative scoring of histopathology sections by a blinded pathologist to estimate neutrophil influx into tissues was concordant with the quantitative MPO levels.
Concordant with ex vivo cytokine measurements, absolute levels of IFN-γ, IL-17, and the neutrophil-acting CXC chemokine, KC, were higher in the kidneys of vaccinated versus control mice (p<0.05 for all comparisons of vaccinated vs. control levels for all three cytokines, in mice infected with C. albicans or S. aureus). After adjusting for infectious burden in individual organs, vaccination markedly increased cytokine levels relative to infectious burden (Fig. 7C).
Histopathology confirmed a marked increase in organism burden in the vaccinated mice versus control mice infected with either organism (Fig. 8). Numerous microabscesses with hyphal and pseudohyphal elements were scattered throughout the kidneys of control mice infected with C. albicans. Microabscesses were also found in the kidneys of vaccinated mice, but most of the abscesses had no fungal elements visible, and those few abscesses with fungal elements contained blastospores or small hyphal fragments. Control mice infected with S. aureus had large renal abscesses with numerous gram positive cocci on Gram stain. Vaccinated mice also had renal abscesses with extensive neutrophil influx, but in most abscesses fewer staphylococcal organisms were seen on Gram stain in vaccinated than control mice (Fig. 8).
One hypothesis regarding the failure to date to develop an effective vaccine against S. aureus or Candida has been the need to simultaneously disrupt multiple virulence factors for such complex pathogens, whereas most vaccines to date have targeted only one virulence factor [33],[34]. However, we have previously confirmed that humoral immunity is neither necessary nor sufficient for rAls3p-N vaccine-induced protection against either organism [6],[7]. Furthermore, homozygous disruption of ALS3 in C. albicans does not result in a loss of pathogenicity in vivo in mice, so the protection mediated by the rAls3p-N vaccine is not the result of abrogation of Als3p virulence functions. The current study confirms that vaccination can be effective by targeting the organism for destruction by increasing the quantity and microbicidal functions of innate phagocytic effectors at the site of infection, irrespective of affecting virulence functions in the organism. Therefore, potential vaccine antigens need not be restricted to microbial virulence factors, and can be expanded to include any target antigen which results in a potent Th1 and/or Th17 immune response against the organism. These data are concordant with the established role of Th17 cells in mediating protection following immunization of mice against Mycobacterium tuberculosis, Helicobacter pylori and Pseudomonas aeruginosa [31],[35],[36].
In unvaccinated animals, deficiency in IFN-γ but not IL-17A exacerbated the severity of iv infection caused by both S. aureus and C. albicans. These results are concordant with recent studies demonstrating that IL-17-deficient mice were not more susceptible to bloodstream infection caused by S. aureus [37] or invasive gastric infection caused by C. albicans [38]. Furthermore, a recent study reported that abrogation of the dectin-2 receptor blocked Th17 induction by C. albicans in mice, but despite the lack of a Th17 response did not affect the ability of mice to clear fungus from tissue during systemic infection [39]. Collectively, these results indicate that Th17 cells/IL-17A are not necessary for normal murine host defense against disseminated candidiasis.
In contrast, IL-17 has been shown to be critical for host defense against cutaneous and oropharyngeal infections caused by S. aureus [37] and C. albicans [40], respectively. Furthermore, our results are discordant with the previous finding that IL-17 receptor-deficiency moderately exacerbated the severity of bloodstream infection caused by C. albicans [41]. The lack of a requirement for IL-17A to mediate normal host defense against disseminated candidiasis likely reflects the ability of IL-17F, which also activates the common IL-17 receptor, to complement an IL-17A deficiency. Differences in C. albicans infecting strain and mouse host strain may also account for differences between our study and the prior. However, a critical point is that IL-17F could not compensate for the requirement for IL-17A in mediating protective, vaccine-induced, adaptive immunity, since IL-17A deficiency abrogated vaccine efficacy.
We confirmed that the rAls3p-N vaccine specifically primed splenic and lymph node lymphocytes to produce high levels of both IFN-γ and IL-17, as well as the neutrophil chemokines, KC and MIP-1α (the latter of which is chemotactic for both mononuclear cells and neutrophils [42]–[47]). The predominant IFN-γ expression in lymph nodes was found in CCR6- Th1 cells which did not produce IL-17 (CD4+CCR6−IFN-γ+IL-17−), and the predominant IL-17 expression in lymph nodes was found in CCR6+ cells. However, we also found substantial numbers of Th1/17 cells, which met or exceeded the frequency of Th17 cells, in the CD4+CCR6+ fraction. The Th1/17 cells were found virtually exclusively in the CCR6+ fraction, and none were found in the CD4+CCR6- fraction. Recent studies have indicated that yeast mannosylated proteins prime Th17 cells via activation of the mannose receptor [48], and that O-linked mannoproteins can activate IFN-γ production via ligation of TLR4 [49]. Since the rAls3p-N protein has O-linked yeast high mannose groups, co-ligation of the mannose receptor and TLR4 on antigen presenting cells may enable induction of Th1, Th17, and Th1/17 cells. The role of specific antigen presenting cells in priming lymphocytes for Th1, Th17, or dual Th1/17 responses is currently under investigation.
We found variations in the total number of surviving mice from experiment to experiment, ranging from as high as 87% to as low as 12.5%. Variations in outcome are most likely accounted for by variations in infectious inoculum and infecting strain. Our challenge model, using the standard SC5314 clinical isolate of C. albicans, is extremely rigorous, and is considerably more rigorous than challenge with other clinical strains of C. albicans [6],[50],[51], as evidenced by the superior efficacy seen in the current study with another clinical bloodstream isolate of C. albicans (15563). We have previously shown that mice infected with the inocula of SC5314 used in these experiments die of overwhelming septic shock [52]. Candidal septic shock causes >50% mortality in humans despite treatment with antifungal therapy [4]. Hence, achievement of survival approaching 50% by vaccination alone is felt to reflect meaningful protection. Furthermore, the experiment in which 12.5% survival was seen in the vaccinated arm was an adoptive transfer experiment, in which immune cells from wild type mice were transferred into IL-17A-/- recipient mice. Thus, while IL-17A production from CD4+ immune T cells can transfer protection, production of IL-17A by other cell types may be required to achieve maximal protection. Specifically, we previously found that immune CD8+ T cells could transfer protection against S. aureus [7], and macrophages or dendritic cells can produce pro-inflammatory cytokines such as IFN-γ, suggesting that these cell types may play an adjunctive role and be required for full vaccine-mediated protection.
We previously reported that cyclophosphamide-induced neutropenia did not completely abrogate vaccine-induced protection during subsequent disseminated candidiasis [53]. In contrast, in the current study, we did find total abrogation of protection against both candidal strains and against S. aureus. The prior study used a different but related vaccine immunogen, rAls1p-N, instead of rAls3p-N. As well, the prior study used Complete Freund's Adjuvant (CFA), not Al(OH)3. The greater efficacy of the former adjuvant may account for the residual efficacy found in neutropenic mice in the former study.
S. aureus and C. albicans express adhesins on their cell surface which possess similar three dimensional shapes [54] and which bind to similar endovascular surfaces (e.g. endothelial cells and subendothelial matrix proteins) and medically relevant plastics [54],[55]. Given these similar virulence mechanisms, it is not surprising that the organisms also infect patients with similar risk factors, including post-operative and trauma patients, patients with central venous catheters, patients on hemodialysis, and patients with compromised phagocytic host defense mechanisms [4],[56],[57]. Finally, our data demonstrate that the host defends itself against both infections by similar mechanisms, and that adaptive immunity to both organisms required CD4+ T cell production of both IFN-γ and IL-17A.
In summary, the rAls3p-N vaccine improved outcomes in mouse models of iv S. aureus and C. albicans infection by inducing upstream, pro-inflammatory, Th1, Th17, and Th1/17 lymphocytes, which enhanced recruitment and activation of neutrophils in infected tissues, thereby reducing tissue infectious burden. Thus, vaccination showed a potential to protect against both infections by targeting the microbes for enhanced destruction by innate effector cells, irrespective of neutralization of microbial virulence factors. Therefore, potential vaccine antigens need not be restricted to microbial virulence factors, and can be expanded to include any target antigen which results in a potent Th1 and/or Th17 immune response against the organisms.
C. albicans SC5314 was supplied by W. Fonzi (Georgetown University), and S. aureus LAC, a USA300 MRSA clinical isolate, was provided by Frank Deleo (NIAID/NIH). C. albicans 15563 is a clinical bloodstream isolate from a patient at Harbor-UCLA Medical Center which is also virulent in our murine model [50]. Candida was serially passaged three times in yeast peptone dextrose broth (Difco) at room temperature prior to infection. S. aureus was grown overnight at 37°C in BHI broth, and then passaged for 4 hours at 37°C in fresh BHI broth.
Female Balb/c or C57BL/6 mice were obtained from Taconic Farms (Bethesda, MD). Congenic IL-17A deficient mice on a Balb/c background were obtained from Y. Iwakura (University of Tokyo) [58]. Vaccinated mice were infected via the tail vein with the appropriate inocula of C. albicans blastospores or S. aureus organisms in PBS, as previously described [7],[52]. In some experiments, mice were made neutropenic by treatment with 230 mg/kg cyclophosphamide 2 days prior to infection, a regimen which results in profound neutropenia for approximately one week [59],[60]. All procedures involving mice were approved by the Los Angeles Biomedical Research Institute animal use and care committee, following the National Institutes of Health guidelines for animal housing and care.
rAls3p-N (amino acids 17 to 432 of Als3p) was produced in Saccharomyces cerevisiae and purified by Ni-NTA matrix affinity purification as previously described [61]. Mice were immunized by subcutaneous (SQ) injection of 300 µg of rAls3p-N in 0.1% Al(OH)3 (Alhydrogel, Brenntag Biosector, Frederikssund, Denmark) in PBS. Control mice received adjuvant alone on the same schedule. Some mice were boosted at 21 days. Mice were infected two weeks following the boost.
Serum and splenic lymphocytes were harvested from vaccinated or control mice, as we have previously described [62]. Lymph node lymphocytes were harvested from cervical and axillary lymph nodes, based on pilot studies with Evans Blue dye lymph node mapping demonstrating that SQ vaccination at the base of the neck drained primarily to these lymph nodes. For adoptive transfers, splenic and lymph node lymphocytes were pooled. CD4+ T lymphocytes were purified by use of the IMag system (BD Pharmingen), as we have described [6],[7]. Purified lymphocytes (5×106 per mouse) were administered iv to congenic, unvaccinated recipient mice. Transferred cell populations were ≥95% pure by flow cytometric analysis. Mice were infected via the tail-vein with C. albicans SC5314 24 h after lymphocyte adoptive transfer.
Intracellular cytokines from lymphocytes were analyzed based on a modification of our previously described method [62]. In brief, cervical and axillary lymph nodes and spleens were dissected from vaccinated or control mice and passed through 70 µm filters. Cells were stimulated for 5 days with rAls3p-N (12.5 µg/ml) in complete media (RPMI 1640, 50 U/ml penicillin, 50 µg/ml streptomycin, 2 mM L-glutamine, 10% FBS, 5 µM 2-ME) in 96 well plates. PMA (50 ng/ml), ionomycin (1 µM), and monensin (10 µg/ml) were added during the final 6 hours of culture. Supernatant was harvested prior to adding monensin for analysis of cytokine content using Cytometric Bead Array Flex kits (BD Pharmingen, La Jolla, CA) or ELISA (for IL-6, TGF-β, and IL-13), per the manufacturer's instructions. Cells were stained on ice with PerCP-anti-CD4 and Alexa647-anti-CCR6 (BD Pharmingen, San Diego), or their isotype control antibodies. The cells were fixed and permeabilized as previously described [62]. Intracellular cytokines were stained with rat FITC-anti-mouse IFN-γ and PE-anti-IL-17, or their isotype controls (BD Pharmingen). Four-color flow cytometry was performed on a Becton-Dickinson FACScan instrument calibrated with CaliBRITE beads (Becton Dickinson, San Jose, CA) using FACSComp software as per the manufacturer's recommendations. Data for each sample were acquired until 10,000 CD4+ lymphocytes were analyzed. Th1 cells were defined as CD4+IFNγ+IL-17−, Th17 cells defined as CD4+IFN-γ−IL-17+, and Th1/17 cells defined CD4+IFNγ+IL-17+.
The killing assay for both C. albicans and S. aureus was modified based on our well-described assay [59],[60]. In brief, RAW murine macrophage cells or murine neutrophil cells were grown in DMEM plus 10% fetal bovine serum. Fresh murine neutrophils were harvested by dextran sedimentation of whole, heparanized blood, followed by centrifugation over Ficoll Hypaque at 500 g for 10 minutes. The RAW cells or neutrophils were added into 24 well plates, the media in the wells was aspirated and the RAW cells or fresh neutrophils were cultured for 4 hours in 10% conditioned media (from vaccinated or control splenic and lymph node lymphocytes exposed to rAls3p-N for 5 days) plus 90% complete media (RPMI + 10% FBS). The conditioned media was then aspirated, and the microorganisms added to the wells in fresh DMEM plus 10% fetal bovine serum. Microorganisms were added to the wells at a ratio of 20∶1 RAW cells to C. albicans, 5∶1 RAW cells to S. aureus, or 10∶1 fresh neutrophils to C. albicans or S. aureus. Media for the wells containing S. aureus contained no antibiotics. The cells were incubated at 37°C for 1 h, at which point 4% blood heart infusion (BHI) agar was directly added to the wells. Plates were incubated overnight at 37°C and colony forming units (CFUs) counted in each well. Killing was defined as the percent reduction CFUs in wells containing co-cultures of phagocytes cells and microorganisms compared to wells just containing microorganisms.
On day 4 post-infection, kidneys (primary target organ) were harvested and homogenized in saline with protease inhibitors (pepstatin, leupeptin, and PMFS). For determination of infectious burden, organ homogenates were quantitatively cultured on Sabourad dextrose agar for C. albicans or tryptic soy agar for S. aureus. Whole organ cytokines were analyzed from kidney homogenates by ELISA (R&D Systems) or Cytometric Bead Array Flex kit for KC (BD Pharmingen, La Jolla, CA), per the manufacturer's instructions. MPO levels were determined by ELISA (Hycult Biotechnology, Uden, Netherlands) of whole organ homogenates. For histopathology, organs were fixed in zinc-buffered formalin, embedded in paraffin, sectioned, and stained with PAS for fungi and H&E and Gram stain for bacteria.
The non-parametric Log Rank test was utilized to determine differences in survival times. The Wilcoxon Rank test was used to compare cytokines, MPO levels, and organ burden across groups. P<0.05 was considered significant.
|
10.1371/journal.pntd.0001018 | Snakebite Mortality in India: A Nationally Representative Mortality Survey | India has long been thought to have more snakebites than any other country. However, inadequate hospital-based reporting has resulted in estimates of total annual snakebite mortality ranging widely from about 1,300 to 50,000. We calculated direct estimates of snakebite mortality from a national mortality survey.
We conducted a nationally representative study of 123,000 deaths from 6,671 randomly selected areas in 2001–03. Full-time, non-medical field workers interviewed living respondents about all deaths. The underlying causes were independently coded by two of 130 trained physicians. Discrepancies were resolved by anonymous reconciliation or, failing that, by adjudication.
A total of 562 deaths (0.47% of total deaths) were assigned to snakebites. Snakebite deaths occurred mostly in rural areas (97%), were more common in males (59%) than females (41%), and peaked at ages 15–29 years (25%) and during the monsoon months of June to September. This proportion represents about 45,900 annual snakebite deaths nationally (99% CI 40,900 to 50,900) or an annual age-standardised rate of 4.1/100,000 (99% CI 3.6–4.5), with higher rates in rural areas (5.4/100,000; 99% CI 4.8–6.0), and with the highest state rate in Andhra Pradesh (6.2). Annual snakebite deaths were greatest in the states of Uttar Pradesh (8,700), Andhra Pradesh (5,200), and Bihar (4,500).
Snakebite remains an underestimated cause of accidental death in modern India. Because a large proportion of global totals of snakebites arise from India, global snakebite totals might also be underestimated. Community education, appropriate training of medical staff and better distribution of antivenom, especially to the 13 states with the highest prevalence, could reduce snakebite deaths in India.
| Earlier hospital based reports estimate about 1,300 to 50,000 annual deaths from snakebites per year in India. Here, we present the first ever direct estimates from a national mortality survey of 1.1 million homes in 2001–03. Full-time, non-medical field workers interviewed living respondents about all deaths. The underlying causes were independently coded by two of 130 trained physicians. The study found 562 deaths (0.47% of total deaths) were assigned to snakebites, mostly in rural areas, and more commonly among males than females and peaking at ages 15–29. Snakebites also occurred more often during the rainy monsoon season. This proportion represents about 45,900 annual snakebite deaths nationally (99% CI 40,900 to 50,900) or an annual age-standardised rate of 4.1/100,000 (99% CI 3.6–4.5), with higher rates in rural areas (5.4) and with the highest rate in the state of Andhra Pradesh (6.2). Annual snakebite deaths were greatest in the states of Uttar Pradesh (8,700), Andhra Pradesh (5,200), and Bihar (4,500). Thus, snakebite remains an underestimated cause of accidental death in modern India, causing about one death for every two HIV-related deaths. Because a large proportion of global totals of snakebites arise from India, global snakebite totals might also be underestimated. Effective interventions involving education and antivenom provision would reduce snakebite deaths in India.
| Alexander the Great invaded India in 326 BC, and was greatly impressed by the skill of Indian physicians; especially in the treatment of snakebites [1]. Since then, India has remained notorious for its venomous snakes and the effects of their bites. With its surrounding seas, India is inhabited by more than 60 species of venomous snakes – some of which are abundant and can cause severe envenoming [2]. Spectacled cobra (Naja naja), common krait (Bungarus caeruleus), saw-scaled viper (Echis carinatus) and Russell's viper (Daboia russelii) have long been recognised as the most important, but other species may cause fatal snakebites in particular areas, such as the central Asian cobra (Naja oxiana) in the far north-west, monocellate cobra (N. kaouthia) in the north-east, greater black krait (B. niger) in the far north-east, Wall's and Sind kraits (B. walli and B. sindanus) in the east and west and hump-nosed pit-viper (Hypnale hypnale) in the south-west coast and Western Ghats [2].
Joseph Fayrer of the Indian Medical Service first quantified human snakebite deaths in 1869 for about half of “British India” (including modern Pakistan, Bangladesh and Burma), finding that 11,416 people had died of snakebites [3]. Subsequent estimates of human deaths from snakebite prior to Indian Independence ranged from 7,400 to 20,000 per year [4]–[6]. Government of India hospitals from all but six states reported only 1,364 snakebite deaths in 2008 [7] but this is widely believed to be an under-report as many victims of snakebite choose village-based traditional therapists and most die outside government hospitals. Community-based surveys in some localities have shown much higher annual mortality rates, ranging widely from 16.4 deaths/100,000 in West Bengal [8] to 161/100,000 in the neighbouring Nepal Terai [9]. However, such focal data cannot be extrapolated to provide national or even state totals because of the heterogeneity of snakebite incidence. These uncertainties have resulted in indirect estimates of annual snakebite mortality in India that varied from approximately 1,300 to 50,000 [6], [7], [10]–[13].
To fill this gap in knowledge, we estimated snakebite deaths directly from a large continuing study of mortality in India.
Ethics approval for the Million Deaths Study (MDS) was obtained from the Post Graduate Institute of Medical Research, St. John's Research Institute and St. Michael's Hospital, Toronto, Ontario, Canada [14]–[15].
Most deaths in rural India take place at home without prior attention by any qualified healthcare worker, so most causes are not medically certified [14]–[15]. Other approaches are therefore needed to help determine the probable causes of such deaths. The Registrar General of India (RGI) organises the Sample Registration System (SRS), which monitors all births and deaths in a nationally representative selection of 1.1 million homes throughout all 28 states and seven union territories of India. India was divided into approximately one million areas for the 1991 census, each with about 1,000 inhabitants. In 1993, the RGI randomly selected 6,671 of these areas to be represented in the SRS. Household characteristics were recorded and then enumerated twice yearly thereafter, documenting new births and deaths, but not the causes of death [16].
Since 2002, one of 800 non-medical field staff (trained by the RGI in appropriate fieldwork methods) visited each SRS area every six months to record a written narrative (in the local language) for each death from families or other reliable informants. In addition to the narratives, answers to standard questions about the deaths were also recorded in the field report. Fieldwork quality control methods were used routinely, including random re-sampling by teams reporting directly to the study investigators [14], [15]. This survey is part of the MDS, which seeks to assign causes to all deaths in SRS areas for the period between 2001–14 [14]–[16], [17]–[19].
These field reports, or ‘verbal autopsies’, were emailed randomly (based on the language of the narrative) to at least two of 130 collaborating physicians trained in disease coding. Physicians worked independently to assess the probable underlying cause of death, assigning each case a three-character International Classification of Diseases (ICD; 10th revision) code [20]. Any differences between the two coders were resolved by anonymous reconciliation between them (asking each to reconsider) or, for persisting differences, adjudication by a third physician (3% or 15/562 of snakebite deaths, and 18% or 22,845/122,848 of all deaths). The physician coders' training and their written guidelines (available online [21]) instructed them to use their best medical judgement to determine the most probable cause of death. Field reports could not be collected on 12% of the identified deaths due to migration or change of residence. As these missing deaths were mostly random, a systematic misclassification in cause of death was unlikely. We used logistic regression to quantify the odds of snakebite versus other deaths for gender, state, religion, education, occupation, place of death and season. Risk is measured compared to the reference group of lowest risk for each variable. Climate data on rainfall and temperature were obtained for each state from the India Meteorological Department [22]–[23].
The proportion of cause specific deaths in each age category was weighted by the inverse probability of household selection within rural and urban sub divisions of each state, to account for the sampling design [16]. Using methods described earlier [14]–[15], [17]–[19], the resulting proportion of deaths from each cause was applied to the United Nations (UN) population division estimates of deaths in India in 2005 [24] (9.8 million, upper and lower limits 9.4–10.3 million) to generate cause- specific death totals and rates. The UN totals (which undergo independent demographic review [24]) were used because the SRS underestimates adult mortality rates by about 10% [25]–[26]. The UN totals are not affected by the 12% of the SRS-enumerated deaths that were unavailable for interview in our survey. Totals for 2005 were used because they: (i) were most complete; (ii) could be compared to the available Indian Census projections for 2006; and (iii) captured information prior to the implementation of a new national health program in rural areas [27]. However, applying the 2001–03 proportions to the 2005 total deaths did not introduce major biases since there was little change in the yearly distribution in snakebite deaths in our survey, or in the annual number of deaths reported from snakebites in routine national hospital surveillance data between 2003 and 2008 [7].
Of the 643 deaths coded by physicians as ICD-10 codes X20–X29 (contact with venomous animals and plants), 523 (81%) were coded as X20 (venomous snakes) and review of these yielded no misclassified causes. Central re-examination of the symptoms and key words found 39 of 45 deaths coded as X27 (animals) and X29 (uncertain) to be snakebite deaths. We excluded 75 deaths coded as X21–X25 (various arthropods), X26 (marine organisms) and X28 (plants).
Among all 122,848 deaths, 2,179 of the deaths that were randomly chosen to be re-interviewed by independent teams were eventually matched to the identical houses and individuals of the MDS. Of the 2,179 re-sampled deaths, 9 were coded as snakebites, and 7 of these were found in the MDS. Thus, the sensitivity and specificity of the SRS field survey, assuming the re-sample deaths are the standard comparison, was 78% (7/9) and 100% (2,170/2,170), respectively.
A total of 562 of the 122,848 deaths (0.47% weighted by sampling probability or 0.46% unweighted) were from snakebites (Table 1). Almost all snakebite deaths (544 or 97%) were in rural areas. More men (330, 59%) than women (232, 41%) died from snakebites (overall ratio of 1.4 to 1). The proportion of all deaths from snakebites was highest at ages 5–14 years. Only 23% (127/562) of the deaths occurred in a hospital or other healthcare facility.
Expressed as national totals, snakebites caused 45,900 deaths in India in 2005 (99% CI 40,900 to 50,900). The age-standardised death rate per 100,000 population per year was 4.1 (99% CI 3.6–4.5) nationally and was 5.4 (99% CI 4.8–6.0) in rural areas.
Figure 1 shows the odds ratios for snakebite deaths versus other deaths, adjusted for age, gender, and for high prevalence states (13 states with age-standardised snakebite death rates greater than 3 per 100,000) versus other states. The risks of snakebite deaths were significantly increased among Hindus and farmers/labourers, deaths occurring outside home, and during the monsoon months of June to September (Figures 1 and 2). In contrast, gender and education were not significantly associated with risk of snakebite death. About 5,000–7,000 snakebite deaths per month occurred during the monsoon period, compared to less than 2,000 deaths in the winter months. Monthly numbers of snakebite deaths correlated with rainfall (R = 0.93, p<.0001) and mean minimum temperature (R = 0.80, p = 0.0017), but not with mean maximum temperature (R = 0.35, p = 0.2585; Figure 2).
Annual age-standardised mortality rates per 100,000 from snakebite varied between states, from 3.0 (Maharashtra) to 6.2 (Andhra Pradesh) in the 13 states with highest prevalence (average 4.5) compared to 1.8 in the rest of the country (Table 2; Figure 3). Total deaths were highest in Uttar Pradesh (8,700), Andhra Pradesh (5,200), and Bihar (4,500). The age and gender of snakebite deaths also varied by region, although these differences were not significant due to the small numbers of snakebite deaths in each state. Deaths at ages 5–14 years were prominent in the states of Jharkhand and Orissa, whereas deaths at older ages were prominent in Andhra Pradesh, Bihar, Madhya Pradesh, and Uttar Pradesh (data not shown). In Bihar, Madhya Pradesh, Maharashtra and Uttar Pradesh, female deaths exceeded male deaths (Table 2).
Snakebite remains an important cause of accidental death in modern India, and its public health importance has been systematically underestimated. The estimated total of 45,900 (95% CI 40,900–50,900) national snakebite deaths in 2005 constitutes about 5% of all injury deaths and nearly 0.5% of all deaths in India. It is more than 30-fold higher than the number declared from official hospital returns [7]. The underreporting of snake bite deaths has a number of possible causes. Most importantly, it is well known that many patients are treated and die outside health facilities – especially in rural areas. Thus rural diseases, be they acute fever deaths from malaria and other infections [19] or bites from snakes or mammals (rabies; [28]), are underestimated by routine hospital data. Moreover, even hospital deaths may be missed or not reported as official government returns vary in their reliability, as shown from a study of snakebites in Sri Lanka [29]. The true burden of mortality from snakebite revealed by our study is similar in magnitude to that of some higher profile infectious diseases; for example, there is one snakebite death for every two AIDS deaths in India [18]. Consequently, snakebite control programmes should be prioritised to a level commensurate with this burden.
Our data suggest underestimation in recent global estimates of mortality from snakebite deaths [10]: the upper bounds of recent annual estimates were 94,000 deaths globally and 15,000 deaths in India. This total for India is only about one-third of the snake bite deaths detected in our study. The incidence of snakebite deaths per 100,000 population per year in a recent community-based study in Bangladesh was similar to ours [30], suggesting that much of South Asia might have thousands more snakebite deaths than is currently assumed. Considering the widely accepted gross underestimation of snakebite deaths in Africa [11], it seems highly probable that well over 100,000 people die of snakebite in the world each year.
A minimal number of non-fatal snakebites in India may be estimated with far less certainty. Indian data from routine public sector hospitals [7] are clearly under-reports of deaths (recording only 1 in 5 of the deaths we estimated to have occurred in hospital). Nonetheless, the ratio of non-fatal bites (about 140,000) to fatal bites (about 2,200) in these hospital data from 2003–08 (about 64∶1) is informative of the relative burden of bites to deaths. Very crudely, even if we halve the fatal/nonfatal bite ratio to 32, this would suggest at least 1.4 million non-fatal bites corresponding to the 45,000 fatal bites. The actual number of non-fatal bites in India may well be far higher, as the community-based study in Bangladesh found about 100 non-fatal bites for each death [30].
Our study has limitations; notably the misclassification of snakebite deaths. However, snakebites are dramatic, distinctive and memorable events for the victim's family and neighbours, making them more easily recognizable by verbal autopsy. We observed a reasonably high sensitivity and specificity when compared to re-sampled deaths. Confusion with arthropod bites and stings is unlikely because of the different circumstances, size and behaviour of the causative animal and the course of envenoming. For example, most deaths from hymenoptera stings result from rapidly evolving anaphylaxis. Kraits (important agents of snakebite death in South Asia) may unobtrusively envenom sleeping victims, who may die after developing severe abdominal pain, descending paralysis, respiratory failure and convulsions [31]. Such deaths might not be associated with snakebite at all. These examples suggest possible underestimation of deaths in our data.
Since the numbers of deaths observed in each state were small, the estimated totals for each state are uncertain. However, the state distribution is broadly consistent with that reported by the RGI survey of deaths in selected rural areas in the 1990s [32]. The marked geographic variation across states in our study is similar to that in a country-wide survey conducted during the period 1941–45, which identified Bengal, Bihar, Tamil Nadu, Uttar Pradesh, Madhya Pradesh, Maharashtra and Orissa as having the highest death rates from snakebite [6]. Moreover, despite the obvious underestimates in hospitalised data [7], their geographical distribution of bites and deaths were similar to what we observed from household reports of deaths. The marked differences in snakebite mortality between states of India may be attributable to variations in human, snake and prey populations, and in local attitudes [8] and health services. The 13 states with the highest snakebite mortality are inhabited by the four most common deadly venomous snakes: Naja naja, Bungarus caeruleus, Echis carinatus and Daboia russelii. With the exception of E. carinatus, which favours open wasteland, these are widely distributed species of the plains and low hills where most Indians live. While some species can inhabit altitudes of up to 2,700 metres [2], this is exceptional and higher mountainous regions have considerably lower death rates.
As found in an earlier study [33], the peak age group of snakebite deaths is 15–29 years (25% or 142/562). However, the relative risk of dying from snakebite versus another cause was greater at ages 5–14 years. The peak age range and gender associated with snakebite mortality varied between states, perhaps reflecting differences in the relative numbers of children and women involved in agricultural work [34]–[35]. The slight excess among Hindus may reflect more tolerance of snakes and greater use of traditional treatments [2]. Snakebites and snakebite fatalities peak during the monsoon season in India [33], [36] and worldwide [10], probably reflecting agricultural activity, flooding, increased snake activity, and abundance of their natural prey.
Only 23% of the snakebite deaths identified in our survey occurred in hospital, consistent with an earlier study from five states [33]. This emphasises three points: (i) hospital-based data reflect poorly the national burden of fatal snakebites; (ii) inadequacy of current treatment of snakebite in India; and (iii) vulnerability of snakebite victims outside hospital. Practicable solutions include strengthening surveillance to allow a more accurate perception of the magnitude of the problem, improving community education to reduce the incidence of snakebites and speed up the transfer of bitten patients to medical care, improving the training of medical staff at all levels of the health service (including implementation of the new WHO guidelines [12]), and deployment of appropriate antivenoms and other interventional tools where they are needed in rural health facilities to decrease case fatality [36]–[38]. In addition, phylogenetic and venom studies are needed to ensure appropriate design of antivenoms to cover the species responsible for serious envenoming.
|
10.1371/journal.ppat.1000912 | The Plasmodium falciparum-Specific Human Memory B Cell Compartment Expands Gradually with Repeated Malaria Infections | Immunity to Plasmodium falciparum (Pf) malaria is only acquired after years of repeated infections and wanes rapidly without ongoing parasite exposure. Antibodies are central to malaria immunity, yet little is known about the B-cell biology that underlies the inefficient acquisition of Pf-specific humoral immunity. This year-long prospective study in Mali of 185 individuals aged 2 to 25 years shows that Pf-specific memory B-cells and antibodies are acquired gradually in a stepwise fashion over years of repeated Pf exposure. Both Pf-specific memory B cells and antibody titers increased after acute malaria and then, after six months of decreased Pf exposure, contracted to a point slightly higher than pre-infection levels. This inefficient, stepwise expansion of both the Pf-specific memory B-cell and long-lived antibody compartments depends on Pf exposure rather than age, based on the comparator response to tetanus vaccination that was efficient and stable. These observations lend new insights into the cellular basis of the delayed acquisition of malaria immunity.
| Plasmodium falciparum (Pf) is a mosquito-borne parasite that causes over 500 million cases of malaria annually, one million of which result in death, primarily among African children. The development of an effective malaria vaccine would be a critical step toward the control and eventual elimination of this disease. To date, most licensed vaccines are for pathogens that induce long-lived protective antibodies after a single infection. In contrast, immunity to malaria is only acquired after repeated infections. Antibodies play a key role in protection from malaria, yet several studies indicate that antibodies against some Pf proteins are generated inefficiently and lost rapidly. The cells that are responsible for the maintenance of antibodies over the human lifespan are memory B-cells and long-lived plasma cells. To determine how these cells are generated and maintained in response to Pf infection, we conducted a year-long study in an area of Mali that experiences a six-month malaria season. We found memory B-cells and long-lived antibodies specific for the parasite were generated in a gradual, step-wise fashion over years despite intense Pf exposure. This contrasts sharply with the efficient response to tetanus vaccination in the same population. This study lends new insights into the delayed acquisition of malaria immunity. Future studies of the cellular and molecular basis of these observations could open the door to strategies for the development of a highly effective malaria vaccine.
| To date, most successful vaccines have targeted pathogens that induce long-lived protective antibodies after a single infection, such as the viruses that cause smallpox, measles and yellow fever [1]. It has proved more difficult to develop highly effective vaccines against pathogens that do not induce sterile immunity such as the human immunodeficiency virus type-1 (HIV-1), Mycobacterium tuberculosis (Mtb), and Plasmodium falciparum malaria [2]. However, unlike HIV-1 and Mtb, clinical immunity to malaria can be acquired, but only after years of repeated Pf infections [3]. Passive transfer studies indicate that antibodies ultimately play a key role in protection from malaria [4], yet several studies show that antibodies to Pf antigens are inefficiently generated and rapidly lost in the absence of ongoing exposure to the parasite (reviewed in [5]). Elucidating the cellular basis of the inefficient acquisition of malaria immunity may ultimately prove critical to the design of an effective malaria vaccine.
Despite the key role that antibodies play in protection from a variety of infectious diseases, remarkably little is known about the cellular basis of acquiring humoral immunity in response to natural infections in humans. This gap in our knowledge is due in large part to the difficulty in studying natural infections in humans when we cannot predict who within a population will be infected with a given pathogen at a given time. Thus, our current understanding of the acquisition of immunity is largely derived from animal models and studies of humans after vaccination. These studies have established that long-lived, antibody-based immunity requires the generation and maintenance of memory B cells (MBCs) and long-lived plasma cells (LLPCs) (reviewed in [6], [7]). This process begins when naïve B cells bind antigen near the interface of B and T cell areas of secondary lymphoid organs. Several studies suggest that high-affinity binding drives naïve B cells to differentiate into short-lived, isotyped switched plasma cells (PCs) within the extra-follicular region which contributes to the initial control of infection. In contrast, lower affinity binding selects for entry of naïve B cells into follicles where germinal centers are formed. After a period of 7–10 days, through the CD4+ T-cell dependent process of somatic hypermutation, the germinal center reaction yields MBCs and LLPCs of higher affinity than the initial wave of short-lived plasma cells (SLPCs). MBCs recirculate and mediate recall responses after re-exposure to their cognate antigen by rapidly expanding and differentiating into PCs, whereas LLPCs residing in the bone marrow constitutively secrete antibody and provide a critical first line of defense against re-infection.
The mechanisms by which antibody responses are maintained over the human life-span remains an open question. In one model, LLPCs survive indefinitely in the bone marrow and independently maintain steady-state antibody levels [8]. Alternative models predict that PCs are replenished by MBCs that proliferate and differentiate in response to persistent [9] or intermittent exposure to antigen, and/or through non-specific by-stander activation (e.g. cytokines or TLR ligands) [10]. Unlike PCs, which are terminally-differentiated, MBCs may be maintained through homeostatic proliferation [11], possibly through exposure to polyclonal stimuli [10]. To address fundamental questions related to the generation and maintenance of MBCs and Abs specific for Pf malaria in children in malaria endemic areas, we conducted a year-long prospective study in a rural village of Mali that experiences an intense, sharply-demarcated six-month malaria season annually. We determined whether Pf infection generates MBCs specific for Pf blood stage antigens, and if so, whether they accumulate with age and cumulative Pf exposure, and also whether their frequency correlates with protection from malaria. In addition, we determined whether acute, symptomatic Pf infection resulted in an increase in the number of Pf-specific MBCs and the levels of specific antibodies, and if so, whether this increase remained stable over a six-month period of markedly reduced Pf transmission. By taking advantage of the tetanus immunization schedule in Mali in which infants and women of child-bearing age are vaccinated, we compared the relative efficiencies of the acquisition of tetanus toxoid (TT)- and Pf-specific MBCs and Ab, and also tested three hypotheses: 1) that growth of the MBC compartment depends on immunological experience rather than age, 2) that Pf infection induces non-specific activation of bystander B cells [12], [13], and 3) that polyclonal activation during heterologous immune responses is a general mechanism for maintaining MBCs and LLPCs [10].
In May 2006 we initiated an observational cohort study in Mali to investigate the mechanisms underlying naturally-acquired malaria immunity. A detailed description of the study site and cohort has been reported elsewhere [14]. The study population was an age-stratified, random sample representing 15% of all individuals living in a small, rural, well-circumscribed, non-migratory community where antimalarial drugs were provided exclusively by the study investigators. During a two-week period one month prior to the abrupt onset of the six-month malaria season, we enrolled 225 individuals in four age groups: 2–4 years (n = 73), 5–7 years (n = 52), 8–10 years (n = 51), and 18–25 years (n = 49). Attendance at scheduled follow-up visits was >99% for children (2–10 years) and 82% for adults (18–25 years) during the one-year study period indicating a high degree of study awareness and participation. For the MBC analysis reported here, a subset of 185 individuals was randomly selected within each of the four age categories. All subsequent data and analysis refer to these 185 individuals. The baseline demographic and clinical characteristics of this subset are shown in Table 1, according to age group. As previously reported [14], only three of the characteristics shown in Table 1 were associated with decreased malaria risk in multivariate analysis—increasing age, sickle cell trait (HbAS), and asymptomatic Pf parasitemia at study enrollment. During the one-year study period there were 380 unscheduled clinic visits, during which 219 cases of malaria were diagnosed, five of which met the WHO criteria for severe malaria [15]. Malaria episodes were defined as an axillary temperature ≥37.5°C, Pf asexual parasitemia ≥5000 parasites/µL, and a non-focal physical exam by the study physician. As expected in this region of Mali, all malaria cases were confined to a six-month period that began in July, peaked in October, and ended by January (Fig. 1A). The incidence of malaria and the proportion of individuals experiencing at least one malaria episode decreased with age, whereas the time to the first malaria episode increased with age (Table 2 and Fig. 1B). Thus, despite intense annual Pf transmission at this study site, malaria immunity is acquired slowly.
We first established baseline levels of IgG+ AMA1-, MSP1- and TT-specific MBCs and Abs in Pf-uninfected, healthy children and adults in May just before the malaria season, a point at which there had been little to no Pf transmission for five months. For this analysis we excluded individuals with asymptomatic Pf parasitemia (8.7% of total cohort; Table 1), because they showed a decreased risk of malaria and tended toward higher frequencies of AMA1- and MSP1-specific MBCs and levels of Ab (data not shown).
We focused our analyses on MBCs and Abs specific for Pf blood-stage antigens because humoral responses are known to be critical to blood-stage immunity [4]. We examined the response to two blood stage proteins, Apical Membrane Antigen 1 (AMA1) and Merozoite Surface Protein 1 (MSP1), because we had previously studied the MBC and Ab responses to these antigens in vaccine trials of Pf-naïve individuals [16]. This afforded the opportunity to compare the acquisition of B cell memory to the same antigens after vaccination versus natural Pf infection. We express MBC data as ‘MBCs per 106 PBMCs’, where ‘MBCs’ refers to the number of antibody secreting cells derived from MBCs during the six-day culture, and ‘106 PBMCs’ refers to the number of PBMCs after culture. In the present study, the mean frequency of AMA1-specific MBCs per 106 PBMCs increased with age (Fig. 2A; 2–4 yr: 1.2 [95% CI: 0.45–1.9]; 5–7 yr: 5.0 [95% CI: −0.2–10.1]; 8–10 yr: 8.9 [95% CI: 4.9–12.9]; 18–25 yr: 37.8 [95% CI: 10.4–65.3]; P<0.001), as did the proportion of individuals with detectable AMA1-specific MBCs (2–4 yr: 8.1%; 5–7 yr: 30.8%; 8–10 yr: 50.0%; 18–25 yr: 54.8%; P<0.001). Similarly, AMA1-specific Ab levels and the proportion of individuals seropositive for AMA1-specific Abs increased with age (Fig. 2A; P<0.001 for both comparisons). There was a positive correlation between the frequency of AMA1-specific MBCs and Ab levels (Spearman's correlation coefficient = 0.35; P = 0.005; Fig. S1).
We observed a similar age-associated increase in the frequency of MSP1-specific MBCs, although the overall frequency was lower than that for AMA1-specific MBCs (Fig. 2B; 2–4 yr: 1.2 [95% CI: 0.55–1.9]; 5–7 yr: 3.2 [95% CI: 1.2–5.2]; 8–10 yr: 5.9 [95% CI: 2.9–9.0]; 18–25 yr: 10.3 [95% CI: 6.3–14.3]; P<0.001). Likewise, the proportion of individuals who had detectable MSP1-specific MBCs (2–4 yr: 9.1%; 5–7 yr: 27.8%; 8–10 yr: 34.3%; 18–25 yr: 47.6%; P = 0.001) was similar to that for AMA1. MSP1-specific Ab levels and the proportion of individuals seropositive for MSP1-specific Abs also increased gradually with age (Fig. 2B; P<0.001 for both comparisons). There was a positive correlation between the frequency of MSP1-specific MBCs and Ab levels (Spearman's correlation coefficient = 0.34; P = 0.004; Fig. S1). Remarkably, despite exposure to 50–60 infective mosquito bites per month at the peak of each malaria season in this area [17], only approximately half of adults had detectable MBCs specific for AMA1 and MSP1, even though most had detectable AMA1- and MSP1-specific antibodies. Of the 72 individuals without detectable AMA1-specific MBCs before the malaria season, 64 (88.9% [95% CI 79.3–95.1]) did not have detectable MSP1-specific MBCs, suggesting that failure to generate MBCs to one Pf antigen is associated with failure to generate MBCs to other Pf antigens.
To understand if the expansion of Pf-specific MBCs with age was driven by repeated exposure to Pf antigens or simply a function of age, we determined the frequency of MBCs specific for an unrelated antigen, tetanus toxoid (TT), with age. In Mali, a single TT vaccine is administered to infants less than six months of age and a second TT vaccine is administered to females around 15 years of age to prevent neonatal tetanus. Thus, we measured TT-specific antibody and MBC responses at least 18 months after TT vaccination, a point at which the TT-specific response is likely to be at steady state. In contrast to what was observed for AMA1- and MSP1-specific MBCs, the frequency of TT-specific MBCs among males did not change significantly from age 2 to 25 years (Fig. 2C) (2–4 yrs: 10.8 [95% CI −7.4–29.0], 5–7 yrs: 7.3 [95% CI 0.7–13.9], 8–10 yrs: 8.0 [95% CI 3.1–12.8], 18–25 yrs: 4.7 [95% CI 1.4–8.1]; P = 0.80). Similarly, the proportion of male adults who were positive for TT-specific MBCs did not differ significantly from male children (2–4 yrs: 25.0%, 5–7 yrs: 33.3%, 8–10 yrs: 40.9%, 18–25 yrs 28.6%; P = 0.80). The slightly higher frequency of TT-specific MBCs in male versus female children was not statistically significant. However, the frequency of TT-specific MBCs was significantly higher in female adults compared to female children (Fig. 2C; mean frequency of TT-specific MBCs per million PBMC by age group (2–4 yrs: 2.9 [95% CI 1.1–4.7], 5–7 yrs: 3.2 [95% CI 0.2–6.1], 8–10 yrs: 3.4 [95% CI 1.1–5.7], 18–25 yrs: 58.7 [95% CI 34.2–83.3]; P<0.001) presumably the result of booster vaccination. Likewise, the proportion of female adults who were positive for TT-specific MBCs was significantly higher as compared to female children (2–4 yrs: 28.1%, 5–7 yrs: 25.0%, 8–10 yrs: 27.3%, 18–25 yrs 88.0%; P<0.001). For both females and males, TT-specific Ab levels mirrored MBC frequencies (Fig. 2C)—clearly increasing from female children to female adults (P<0.001), while not changing significantly by age in males (P = 0.44). Overall, TT-specific Ab levels and MBC frequencies correlated (Spearman's correlation coefficient = 0.48; P<0.001; Fig. S1). The observation that Pf-specific MBCs increased with age while TT-specific MBCs in individuals who received no booster vaccine did not increase and tended to decrease slightly with age indicates that the increase in Pf-specific MBCs is driven by repeated antigen exposure and is not simply a function of age. Of note, the size of the total IgG+ MBC compartment, as reflected in the peripheral blood, increased with age (Fig. 3; P<0.001), consistent with the maturation of the total MBC compartment with immunological experience.
To assess the Pf-specific MBC and Ab responses to acute malaria, and to determine the stability of this response during a period of little to no Pf transmission, we measured the frequencies of MBCs and Ab levels specific for AMA1 and MSP1 14 days after the first episode of malaria (convalescence), and in a cross-sectional survey at the end of the following dry season (month 12), and compared these frequencies to the pre-malaria season baseline (month 0; as detailed above). Malaria episodes were defined as an axillary temperature ≥37.5°C, Pf asexual parasitemia ≥5000 parasites/µL, and a non-focal physical exam by the study physician. Because few adults experienced malaria (Table 2), this analysis only included children aged 2–10 years (see Fig. 4 for sample sizes at each time point). The mean frequency of AMA1-specific MBCs in children aged 2–10 years increased from month 0 to convalescence (Fig. 4A; month 0: 4.7 [95% CI: 2.8–6.6]; convalescence: 13.4 [95% CI: 2.7–24.1; P = 0.006] and then decreased from convalescence to month 12 (Fig. 4A; month 12: 5.9 [95% CI: 2.4–9.4]; P = 0.93 versus convalescence) to a point just above the frequency at month 0 (Fig. 4A; P = 0.021, month 0 vs. month 12). Likewise, the level of AMA1-specific Ab increased from month 0 to convalescence (Fig. 4A; month 0: 422.8 [95% CI: 228.7–617.0]; convalescence: 797.2 [95% CI: 460.0–1134.7; P<0.001], and then decreased from convalescence to month 12 (Fig. 4A; month 12: 535.5 [95% CI: 283.8–787.2]; P<0.001 versus convalescence], to a point just above month 0 levels (Fig. 4A; P = 0.040, month 0 vs. month 12).
The MSP1-specific MBC and Ab responses followed a similar pattern. The mean frequency of MSP1-specific MBCs in children aged 2–10 years increased from month 0 to convalescence (Fig. 4B; month 0: 3.3 [95% CI: 2.0–4.6]; convalescence: 4.8 [95% CI: 2.9–6.8; P = 0.002] and then decreased from convalescence to month 12 (Fig. 4B; month 12: 4.5 [95% CI: 2.4–6.6]; P = 0.71 versus convalescence) to a point just above the frequency at month 0 (Fig. 4B; P = 0.156, month 0 vs. month 12). Likewise, the level of MSP1-specific Ab increased from month 0 to convalescence (Fig. 4B; month 0: 14.6 [95% CI: 10.5–18.6]; convalescence: 302.6 [95% CI: 111.7–493.4; P<0.001], and then decreased from convalescence to month 12 (Fig. 4B; month 12: 31.1 [95% CI: 5.5–56.6]; P<0.001 versus convalescence], to a point just above month 0 levels (Fig. 4B; P = 0.052, month 0 vs. month 12).
To determine if malaria induces non-specific activation of ‘bystander’ MBCs, we compared the frequencies of TT-specific MBCs and Ab levels before the malaria season (month 0) to that 14 days after acute malaria (convalescence). We observed a small, but statistically significant increase in the frequency of TT-specific MBCs from month 0 to convalescence (Fig. 4C; month zero: 7.1 [95% CI: 3.1–11.2]; convalescence: 8.4 [95% CI: 5.0–11.8; P = 0.012) that did not change significantly at month 12 (month 12: 9.1 [95% CI: 3.2–15.4]; P = 0.974 versus convalescence]. In contrast, TT-specific Ab levels decreased slightly from month 0 to convalescence, and again from convalescence to month 12, although neither decline was statistically significant (Fig. 4C; month 0: 0.58 [95% CI: 0.5–0.7]; convalescence: 0.57 [95% CI: 0.5–0.7; P = 0.063]; month 12: 0.54 [95% CI: 0.4–0.6]; P = 0.525 versus convalescence). Collectively these results indicate that malaria infection results in an increase in the frequencies of both Pf-specific, and bystander MBCs. However, malaria selectively induces Pf-specific Ab production but does not appear to drive the differentiation of bystander naïve and memory B cells into PCs.
By FACS we determined the proportion of B cell subsets in individuals (2–4 yrs [n = 38], 5–7 yrs [n = 21], 8–10 yrs [n = 23], 18–25 yrs [n = 27]) before the malaria season (Fig. 5A). With increasing age, and as a percentage of total CD19+ B cells we observed a decrease in immature B cells (CD19+ CD10+; P<0.001) and naïve B cells (CD19+ CD27− CD21+ CD10−; P = 0.047) and an increase in resting IgG+ MBCs (CD19+ CD27+ CD21+; P<0.001) and activated IgG+ MBCs (CD19+ CD27+ CD21− CD20+ CD10−; P<0.001). The increase with age of classical MBCs is consistent with the increase in total IgG+ MBCs we observed using the MBC ELISPOT assay (Fig 3).
In a subset of 87 individuals from this same study cohort, we previously reported that Pf exposure is associated with an expanded subset of ‘atypical’ MBCs that express FCRL4 and are hyporesponsive to in vitro stimuli [18], similar to the ‘exhausted’ MBCs described in viremic, HIV-infected individuals [19]. Atypical MBCs are defined as CD19+ CD27− CD21− CD20+ CD10− and typically represents <4% of circulating CD19+ B cells in healthy U.S. adults [19]. Here, analyzing a larger number of individuals in the cohort, we confirmed that this subset of MBCs is expanded in Malian children and adults compared to malaria-naïve U.S. adults (U.S. adults: 1.4% [95% CI: 0.9–1.8]; Malian children aged 2–10 years: 10.2% [95% CI: 8.7–11.8], P<0.001 versus U.S. adults; Malian adults aged 18–25 years: 14.8% [95% CI: 11.0–19.1], P<0.001 versus U.S. adults). Thus, in addition to the increase in classical MBCs, an ‘atypical’ MBC subset is expanding with age in this study population.
We investigated the impact of acute malaria on the relative proportion of B cells in each subset in children aged 2–10 years. Compared to the pre-malaria season baseline (month 0), there were no significant changes in the percent of lymphocytes that were CD19+ 14 days after acute malaria. Within the CD19+ B cell population there were no significant changes in the percent of immature B cells, naïve B cells, or resting MBCs, after acute malaria. Moreover, there was no change in the proportions of resting and atypical MBCs that were IgG+. However, we observed a decrease in the percentage of total atypical MBCs (Fig. 5B; month 0: 10.9% [95% CI: 9.4–12.4], convalescence: 8.7% [95% CI: 7.3–10.2]; P = 0.027), and an increase in activated MBCs following acute malaria (month 0: 1.6 [95% CI: 1.2–2.0], convalescence: 1.9 [95% CI: 1.4–2.4]; P = 0.09). Within the activated MBC subset there was a significant increase in the proportion that were IgG+ (month 0: 59.0% [95% CI: 56.0–62.1], month 12: 62.8% [95% CI: 59.2–66.3]; P<0.001). The decrease in the proportion of atypical MBCs in the peripheral blood suggests that this subpopulation may be trafficking out of the circulation into tissues in response to acute malaria.
We determined prospectively whether AMA1- or MSP1-specific Ab levels or MBC frequencies measured just prior to the six month malaria season were associated with the subsequent risk of malaria. For this analysis a malaria episode was defined as an axillary temperature ≥37.5°C, Pf asexual parasitemia ≥5000 parasites/µL, and a non-focal exam by the study physician. Because the incidence of malaria was very low in adults during the study period (Table 2), they were excluded from this analysis. Three measures of malaria risk were analyzed: 1) whether or not malaria was experienced, 2) the incidence of malaria, and 3) the time to the first malaria episode. In the corresponding multivariate regression models (logistic, Poisson, and Cox regression) which controlled for age, sickle cell trait, and concurrent asymptomatic Pf parasitemia, we found no correlation between malaria risk and AMA1- or MSP1- specific Ab levels or MBC frequencies. As discussed below, this finding was not unexpected based on the observation that the malaria vaccine candidates AMA1 and MSP1 did not confer protection against malaria in clinical trials [20], [21].
In this year-long prospective study of children and adults in an area of intense, annual, sharply demarcated Pf transmission, we show that MBCs specific for Pf can be acquired, but only gradually in a stepwise fashion over years of repeated Pf exposure. MBCs specific for two Pf antigens, AMA1 and MSP1, increased in frequency in response to acute Pf infection, and then contracted during a six-month period of decreased Pf exposure to a point slightly above pre-infection levels. Cross-sectional analysis of individuals aged 2–25 years just before the malaria season indicated that this step-wise, incremental increase in Pf-specific MBCs with each malaria season contributes to the gradual expansion of the Pf-specific MBC compartment with cumulative Pf exposure. By comparison, the stable frequency of TT-specific MBCs with age after immunization in infancy indicates that growth of antigen-specific MBC compartments does not simply occur with age, but requires repeated antigen exposure. We do not formally know if the gradual gain in Pf-specific MBCs is in fact due to an increase in long-lived MBCs, or whether those MBCs require Pf-stimulation and would be lost if Pf transmission did not resume after the six-month dry season. In another setting, namely in an area of Thailand with low Pf transmission, Wipasa et al. [22] recently reported that nearly half of adults studied had acquired long-lived Pf-specific MBCs as a result of infrequent malaria infections. It will be of genuine interest to understand the cellular and molecular mechanisms at play in the generation of MBCs under these very different conditions of exposure of children versus adults as these could have significance with regard to vaccine development. Moreover, recent studies in mouse models are revealing multiple, phenotypically and functionally distinct populations of MBCs [23], [24] and it will be of interest to further characterize Pf-specific MBCs in different malaria endemic settings.
The study described here provides a rare view of the acquisition and maintenance of human B cell memory. Most prospective studies of human B and T cell immunological memory have evaluated responses to vaccination rather than natural infection, in part because of the difficulty of predicting who within a population will be infected with a given pathogen at a given time. In response to a single vaccination, several studies have described an expansion and contraction of vaccine-specific MBCs [25], [26] and CD8+ memory T cells [27]. In one of the few longitudinal studies of the MBC response to natural infection, Harris et al. examined antigen-specific MBC responses of patients presenting with acute Vibrio cholerae infection, a pathogen that elicits long-term protective immunity against subsequent disease [28]. In contrast to our results, they observed that the majority of patients acquired IgA and IgG MBCs specific for two Vibrio cholerae antigens and that these persisted up to one year after infection.
Whereas MBCs mediate recall responses to reinfection by rapidly expanding and differentiating into PCs, LLPCs residing in the bone marrow constitutively secrete antibody in the absence of antigen and thus provide a critical first line of defense against reinfection [6]. Logistical constraints precluded the direct measurement of circulating PCs in this study. However, we took advantage of the discrete six-month dry season, a period of little to no Pf transmission, to infer the relative contributions of SLPCs and LLPCs to the Pf-specific IgG response based on a serum IgG half-life of ∼21 days [29]. Two weeks after acute malaria, AMA1- and MSP1-specific Ab levels increased significantly and then decreased over a six-month period to a point just above pre-infection levels, indicating that the majority of PCs generated in response to acute Pf infection were short-lived. This observation is consistent with previous studies that described rapid declines in Pf-specific Ab within weeks of an acute malaria episode [30], [31]. We infer that the small net increase in Pf-specific Ab at the end of the six-month dry season represents the acquisition of Pf-specific LLPCs. Because Pf transmission resumes after the six-month dry season, we cannot estimate the long-term decay rate of Pf-specific Ab in the absence of reinfection. It remains to be seen whether long-term decay rates of Pf-specific Ab are comparable to rates of Ab decay after exposure to common viral and vaccine antigens such as mumps and measles, for example, which elicit Ab with half-lives exceeding 200 years [32]. The small incremental gains in AMA1- and MSP1-specific Abs in response to acute malaria mirrors the gradual exposure-related increase in Pf-specific MBCs, consistent with the long-lived Abs being the products of LLPCs derived from MBCs. Unlike the response to some other pathogens, such as measles, which induce long-lived protective Abs after a single exposure, it may be that repeated exposure to the Pf parasite is necessary to ‘fill’ the Pf-specific LLPC compartment to the point where basal levels of circulating Abs to any given Pf antigen reach a protective threshold. In a separate study of this cohort, we observed a similar pattern of transient increases during the malaria season of Abs specific for a large number of Pf antigens using protein microarrays [33] suggesting that malaria induces a relatively high SLPC-to-LLPC ratio that is not exclusively a function of the inherent qualities of any given antigen per se.
In contrast to the highly efficient immune response to a single smallpox vaccination, which generates long-lived (>50 years) MBCs in nearly all vaccinees [34], a remarkably high proportion of adults in the present study did not have detectable AMA1- or MSP1-specific MBCs despite annual exposure to 50–60 infective mosquito bites per person per month at the height of the malaria season [17], similar to what Dorfman et al. observed in a cross-sectional study in Kenya [35]. Importantly, most female adults in the present study had detectable TT-specific MBCs three to ten years after a single TT booster vaccine in adolescence. We previously reported that AMA1- and MSP1-specific MBCs were reliably generated in Pf-naïve U.S. adults following just two vaccinations [16]. Taken together, these observations indicate that the relatively inefficient generation and/or maintenance of Pf-specific MBCs in response to natural Pf infection cannot be ascribed entirely to inherent deficiencies in the antigens themselves. Collectively, these observations raise a central question: if AMA1 and MSP1-specific MBCs and Abs can be efficiently generated by vaccination of Pf-naïve adults, and TT-specific MBCs and Abs can be efficiently generated by vaccination of Pf-exposed individuals in this cohort, what underlies the inefficient acquisition and/or maintenance of AMA1 and MSP1-specific MBCs and Abs in response to natural Pf infection? One simple answer, in addition to parasite antigenic variation [36], [37], might be that the enormous number of antigens encoded by the over 5,400 Pf genes overwhelms the immune system's capacity to select for and commit a sufficient number of MBCs and LLPCs specific for any given Pf antigen to a long-lived pool [38]. If immunity to clinical malaria requires high levels of antibodies to a large number of Pf proteins, then the inability to commit large numbers of MBCs and LLPCs specific for any given Pf antigen during any given infection, as shown here, may explain, in part, why malaria immunity is acquired slowly. In this scenario the Pf-infected individual is capable of the normal generation and maintenance of MBCs and LLPCs, but acquiring a sufficient number of MBCs and LLPCs to a large number of antigens may simply take years.
It is also possible that Pf infection disrupts the immune system's ability to generate or maintain MBCs or LLPCs. The differentiation of B cells into long-lived MBCs depends to a great extent on the affinity of their BCRs for antigen. Recently, evidence was presented that affinity maturation of B cells may fail to occur in the absence of adequate Toll-like receptor (TLR) stimulation [39]. We recently reported that Malian adults were relatively refractory to CpG, a TLR9 agonist incorporated into two subunit malaria vaccine candidates [40], raising the possibility that the slow acquisition of MBCs observed here may be due to a failure of B cells to undergo affinity maturation during Pf infection. Although our data do not directly address the role of apoptosis in the gradual acquisition of Pf-specific MBCs, it is worth noting that we found no evidence of Pf-induced ablation of Plasmodium-specific MBCs, as was observed in mice four days after Plasmodium yoelii infection [41]. The relatively inefficient response to natural Pf infection also does not appear to be due to a persistent, Pf-induced general immunosuppression as the frequency of TT-specific MBCs increased significantly in most adult females in response to a single TT booster vaccination, an increase that appeared to be maintained for years. In an experimental model of lymphocytic choriomeningitis virus (LCMV) infection, a high antigen-to-B cell ratio disrupted germinal center formation and the establishment of B cell memory [42]. It is plausible that a similar mechanism is at play during the blood stage of Pf infection when the immune system encounters high concentrations of parasite proteins. Indeed, germinal center disruption is observed in mice infected with P. berghei ANKA [43] and P. chabaudi [44]. It is also possible that specific parasite products selectively interfere with the regulation of B cell differentiation [45] or with the signals required for sustaining LLPCs in the bone marrow [46]. It is also conceivable that the disproportionately high level of class-switched SLPCs we observed in response to Pf infection arises from pre-diversified IgM+IgD+CD27+ (marginal zone) B cells—analogous to the rapid protective response against highly virulent encapsulated bacteria that do not elicit classical T-dependent responses [47]. These and other hypotheses could be tested by applying systems biology methods [48] and targeted ex vivo and in vitro assays to rigorously conducted prospective studies of Pf-exposed populations.
We previously reported that Pf exposure is associated with a functionally and phenotypically distinct population of FCRL4+ hypo-responsive atypical MBCs [18], similar to the ‘exhausted’ MBCs described in HIV-infected individuals [19]. In this study, with a larger sample size, we confirmed that Pf exposure is associated with an expansion of FCRL4+ MBCs. The accumulation of atypical MBCs could be linked to the slow acquisition of Pf-specific MBCs, as naïve B cells in response to Pf infection could have a propensity to differentiate into atypical rather than classical MBCs. We also observed that the FCRL4+ MBC population decreased in the peripheral circulation two weeks after acute malaria suggesting that these MBCs are directly involved in the response to Pf infection, possibly trafficking to secondary lymphoid tissues. Although the function of FCRL4+ MBCs is not established, Moir et al. [19] suggested that FCRL4+ ‘exhausted MBCs’ contribute to the B cell deficiencies observed in HIV-infected individuals. In contrast, Ehrhardt et al.[49], who first described FCRL4+ ‘tissue-like MBCs’ in lymphoid tissues associated with epithelium, suggested that these cells may play a protective role during infections. At present, the factors that underlie the expansion of atypical MBCs in this study population are not known. Genetic or environmental factors that are associated with Pf transmission but not accounted for in this study could explain this observation. It will be of interest to understand the origin, antigen-specificity, and function of FCRL4+ MBCs in the context of Pf infection and the potential impact of these MBCs on the ability of children to respond to malaria vaccines.
In multivariate analysis we found no correlation between the frequency of MBCs and levels of Abs specific for AMA1 or MSP1 and malaria risk. This is not necessarily unexpected in light of recent clinical trials that showed that vaccination with either AMA1 or MSP1 did not confer protection [20], [21]. Furthermore, we suspect that the frequency of MBCs per se may not reliably predict clinical immunity to malaria regardless of antigen specificity. Malaria symptoms only occur during the blood stages of Pf infection and can begin as early as three days after the blood stage infection begins [50].Because the differentiation of MBCs into PCs peaks ∼6–8 days after re-exposure to antigen [10], there may not be sufficient time for MBCs specific for Pf blood stage antigens to differentiate into the antibody-secreting cells that would prevent the onset of malaria symptoms. In contrast, the longer incubation period of other pathogens allows MBCs to differentiate into protective antibody-secreting cells before symptoms develop. For example, follow-up studies of hepatitis B vaccinees have shown that protection can persist despite the decline of hepatitis B-specific antibodies to undetectable levels [51], presumably due to the recall response of persistent MBCs. Thus, protection against the blood stages of malaria may depend on achieving and maintaining a critical level of circulating antibody that can rapidly neutralize the parasite. MBCs may contribute to the gradual acquisition of protective immunity by differentiating into LLPCs with each Pf infection.
Here we also provide evidence concerning the mechanism by which MBCs and LLPCs are maintained. We observed a modest but statistically significant increase in TT-specific MBCs two weeks after acute malaria, in support of the hypothesis that MBCs are renewed by polyclonal or ‘bystander’ activation [10]. The stable frequency of TT-specific MBCs with age suggests that the small increases associated with Pf-induced polyclonal activation are matched by the rate of loss of senescent TT-specific MBCs. It has also been proposed that non-specific polyclonal stimulation maintains long-lived Ab responses by driving MBCs to differentiate into SLPCs or LLPCs [10]. Similarly, it has been hypothesized that Plasmodium infection generates large amounts of non-specific Ig [52] through polyclonal B cell activation [12], [13]. However, despite the presence of TT-specific MBCs and their expansion following Pf infection, we did not observe a concomitant increase in TT-specific IgG. This finding is consistent with recent human studies that demonstrate a lack of bystander IgG production after heterologous vaccination or viral infection [32], [53]; as well as studies in mice that demonstrate PC persistence after MBC depletion [54], and the failure of MBCs to differentiate into PCs in vivo upon TLR4 and 9 activation [55]. This finding does not represent an overt inability of TT-specific MBCs to differentiate into PCs, since adult females in this study had a sharp increase in tetanus IgG after a single tetanus booster. It is possible that bystander MBCs specific for antigens other than TT differentiate into PCs after Pf infection, but based on the results of this study we hypothesize that the preponderance of IgG produced in response to malaria is specific for the ∼2400 Pf proteins expressed during the blood-stage of infection [56], and that increases in ‘non-specific’ IgG reflect boosting of cross-reactive B cells [57], [58]. From a basic immunology perspective, these data support a model in which non-specific stimuli contribute to MBC self-renewal, but not to the maintenance of LLPCs. Studies of other Ab specificities and isotypes before and after malaria and other infections would test this hypothesis further. Although a recent mouse study showed that MBCs do not proliferate in vivo after immunization with an irrelevant antigen [59], this may reflect the difference in requirements for MBC maintenance in mammals with relatively short life spans.
It is of general interest to determine which parasite products are responsible for the polyclonal activation of MBCs observed here. Studies in vitro suggest that Pf drives polyclonal MBC activation by the cysteine-rich interdomain regions 1α (CIDR1α) of the Pf erythrocyte membrane protein 1 (PfEMP1) [13], [60], but it is conceivable that Pf-derived TLR agonists [61], [62] or bystander T cell help [63], [64], [65] also contribute to MBC proliferation in the absence of BCR triggering [66].
Animal models have provided important insights into the immunobiology of Plasmodium infection [67], but ultimately, despite obvious experimental limitations, it is critical to investigate the human immune response to Pf in longitudinal studies since findings from animal models do not always mirror human biology or pertain to the clinical context [68], [69]. Key challenges for future studies will be to determine the molecular basis of the inefficient generation of MBCs and LLPCs in response to Pf infection and to determine the longevity of these cells in the absence of Pf transmission over longer periods of time. Greater insight into the molecular and cellular basis of naturally-acquired malaria immunity could open the door to strategies that ultimately prove useful to the development of a highly effective malaria vaccine.
The ethics committee of the Faculty of Medicine, Pharmacy, and Odonto-Stomatology, and the institutional review board at the National Institute of Allergy and Infectious Diseases, National Institutes of Health approved this study (NIAID protocol number 06-I-N147). Written, informed consent was obtained from adult participants and from the parents or guardians of participating children.
This study was carried out in Kambila, a small (<1 km2) rural village with a population of 1500, located 20 km north of Bamako, the capital of Mali. Pf transmission is seasonal and intense at this site from July through December. The entomological inoculation rate measured in a nearby village was approximately 50–60 infective bites per person per month in October 2000 and fell to near zero during the dry season [17]. A detailed description of this site and the design of the cohort study has been published elsewhere [14].
In May 2006, during a two-week period just prior to the malaria season, 225 individuals aged 2–10 years and 18–25 years were enrolled after random selection from an age-stratified census of the entire village population. Enrollment exclusion criteria were hemoglobin <7 g/dL, fever ≥37.5°C, acute systemic illness, use of anti-malarial or immunosuppressive medications in the past 30 days, or pregnancy. All analysis in the present study pertains to an age-stratified subset of individuals (n = 185) randomly selected from those who had complete sets of PBMC samples over the entire study period. From May 2006 through May 2007, participants were instructed to report symptoms of malaria at the village health center, staffed 24 hours per day by a study physician. For individuals with signs or symptoms of malaria, blood smears were examined for the presence of Pf. Patients with positive smear results (i.e. any level of parasitemia) were treated with a standard 3-day course of artesunate plus amodiaquine, following the guidelines of the Mali National Malaria Control Program. Anti-malarial drugs were provided exclusively by the study investigators. Children with severe malaria were referred to Kati District Hospital after an initial parenteral dose of quinine. For research purposes, a malaria episode was defined as an axillary temperature ≥37.5°C, Pf asexual parasitemia ≥5000 parasites/µL, and a nonfocal physical examination by the study physician. Severe malaria, as defined by the WHO [15], was included in this definition. Three clinical endpoints were used to evaluate the relationship between Pf-specific immune responses and malaria risk: 1) whether or not malaria was experienced, 2) the incidence of malaria, and 3) the time to the first malaria episode. Blood smears were prepared and venous blood samples collected during the two-week enrollment period (month 0), 14 days after the first episode of malaria (convalescence), and during a two-week period at the end of the six-month dry season (month 12). Hemoglobin was typed from venous blood samples. Stool and urine were examined at enrollment for the presence of helminth infections. Venous blood samples from ten healthy U.S. adult blood bank donors were analyzed as controls. Travel histories for these U.S. adults were not available, but prior exposure to Pf is unlikely.
Blood samples (8 ml for children and 16 ml for adults) were drawn by venipuncture into sodium citrate-containing cell preparation tubes (BD, Vacutainer CPT Tubes) and transported 20 km to the laboratory where they were processed within three hours of collection. Plasma and PBMCs were isolated according to the manufacturer's instructions. Plasma was stored at −80°C. PBMCs were frozen in fetal bovine serum (FBS) (Gibco, Grand Island, NY) containing 7.5% dimethyl sulfoxide (DMSO; Sigma-Aldrich, St. Louis, MO), kept at −80°C for 24 hours, and then stored at −196°C in liquid nitrogen. For each individual, PBMC and plasma samples from all time points were thawed and assayed simultaneously.
Thick blood smears were stained with Giemsa and counted against 300 leukocytes. Pf densities were recorded as the number of asexual parasites/µl of whole blood, based on an average leukocyte count of 7500/µl. Each smear was evaluated separately by two expert microscopists blinded to the clinical status of study participants. Any discrepancies were resolved by a third expert microscopist.
Hemoglobin was typed by high performance liquid chromatography (HPLC; D-10 instrument; Bio-Rad, Hercules, CA) as previously described [14].
At enrollment, duplicate stool samples were examined for Schistosoma mansoni eggs and other intestinal helminths using the semi-quantitative Kato-Katz method. To detect Schistosoma haematobium eggs, 10 ml of urine were poured over Whatman filter paper. One or two drops of ninhydrine were placed on the filter and left to air dry. After drying, the filter was dampened with tap water and helminths were eggs detected by microscopy.
Latitude and longitude coordinates of study subjects' households were measured by a handheld global positioning system receiver (GeoXM; Trimble) and reported earlier [14].
ELISAs were performed by a standardized method as described previously [70]. For both AMA1 and MSP1, a 1∶1 mixture of FVO and 3D7 AMA1 and MSP1 isotypes was used to coat the ELISA plates. The limit of detection for the AMA1 and MSP1 ELISA is based on the range of values that gives reproducible results at the Malaria Vaccine and Development Branch at NIAID where the assay is routinely performed. More specifically, the limit of detection is the ELISA unit value at the lowest point on the standard curve, multiplied by the dilution factor at which samples are tested. The minimal detection levels for the MSP1 and AMA1 ELISA assays were 11 and 33 ELISA units, respectively. For analysis, all data below the minimum detection level were assigned a value of one half the limit of detection (i.e. 6 units for MSP1, 17 units for AMA1). The limit of detection for the TT ELISA was not determined because we did not have access to TT-naïve serum.
Antigen-specific MBCs were quantified by a modified version of the method developed by Crotty et al [71]. We found that adding IL-10 to the cocktail of polyclonal activators resulted in a six-fold increase in the efficiency of the assay (Weiss et al., unpublished observation). Briefly, PBMCs were thawed and cultured in 24 well plates at 37°C in a 5% CO2 atmosphere for six days in media alone (RPMI 1640 with L-Glutamine, Penicillin/Streptomycin 100 IU/ml, 10% heat-inactivated FBS, 50 µM β-Mercaptoethanol) or media plus a cocktail of polyclonal activators: 2.5 µg/ml of CpG oligonucleotide ODN-2006 (Eurofins MWG/Operon, Huntsville, AL), Protein A from Staphylococcus aureus Cowan (SAC) at a 1/10,000 dilution (Sigma-Aldrich, St. Louis, MO), pokeweed mitogen at a 1/100,000 dilution (Sigma-Aldrich), and IL-10 at 25 ng/ml (BD Biosciences). Cells were washed and distributed on 96-well ELISPOT plates (Millipore Multiscreen HTS IP Sterile plate 0.45 um, hydrophobic, high-protein binding) to detect antibody-secreting cells (ASCs). ELISPOT plates were prepared by coating with either: a 10 µg/ml solution of polyclonal goat antibodies specific for human IgG (Caltag) to detect all IgG-secreting cells; a 1% solution of bovine serum albumin (BSA) as a non-specific protein control; or 5 µg/ml solutions of either tetanus toxoid (TT), AMA1, or MSP1 to detect antigen-specific ASCs. For AMA1 and MSP1, a 1∶1 mixture of FVO and 3D7 isotypes was used to coat the ELISPOT plates. Plates were blocked by incubation with a solution of 1% BSA. For the detection of antigen-specific ASCs, cells were plated in duplicate in eight serial dilutions beginning with 5×105 cells/well. For detection of total IgG ASCs cells were plated at six serial dilutions beginning at 4×104 cells/well. After a five hour incubation of the cells in the ELISPOT plates, plates were washed four times each in PBS and PBS-Tween 20 0.05% (PBST), and incubated overnight with a 1∶1000 dilution of alkaline phosphatase-conjugated goat antibodies specific for human IgG (Zymed) in PBST/1% FCS. Plates were washed four times each in PBST, PBS, and ddH2O; developed using 100 µl/well BCIP/NBT for 10 minutes; washed thoroughly with ddH2O and dried in the dark. ELISPOTS were quantified using Cellular Technologies LTD plate-reader and results analyzed using Cellspot software. Results are reported as frequencies of MBCs per 106 PBMCs after the six-day culture. The limit of detection of the MBC ELISPOT assay for this analysis was five ASCs per 106 PBMC based on the average number of ASCs on the BSA control. Assay failure was defined as fewer than 1000 IgG+ ASCs per 106 PBMCs after the six-day culture which resulted in the exclusion of 15% of individuals at month 0, 13.2% 14 days after the first malaria episode, and 7.3% at month 12. For individuals with a limited number of PBMCs, priority was given to performing the ELISPOT assay for MSP1, then TT, and then AMA1.
All phenotypic analyses were performed using mouse mAbs specific for human B cell markers conjugated to fluorophores as previously reported [18]. Fluorophore-conjugated mAbs specific for the following markers were used: PECy7-CD19, PE-CD20, APC-CD10, APC-CD27, PE-IgG (BD Biosciences, San Jose, CA) and FITC-CD21 (Beckman Coulter, Fullerton, CA). A four-color, two-stain strategy was used to identify B cell subsets as follows: plasma cells/blasts (CD19+ CD21− CD20−), naive B cells (CD19+ CD27−CD10−), immature B cells (CD19+ CD10+), classical MBCs (CD19+ CD27+ CD21+), atypical MBCs (CD19+ CD21− CD27− CD10−) and activated MBCs (CD19+ CD21− CD27+CD20+). FACS analyses were performed on a FACSCalibur flow cytometer (BD Biosciences) using FlowJo software (Tree Star, Ashland, OR).
Data were analyzed using STATA (StataCorp LP, Release 10.0) and GraphPad Prism for Windows (GraphPad Software, version 5.01).The Kruskal-Wallis test was used to compare continuous variables between groups, and the Fisher's exact test was used to compare categorical variables. The Wilcoxon matched-pairs signed-rank test was used to compare measurements of the same parameter at two time points for the same individual. The correlation between different continuous measures was determined by using the Spearman correlation coefficient. The malaria-free probability over the twelve-month study period was estimated by the Kaplan-Meier curve, and the time to the first malaria episode was compared by the log rank test. Cox's proportional hazards model was used to assess the effect of the following factors on the hazard of malaria: age, gender, weight, ethnicity, distance lived from study clinic, self-reported bednet use, hemoglobin type, antigen-specific MBC frequencies and Ab levels. The same list of variables was included in logistic and Poisson regression models to determine their impact on the odds and incidence of malaria episodes, respectively. For all tests, two-tailed p values were considered significant if ≤0.05.
|
10.1371/journal.pgen.1007659 | Ionizing radiation induces stem cell-like properties in a caspase-dependent manner in Drosophila | Cancer treatments including ionizing radiation (IR) can induce cancer stem cell-like properties in non-stem cancer cells, an outcome that can interfere with therapeutic success. Yet, we understand little about what consequences of IR induces stem cell like properties and why some cancer cells show this response but not others. In previous studies, we identified a pool of epithelial cells in Drosophila larval wing discs that display IR-induced stem cell-like properties. These cells are resistant to killing by IR and, after radiation damage, change fate and translocate to regenerate parts of the disc that suffered more cell death. Here, we report the identification of two new pools of cells with IR-induced regenerative capability. We addressed how IR exposure results in the induction of stem cell-like behavior, and found a requirement for IR-induced caspase activity and for Zfh2, a transcription factor and an effector in the JAK/STAT pathway. Unexpectedly, the requirement for caspase activity was cell-autonomous within cell populations that display regenerative behavior. We propose a model in which the requirement for caspase activity and Zfh2 can be explained by apoptotic and non-apoptotic functions of caspases in the induction of stem cell-like behavior.
| Ionizing Radiation (IR), alone or in combination with other therapies, is used to treat an estimated half of all cancer patients. Yet, we understand little about why some tumors cells respond to treatment while others grow back (regenerate). We identified specific pools of cells within a Drosophila organ that are capable of regeneration after damage by IR. We also identified what it is about IR damage that allows these cells to regenerate. These results help us understand how tissues regenerate after IR damage and will aid in designing better therapies that involve radiation.
| Regeneration is essential to tissue homeostasis and health. Conversely, regeneration of tumors after treatment leads to tumor recurrence and treatment failure. Understanding mechanisms that underlie regeneration is therefore important not only for understanding basic biology but also for optimizing treatment of diseases like cancer. Our understanding of regeneration has benefited immensely from experimental systems with dedicated stem cells that form the cellular basis for regeneration. Examples include regeneration of vertebrate gut and Drosophila intestine [1–3]. Tissues also regenerate despite the lack of a dedicated stem cell pool. A prime example is the vertebrate liver, which regenerates by proliferation of the surviving cells of each cell type [4–6]. If proliferation of hepatocytes is blocked during liver regeneration, however, biliary epithelial cells can dedifferentiate, proliferate and re-differentiate into hepatocytes [4–6]. Such plasticity has been documented in other mammalian organs [7–9], and in some models of amphibian limb and fish fin regeneration [10]. This report addresses the molecular basis for cell fate plasticity during regeneration using Drosophila larval cells as a model.
Drosophila larval imaginal discs are precursors of adult organs. Imaginal discs lack a dedicated stem cell pool yet can regenerate fully even after surgical ablation of 25% of the disc, after genetic ablation of a disc compartment (e.g. by expressing a pro-apoptotic gene in the anterior compartment), or after exposure to doses of ionizing radiation (IR) that kills about half of the cells [11, 12]. We recently identified a previously unknown mode of regeneration in Drosophila larval wing discs, whereby epithelial cells acquire stem cell-like properties during regeneration after damage by IR [13]. These properties include resistance to killing by IR, the ability to change cell fate, and the ability to translocate to areas of the wing disc with greater need for cell replenishment. The ability to behave like stem cells in response to IR is limited to certain cells within the continuous epithelium of the wing disc. Specifically, a subset of future hinge cells (see Fig 1P for the fate map in the wing disc) is protected from IR-induced apoptosis by the action of STAT92E (Drosophila STAT3/5, to be called ‘STAT’ hereafter) and by Wg (Drosophila Wnt1)-mediated repression of pro-apoptotic gene reaper [13]. These hinge cells lose the hinge fate and translocate to the pouch region that suffers more apoptosis and participate in regenerating the pouch. Without IR, these cells differentiate into the adult wing hinge, indicating that cell fate plasticity is IR-induced.
In above-described studies, regeneration of the pouch by the hinge was observed in nearly all irradiated discs [13, 14]. In about 20% of irradiated discs, we observed, in addition, abnormal regeneration that produced an ectopic wing disc [14]. Ectopic discs were wing discs based on staining for the protein markers Ubx and Wg, and were composed of an ectopic pouch and an ectopic hinge [14]. Ectopic discs were neither duplications (e.g. not pouch-to-pouch) nor transdeterminations (e.g. not leg-to-wing) described in classical studies of regeneration after surgical ablation [15]. Our efforts to dissect the cellular origin for the ectopic discs showed that cells of the hinge that regenerate the pouch are unlikely to be responsible for ectopic discs [14]. Therefore, we hypothesized that there are additional pools of cell in the wing disc that show stem cell like properties after IR damage by participating in abnormal regeneration to produce an ectopic wing disc.
Here, we report the mapping of cell lineages during regeneration of Drosophila larval wing discs following damage by X-rays, a type of IR. To express lineage tracers, we used FlyLight GAL4 drivers that display simple expression patterns because they use small (~3kb) enhancers from various genes [16]. In addition to the subset of future hinge cells we previously identified as capable of behaving like stem cells [13], two more cell populations, in the notum and in the dorsal-posterior hinge/pleura region, were found to show this potential. While the previously identified hinge cells are responsible for normal regeneration to restore the wing disc, the newly identified regenerative cells undergo abnormal regeneration to produce ectopic discs. Cells of the pouch, we find, lack the capacity for plasticity and do not change fate or translocate. Of possible consequences of X-ray exposure, we identified caspase activity as an essential determinant for inducing stem cell-like properties, and further localize this requirement to the regenerative cells. Zfh2, a transcription factor and a STAT effector, we found, was needed to regulate IR-induced caspase activity and for fate change. Cancer treatments including IR induce stem cell-like properties in non-stem cancer cells [17–20], but we understand little about what consequences of IR induces stem cell-like properties. This report describes similar phenomena in Drosophila cells and offers molecular insights into IR-induced cell plasticity.
We are using the published G-trace system to monitor cell lineages in the larval wing discs [21]. In this system, GAL4 drives the expression of UAS-RFP (real time expression, Fig 1N). GAL4 also drives the expression of UAS-FLP recombinase, which causes a recombination event to excise transcription ‘STOP’ sequences, resulting in stable GFP expression (lineage expression). Thus, even if cells change fate and lose RFP expression, their clonal descendants would be marked with GFP (becoming GFP+RFP-). We used G-trace to test a collection of GAL4 drivers, each active within a different subset of cells in Drosophila 3rd instar larval wing discs.
30A-GAL4 was used in our recent studies and is active in a subset of hinge cells and a few (fewer than 10) of notum cells (Fig 1A–1C and 1J–1L, arrows point to RFP+ notum cells; [13, 14]). The dynamics of 30A-GAL4 activity is such that cells expressing it show stable lineage; very few were GFP+RFP-. In contrast, another hinge driver, FlyLight R73G07-GAL4 produced GFP+ cell in most of the disc including the pouch, the hinge, and most of the notum, even though RFP is restricted to the hinge in 3rd instar wing discs (Fig 1D–1F). R73G07 is a 3028 bp enhancer from the zfh2 locus and is apparently active in most cells of the wing disc before becoming restricted to the hinge. Zfh2 is a transcription factor important for wing development [22]. Temporal restriction of R73G07-GAL4 activity with repressor GAL80ts according to the temperature shift protocol shown in Fig 1M confined GFP to the hinge and the pouch (Fig 1G–1I). Increasing larval age from 4–5 days to 5–6 days after egg deposition (AED) before temperature shift to induce GAL4 did not restrict the GFP+ domain further. Aging the larvae beyond 5–6 days AED before inducing GAL4 may help eliminate GFP in the pouch and restrict it to the hinge, but this schedule is incompatible with our goal because we need to monitor regeneration for 72 h after IR before losing the larvae to pupariation.
These results illustrate that while some GAL4 drivers show stable lineage expression and could be used to monitor fate changes after irradiation, others show lineage changes without IR. This was confirmed using fifteen additional FlyLight GAL4 drivers (S1 Fig). Therefore, we used GAL80ts and the protocol shown in Fig 1M in all subsequent experiments to identify regenerative cell populations, even if their lineages were stable as in the 30A-GAL4 example. Although we selected FlyLight drivers with apparently exclusive expression in the disc region of interest (https://flweb.janelia.org/), many, we found, show additional expression elsewhere in the wing disc and are unsuitable for lineage tracing (S1 Fig). Some GAL4 drivers were active in the cells of the peripodial membrane that covers the wing disc epithelium on the apical side, and in wing-disc associated tracheal cells on the basal side. In such cases, peripodial and tracheal cells could be identified based on their larger nuclear size compared to columnar epithelial cells and on their location in optical sections that book-end the columnar epithelium (S2 Fig). Our analyses focused on the columnar epithelium by excluding other optical sections.
Antibody staining shows that Zfh2 protein expression resembles R73G07-GAL4>RFP expression (compare Fig 1H and 1K). In contrast, 30A>RFP is expressed in only a subset of these cells (compare Fig 1K and 1L). Of relevance to subsequent sections is the expression of R73G07-GAL4 but not 30A-GAL4 in the dorsal-posterior hinge and the pleura (arrowheads in Fig 1J–1L and 1O–1P).
We used 4000R of X-rays for experiments and analyzed regenerated discs 72 h after irradiation. This level of IR kills more than half of the cells but the discs could still regenerate to produce viable adults [11, 12]. In our published studies of lineage tracing after irradiation, 30A-GAL4 expressing hinge cells translocated to the pouch but showed little movement dorsally towards the notum [13, 14]. In contrast, R73G07-GAL4>G-trace experiments show GFP+ cell populations that extend from the hinge dorsally along both the anterior and posterior margins of the wing disc (Fig 2). The extending GFP+ cell population is contiguous with both the hinge (arrowheads) and the pleura (white arrows, Fig 2F–2H and 2J–2L). Some GFP+ cells in the notum lack RFP (for example, Fig 2H arrow) while others express RFP (for example, Fig 2L arrow). In discs that show an ectopic disc (Fig 2M–2P, yellow arrows), many cells of the ectopic disc are GFP+RFP+. To better understand the source of GFP+ cells in the notum, we repeated the experiment but analyzed the discs at different times after IR.
We analyzed R73G07>G-trace larvae from the same cohort at 24, 48 and 72 h after IR, in two independent time course experiments (Fig 3). Wing discs, we found, fell into four categories depending on the abundance and location of GFP/RFP cells in the notum. Un-irradiated discs showed very few GFP/RFP cells in the notum (Fig 2A–2D, ‘category 1’). Category 2 discs showed GFP+RFP+ cells spreading dorsally from the hinge (arrowhead) and the pleura (arrow, Fig 3A and 3B, magnified in C). Category 2 predominates at 24 h after IR (Fig 3O). The presence of RFP in the hinge cells that had spread into the notum could be due to the persistence of GAL4, RFP, or both. The movement of hinge/pleura cells was seen along both the anterior and the posterior disc margins, but not in the central portion of the disc (Fig 3B). In category 3 discs, GFP+ cells in the notum increased in number compared to category 2, were found deeper (more dorsal) in the notum, and most lacked RFP (Fig 3E–3H). Category 3 predominated at 48 h after IR (Fig 3O). We interpret these data to mean that cells continued to translocate into the notum from the R73G07-GAL4 domain between 24 and 48 h after IR, with many terminating R73G07>RFP expression, indicative of fate change. As in category 2 discs, GFP+ cell population in the notum of category 3 discs appeared contiguous with both the hinge (arrowhead) and the pleura (arrow, Fig 3E, magnified in F). Moreover, GFP+ cells in the notum were more numerous in the posterior half (post) than in the anterior half (ant) in some discs (Fig 3G, magnified in H), which may explain the finding that ectopic discs seen at 72 h after IR always appear along the posterior wing margin (e.g. Fig 2M–2P; [14]). In category 4 discs, GFP+ cells in the notum were more numerous and more dorsal than in category 3, and most expressed RFP. In some category 4 discs, GFP+RFP+ cells were contained within the notum (similar to Fig 2I–2L) while in others GFP+RFP+ cells were in an ectopic disc (Fig 3I–3N, similar to Fig 2M–2P). Category 3 still predominated at the 72 h time point but the fraction of category 4 increased between 48 and 72 h after IR in both time courses (Fig 3O). Likewise, ectopic discs appeared between 48 and 72 h after IR, which agrees with our published results using a different GAL4 driver [14]. RFP+GFP+ cells of the ectopic discs appear contiguous with cells of the hinge and the pleura (arrowhead and arrow, respectively, in Fig 3I–3K).
How were GFP+RFP+ cells in the notum of category 4 discs produced? There are three possibilities. First, they are translocated hinge/pleura cells that never lost their original fate. Second, they are GFP+RFP- cells in class 3 discs that re-gained their hinge/pleura fate to re-express RFP. Third, they formed de novo and bear no relation to the hinge/pleura of the primary disc. The finding that GFP+RFP+ cells in the notum appear contiguous with the primary hinge and the pleura makes us favor the first two possibilities, which are not mutually exclusive. In our published time courses, we never saw cells of the 30A domain spreading into the notum [13, 14]. We interpret these data to mean that hinge cells that translocate into the notum originate from part the hinge outside the 30A domain (arrowheads in Fig 1J–1L and 1O).
Confocal imaging and close inspection of each optical section showed that ectopic discs included cells that lacked both RFP and GFP (Fig 3L–3N, arrows). In these experiments, the only cells that lacked RFP and GFP were notum cells, suggesting that cells of the notum also contribute to ectopic discs. We addressed this possibility directly by lineage-tracing with a notum-specific GAL4 driver (Fig 4).
R76A10-GAL4, bearing an enhancer fragment from the tailup locus, is active exclusively in a subset of notum cells (Fig 4A). Without IR, cell fate in this domain was stable as seen by GFP/RFP overlap (Fig 4A). At 72 h after IR, GFP/RFP overlap looked similar to–IR in most discs (Fig 4B). The exceptions were irradiated discs with ectopic growths, where we observed an expansion of the GFP+ cell population beyond the RFP+ area (Fig 4C, 4D and 4G). The lack of RFP in these cells suggests that they had lost their original fate as detected by R76A10-GAL4>RFP expression. Such GFP+RFP- cells were part of the ectopic disc, although the extent of their contribution to the ectopic disc and their location within the ectopic disc varied from disc to disc (arrows in Fig 4C, 4D and 4G). Nubbin is a pouch marker that is not normally expressed in the notum (Fig 4F). Some GFP+RFP- former notum cells in ectopic discs expressed Nub (Fig 4G–4G”‘), indicating that notum-to-pouch fate change occurred as these cells participated in the formation of an ectopic disc.
In our previous studies, cells of the pouch, marked with rn-GAL4>G-trace, did not change fate or translocate after irradiation [13]. Even in experiments when we directed cell death to the hinge and left the pouch cells alive, the hinge was repaired with the hinge cells and not the pouch cells [13]. We confirmed these findings in new experiments with rn-GAL4 as well as two additional pouch drivers, R42A07-GAL4 (from the dve locus) and R85E08-GAL4 (from the salm locus) (S3 Fig).
Collectively, the data in Figs 2–4 show that within a wing disc, cells in the hinge, the pleura and the notum can change fate and translocate after irradiation. Of these, hinge cells in the 30A-GAL4 domain translocate and change fate in nearly all irradiated discs to regenerate the pouch ([13]; see also Fig 5). In contrast, hinge cells outside the 30A domain, pleura and notum cells produce an ectopic disc in a fraction of irradiated discs (summarized in Fig 4E).
Drosophila wing disc is sub-divided into compartments, Anterior/Posterior and Dorsal/Ventral, for example, with cell lineages restricted to each compartment during development. During regeneration after ablation of a specific compartment, compartment boundaries collapse and are rebuilt, with some cells switching compartment identities [23]. In these models, one compartment suffered massive damage while the others were untouched. In contrast, damage by IR is scattered throughout the disc. Using the same compartment-specific GAL4 drivers as in the published study, ci-, ap-, and en-GAL4, we found that compartment boundaries remained intact in the primary disc during regeneration after IR damage (S4 Fig). In the same experiments, ectopic discs showed fluid compartment boundaries (further discussed in DISCUSSION).
Regenerative cells proliferate, change fate, and change location, in order to rebuild damaged tissue. Two aspects of regenerative behavior studied here, cell fate change and translocation, do not occur without IR (for example, Fig 1A and Fig 4A). Therefore, we next addressed how IR exposure is linked to these two aspects of regenerative behavior. IR has many effects on cells including DNA breaks, cell cycle arrest by checkpoints, and apoptosis. Of these, apoptosis has been demonstrated to induce proliferation of the surviving cells in a phenomenon known as Apoptosis-induced-Proliferation or AiP (reviewed in [24]). Therefore, we investigated whether apoptosis is also required for the induction of cell fate change and translocation after IR in the primary disc and for the formation of ectopic discs. In these experiments, we used 30A-GAL4>G-trace (expression pattern in Fig 1A–1C and 1J–1L, Fig 5A), which we used previously to show that former hinge cells lose GAL4 expression (become GFP+RFP-) and translocate towards the pouch ([13]; arrow in Fig 5B, which is from new experiments that reproduced the published findings). We confirmed that such cells gain the pouch fate as detected by Vestigial Quadrant Enhancer-lacZ or VgQ-Z (Fig 5C and 5D, arrows point to GFP+ former hinge cells that express VgQ-Z; [25]). We showed previously that regenerative behavior of the hinge cells could be quantified by measuring the GFP+RFP- area inside the 30A circle (Fig 5A, quantified in M; see figure legend and [13] for quantification method). We showed previously that ectopic disc formation could be quantified by classifying the discs according to 30A-GAL4>RFP pattern (Fig 5N; see figure legend and [14] for quantification method). As reported previously, unirradiated w1118 controls show classes 0 and I only; ectopic disc classes II-IV are IR-induced ([14]; Fig 5O).
Chromosome deficiency H99 deletes several genes including three pro-apoptotic genes, hid, rpr and skl; H99 heterozygous wing discs show delayed and reduced IR-induced caspase activation and apoptosis [26]. We expressed 30A-GAL4>G-trace in this background, and saw a reduction in fate change and translocation by the hinge cells (Fig 5, compare F to B, quantified in Fig 5M). We conclude that caspase activity and/or cell death is required for IR-induced regenerative behavior of the hinge cells. H99 heterozygosity also prevented ectopic disc formation in irradiated discs (Fig 5O), indicating that IR-induced apoptosis/caspase activity is also needed for the formation of ectopic discs.
H99 deficiency reduces apoptosis and caspase activity throughout the disc. Prior studies of regeneration in larval wing discs showed that apical caspase activity is needed in the dying cells to stimulate the neighbors to proliferate [24]. But no study we are aware of has addressed the need for caspase activity in the regenerative cells. Yet, there is mounting evidence for the role of caspases in cell fate changes [27, 28]. Our identification of regenerative cells in the hinge and the ability to target UAS-transgenes using 30A-GAL4 allowed us to address this possibility. We found that 30A-GAL4-driven expression of UAS-p35, an inhibitor of effector caspases, inhibited the translocation and fate change of the hinge cells (Fig 5, compare H to B, quantified in Fig 5M). 30A>p35 also inhibited the formation of ectopic discs (Fig 5O). We conclude that effector caspase activity is needed cell-autonomously (within the red cells in Fig 5N) for IR-induced regenerative behavior.
In reciprocal experiments, we expressed p35 in the pouch, to ask if effector caspase activity is needed in the pouch for the hinge to change fate and translocate. This is of interest because we and others have shown that dying cells secrete signals that change the behavior of surviving cells, including mitogenic signals [24] and ‘do not die’ signals [29]. We followed an experimental set-up used previously to show that when pouch cells were killed by rn-GAL4>UAS-hid, a pro-apoptotic gene, regeneration occurred with cells that immigrated from the hinge ([30]; the role of caspases were not addressed in this study). In these experiments, because GAL4>UAS was used in the pouch, the hinge cells were marked using the orthologous QF>QUAS system. We used the same GH146-QF driver, which is active in two pockets of hinge cells ([30]; see also Fig 6A and 6B), to mark the hinge, rn-GAL4 to express p35 in the pouch, and induced cell killing by IR (Fig 6). We found that GFP+ former hinge cells translocated into the pouch but only in irradiated discs (compare Fig 6B to 6F and 6J). p35 inhibits effector caspases but not initiator caspases in Drosophila [31, 32]; expression of p35 in the context of apoptosis induction results in ‘undead’ cells that initiate but cannot complete the apoptosis program and instead produces tissue overgrowth and disorganization [33]. We saw such overgrowth in the pouch of p35 expressing cells in the context of IR-induced apoptosis (arrowheads in Fig 6G and 6K). It was possible to discern the pouch using DNA staining at 48 h after IR but not at 72 h after IR because of overgrowth. Regardless, we observed GFP+ cells in the pouch (Fig 6H) or the pouch area (Fig 6K and 6L), and conclude that effector caspase activity is not needed in the pouch for the hinge cells to lose RFP and translocate.
Apical caspase Dronc is needed to activate effector caspases [34] and is required for X-ray induced apoptosis in the wing [35] and the eye [36] discs. We co-expressed a catalytically inactive C->A mutant (UAS-proDroncDN) [37], and found that both fate change and translocation of hinge cells and ectopic disc formation were inhibited to similar extent as did H99/+ and p35 (Fig 5I and 5J, quantified in M and O).
Dronc, together with stress-responsive JNK kinase, also functions within dying cells for mitogenic signaling [24]. This function of Dronc is considered separable from its function in activating effector caspases. In the rn-GAL4>hid model described above, inhibition of JNK in the pouch by overexpression of phosphatase Puc inhibited the immigration of hinge cells into the pouch [30]. We performed a reciprocal experiment to ask if JNK activity was needed in the hinge in our IR-based model. Co-expression of UAS-Puc with 30A-GAL4>G-trace had a statistically significant effect on the translocation of the hinge cells (Fig 5K and 5L, quantified in Fig 5M). But the effect was minor; normalized GFP+RFP- area decreased from 0.56±0.14 in ‘w1118+IR’ to 0.36±22 in ‘puc+IR’, p = 0.02). In contrast, puc had a stronger inhibitory effect on the formation of ectopic discs (Fig 5O). This is similar to our previous finding that depletion of nucleosome remodeling factor Nurf-38 with 30A-GAL4>RNAi had a greater effect on ectopic disc formation than on hinge translocation [14]. We had reported that while ectopic disc formation requires a temperature shift (irradiated larvae kept at 25°C throughout the experiment do not form ectopic discs; [14]), a temperature shift is not necessary for the hinge-to-pouch regeneration [13], further distinguishing the two modes of regeneration.
Zfh2 is a transcription factor and a downstream effector of JAK/STAT signaling during wing development [38]. In addition, Zfh2 has been shown to prevent apoptosis in tissues under stress [39]. During normal development, Zfh2 expression is confined to the hinge in the 3rd instar wing disc (Fig 1K). We published before that in irradiated discs, the hinge is protected from apoptosis ([13]; Fig 7A, brackets indicate the dorsal hinge). Therefore, we asked if Zfh2 has a role in preventing IR-induced apoptosis in the hinge. Zfh2 is required for wing hinge development [22, 38]. To deplete it we followed a protocol, described in Fig 7 legend, which is similar to what we used in published studies to deplete STAT and Wg that are also required for hinge development [14]. Zfh2 was conditionally depleted by en-GAL4-driven RNAi in the posterior half of the disc while the anterior half served as an internal control. We found that this treatment increased IR-induced apoptosis specifically in the posterior hinge (Fig 7B and 7C, compare arrows to arrowheads; brackets indicate the dorsal hinge). This increase was detected using Acridine Orange, which is excluded from live cells but penetrates dying cells [40, 41], and by antibody staining for cleaved caspase Dcp-1. Without Zfh2 RNAi, cell death was similar in anterior and posterior hinge (Fig 7A) and Zfh2 RNAi did not induce apoptosis without IR (Fig 7E). We conclude that Zfh2 is needed to protect the hinge cells from IR-induced apoptosis.
We next investigated the consequences of conditional Zfh2 depletion on the regenerative behavior of hinge cells, using the 30A-GAL4>G-trace (Fig 7F and 7G). Similarly treated w1118 controls show the translocation of hinge cells towards the pouch (Fig 7F, quantified in H). Depletion of Zfh2 inhibited the appearance of GFP+RFP- cells in the pouch area (Fig 7G, quantified in H). Collectively, these data suggest a cell-autonomous requirement for Zfh2 in the hinge to temper IR-induced apoptosis and to promote IR-induced fate change and translocation. As in the case of H99/+ and p35, 30A-GAL4>Zfh2 RNAi also prevented ectopic disc formation (Fig 7I).
Our published data offer an explanation for how the hinge is protected from IR-induced caspase activation and apoptosis [13]. After irradiation, hid mRNA increases throughout the disc. rpr is transcriptionally induced only in the notum and the pouch and not in the hinge where it is repressed by Wg. Thus, while the cells in the notum and the pouch experience increased hid and rpr and undergo apoptosis, only hid is induced in the hinge and appears insufficient for apoptosis ([13], summarized in Fig 7J). These published data allow us to interpret the results reported here (modeled in Fig 7K). During normal development (-IR), cells of the hinge lack effector caspase activity and do not die, change fate, or translocate. The same situation applies in irradiated discs expressing 30A-GAL4>p35 (+IR+p35). After irradiation (+IR), hid is induced in the hinge but not rpr, which we propose leads to caspase activity that is insufficient for apoptosis but sufficient for fate change and translocation; 30A-GAL4>p35 blocks this intermediate caspase activity to inhibit regenerative behavior in irradiated discs. Upon depletion of Zfh2 in the hinge (this report) or inhibition of STAT/Wg activity [13], IR induces sufficient caspase activity, which is compatible with apoptosis but not with fate change and translocation. Thus, Zfh2 serves to keep alive cells with radiation stress and a low level of effector caspase activity, thereby allowing them to adopt new fate and location. This is in agreement with the report that Zfh2 is needed to keep alive JNK-active regenerative cells in a genetic ablation model in the wing disc [39]. Similar regulatory mechanisms may apply in the irradiated notum but with a lot fewer cells displaying intermediate effector caspase activity, which can explain why ectopic discs form only in a small fraction of irradiated discs.
One prediction of the above model is that an irradiated wing disc contains cells that activated caspases but did not die and instead contributed to the regenerated disc. To test this prediction, we used ‘CaspaseTracker biosensor’, which is a membrane-tethered GAL4 that is ubiquitously expressed in all cells in Drosophila [42, 43]. In the presence of active effector caspases, the tether is cleaved at DQVD to release GAL4, which enters the nucleus to activate G-trace (Fig 8D). CaspaseTracker>RFP is not a real time reporter of caspase activity because it takes several hours to express [42], but CaspaseTracker>GFP lineage expression effectively marks cells that activated caspases but did not die. The sensitivity of CaspaseTracker is such that there is a substantial number of GFP+ cells without IR or without GAL80ts [43]. Therefore, we used GAL80ts and optimized the conditions to express CaspaseTracker within a narrow (6 h) window immediately following irradiation (Fig 8E). This protocol produced very few GFP+ cells in unirradiated discs (Fig 8A), and allowed us to detect changes in irradiated discs (Fig 8B). Caspase-resistant controls in which DQVD has been mutated to DQVA [42] do not show GFP+ cells even with irradiation (Fig 8C), indicating that we are capturing caspase activity. Clusters of CaspaseTracker>GFP+ cells were found throughout the irradiated disc (Fig 8B). We do not know how their origin is spatially distributed within the disc; addressing this issue will require a non-FLP/FRT-based lineage tracing system for use in conjunction with CaspaseTracker, which we plan to develop in the future. Most GFP+ clusters were composed of at least 8 cells each (Fig 8B’), supporting the idea that irradiated discs include cells that activated effector caspases but did not die and instead went through at least three cell cycles.
In tumor biology, the concept of cancer stem cells has been controversial, but there is agreement that within a tumor, some cancer cells are better than others at re-initiating tumor growth [15, 16]. Eradication of ‘Cancer Stem-like Cells’ (CSCs) is considered necessary for successful therapy. However, not only do CSCs generate non-stem cancer cells, non-stem cancer cells also are capable of converting to CSCs. Even more concerning, cancer treatments including IR converts non-stem cancer cells from a variety of cancer types into cells with CSC markers that can initiate new tumors in culture and in vivo [19–21]. An estimated 50% of cancer patients receive IR, alone or as part of their treatment (www.cancer.org). Yet, we know very little about what aspects of IR exposure induce CSCs. The finding that IR induces non-stem cells of the Drosophila larval wing discs to exhibit stem cell like properties allowed us to fill in the gaps in this knowledge.
This study identified multiple specific pools of cells that show IR-induced changes in cell fate and location. A subset of hinge cells can do so, to become part of the regenerated pouch in nearly all irradiated discs. A different subset of cells in the hinge, together with cells of the pleura and the notum, produce an ectopic disc in about a tenth of irradiated discs. There are, however, limits to fate changes. For example, we observed very little disruption of A/P and D/V boundaries in the primary hinge, which means hinge cells that changed fate and translocated remained in their original quadrant. The A/P boundary is already established and the ventral marker Ap is already active in early 2nd instar disc when the distinction between the hinge and the pouch is yet to be made [44]. Thus, the hinge-to-pouch conversion we detect reflects a developmentally more recent fate choice than A/P or D/V choices, identifying a limitation in IR-induced fate change. The fluidity of the A/P boundary in ectopic discs (S4 Fig), together with notum-to-pouch conversion in their generation suggest that fate changes during ectopic disc formation reflects events further back in development. But ectopic disc formation is rare and requires a temperature shift protocol, again illustrating that IR-induced fate change is not limitless. In another example, cells of the pouch display little indication that they change fate or translocate (S3 Fig) and not even when we directed cell death to the hinge and left the pouch cells alive [13]. This parallels how IR induces Cancer Stem Cell-like properties in some cancer cells but not others. Understanding how IR-induced fate changes are limited would be an important future goal.
There are several insightful reports of regeneration after genetic ablation where cell death is directed to a specific compartment of the wing disc (for example, [45–48]). Different models of regeneration including ours rely on common factors such as STAT and Wg, but also show different molecular requirements. For example, heterozygotes of transcriptional factor CtBP increased the incidence of ectopic discs in a genetic ablation model [48]. But the same alleles of CtBP, we found, did not increase IR-induced ectopic discs or hinge-to-pouch fate change and translocation (S5 Fig). Even within genetic ablation models, the choice of apoptotic gene used to kill cells can have different outcomes on the regenerative behavior of the surviving cells. For example, ectopic discs formed in the notum when the pouch was ablated with Eiger (Drosophila TNF) but not with Rpr [48].
A recent study used genetic ablation to probe the regenerative potential of the notum. After ablation of the pouch and the hinge, the notum showed no increase in proliferation, did not regenerate the hinge or the pouch, and instead duplicated itself. The authors concluded that the notum cells have little regenerative potential [46]. This is in sharp contrast to IR-induced regenerative behavior we see in the notum. After IR, we detect a 3-fold increase in mitotic activity in the notum [14], and lineage tracing shows the notum contributes to the ectopic discs (this report). Our results parallel more closely what happens after genetic ablation of the pouch, which resulted in the production of ectopic wing discs in some mutant backgrounds [48]. Lineage tracing suggested that notum cells changed fate to contribute to the ectopic disc in that model, which agrees with our results. We add to this picture by identifying additional pools of cells in the dorsal-posterior hinge/pleura that contribute to the ectopic discs (Fig 4E).
Collectively, these data illustrate how different regenerative models rely on different molecules and cellular behaviors, hence the need to study each to learn the range of possibilities. This, we believe, is particularly true in the case of IR where damage (i.e. cell death) is not confined to any particular compartment but scattered throughout the wing disc in a reproducible pattern [13].
One key gap concerns the question ‘what are the consequences of IR that induce stem cell-like behavior?’ The answer, we report here, is caspase activity. Surprisingly, we detect this requirement in regenerative cells that translocate and change fate. We propose that the outcome depends on the extent of effector caspase activity as modeled in Fig 7K. Regeneration typically relies on surviving cells proliferating and re-programming to replace cells lost to cell death or surgical removal. Prior work on Drosophila wing discs found that cell death itself induces Apoptosis-induced Proliferation in the surviving cells. In AiP, signaling through apical caspase Dronc and JNK in dying cells act through Dpp (TGF–β) and Wg to promote cell division in the surviving cells [33, 49–51]. The role of Dronc in AiP is in addition to its role in activating effector caspases and apoptosis. This and other similar mechanisms that operate in other larval discs explain the proliferative aspect of regeneration, but the re-programming aspect remained to be better understood. Our study fills the gap in the knowledge by identifying the role of caspases. We note a key difference between AiP and fate change/translocation. The former requires apical caspase Dronc but not effector caspases [24], while the latter requires both apical and effector caspases (this report).
The requirement for caspases we identified lies within the regenerative cell population, e.g. the hinge. This could be because a few apoptotic cells within this population stimulate others to display regenerative behavior. If such instigators exist, they must be very small in number because 30A-GAL4>p35 experiments showed no sign of overgrowth indicative of undead cells, even when they were readily visible in the pouch (Fig 6). Alternatively, caspases could play a non-apoptotic role within the regenerative cell population. There is precedent for non-apoptotic roles of effector caspases ([50, 52–54]; reviewed in [27]). In a particularly relevant study, effector caspase CED-3 in C. elegans was found to cleave cell fate determinant Lin-28, which is unrelated to apoptosis, in order to ensure that cell fate changes and developmental transitions occur normally [28]. The emerging view is that for a cell to change fate, it is insufficient to change only the transcriptional activity. Transcripts and proteins associated with the old fate must also be eliminated, and miRNAs and caspases work together to complete this task [28]. An intriguing hypothesis in Drosophila is that IR activates caspases that similarly down-regulate key fate determinants (to terminate the hinge identity, for example) and allow cell fate changes to reach completion. Validating this hypothesis will require the identification and functional analysis of caspase targets in this process.
These stocks are described in Flybase: w1118, 30A- GAL4 (on Ch II, Bloomington stock# or BL37534), Ptub-GAL80ts (on Ch III), rn-GAL4 (on Ch III), en-GAL4 (on Ch II), ci-GAL4 (on Ch II), ap-GAL4 (on Ch II), UAS-p35 (on Ch III). These stocks are described in publications: vgQ-lacZ [25], UAS-Zfh2 RNAi [22], UAS-DroncDN [37], UAS-puc [55], and CaspaseTracker and caspase resistant stocks [43]. The stock used for lineage tracing is also described in Flybase; w*; P{UAS-RedStinger}4, P{UAS-FLP.D}JD1, P{Ubi-p63E(FRT.STOP)Stinger}9F6 /CyO (BL28280). Genotypes for some BL stocks are in S1 Table and include FlyLight stocks [16]. Stocks to express QF and rn-GAL4 were generated by standard Drosophila recombinant protocols, using starting stocks listed in S1 Table.
Larvae were raised on Nutri-Fly Bloomington Formula food (Genesee Scientific). The cultures were monitored daily for signs of crowding, typically seen as ‘dimples’ in the food surface as larvae try to increase the surface area for access to air. Cultures were split at the first sign of crowding. Larvae in food were placed in petri dishes and irradiated in a Faxitron Cabinet X-ray System Model RX-650 (Lincolnshire, IL) at 115 kv and 5.33 rad/sec.
Antibodies to Zfh2 (1:400, rat polyclonal, [56]), Ci (1:500, rat monoclonal 2A1, deposited into Developmental Biology Hybridoma Bank by R. Holmgren [57]), Nubbin (1:50, mouse monoclonal 2D4, deposited into Developmental Biology Hybridoma Bank by Michalis Averof), cleaved caspase Dcp1 (1:100, rabbit polyclonal, Cell Signaling #9578S) and fluorescently labelled secondary antibodies (1:200, Jackson) were used (see also S1 Table). In all experiments, wing discs were dissected in PBS, fixed in 4% para-formaldehyde in PBS for 30 min, and washed three times PBTx (0.1% Triton X-100). For antibody staining, the discs were washed in PBS instead of PBTx after the fixing step, permeabilized in PBTx with 0.5% Triton X-100 for 10 min and rinsed in PBTx. The discs were blocked in 5% Normal Goal Serum in PBTx for at least 30 min and incubated overnight at 4°C in primary antibody in block. The discs were rinsed thrice in PBTx and incubated in secondary antibody in block for 2 h at room temperature. Stained discs were washed in PBT. The discs were counter-stained with 10 μg/ml Hoechst33342 in PBTx for 2 min, washed 3 times, and mounted on glass slides in Fluoromount G (SouthernBiotech).
With the exceptions noted below, the discs were imaged on a Perkin Elmers spinning disc confocal attached to a Nikon inverted microscope, using an SDC Andor iXon Ultra (DU-897) EM CCD camera. The NIS- Elements acquisition software’s large image stitching tool was used for the image capture. 15–20 z-sections 1 um apart were collected per disc. Sections that exclude the peripodial cells were collapsed using ‘maximum projection’ in Image J. The exceptions are images in Fig 7A–7E, Fig 8C and S1 Fig, which were acquired on a Leica DMR compound microscope using a Q-Imaging R6 CCD camera and Ocular software.
For sample size justifications, we used a simplified resource equation from [58]; E = Total number of animals − Total number of groups, where E value of 10–20 is considered adequate. When we compare two groups (w1118 vs H99/+, for example), 6 per group or E = 11 would be adequate. All samples subjected to statistical analysis exceed this criterion. Two tailed student t-tests were used to analyze the fate change and translocation of the hinge (Fig 5M and Fig 7H, S5 Fig) and Fisher Exact Test was used to analyze ectopic disc formation (Fig 5O, Fig 7I and S5 Fig). In the latter application, IR-induced classes (II-IV) were binned together to compare the number of class 0-I discs and class II-IV discs for one condition against the same for another condition.
|
10.1371/journal.ppat.1001081 | Structures of Receptor Complexes of a North American H7N2 Influenza Hemagglutinin with a Loop Deletion in the Receptor Binding Site | Human infections with subtype H7 avian influenza viruses have been reported as early as 1979. In 1996, a genetically stable 24-nucleotide deletion emerged in North American H7 influenza virus hemagglutinins, resulting in an eight amino acid deletion in the receptor-binding site. The continuous circulation of these viruses in live bird markets, as well as its documented ability to infect humans, raises the question of how these viruses achieve structural stability and functionality. Here we report a detailed molecular analysis of the receptor binding site of the North American lineage subtype H7N2 virus A/New York/107/2003 (NY107), including complexes with an avian receptor analog (3′-sialyl-N-acetyllactosamine, 3′SLN) and two human receptor analogs (6′-sialyl-N-acetyllactosamine, 6′SLN; sialyllacto-N-tetraose b, LSTb). Structural results suggest a novel mechanism by which residues Arg220 and Arg229 (H3 numbering) are used to compensate for the deletion of the 220-loop and form interactions with the receptor analogs. Glycan microarray results reveal that NY107 maintains an avian-type (α2-3) receptor binding profile, with only moderate binding to human-type (α2-6) receptor. Thus despite its dramatically altered receptor binding site, this HA maintains functionality and confirms a need for continued influenza virus surveillance of avian and other animal reservoirs to define their zoonotic potential.
| Influenza virus adaptation to different hosts usually results in a switch in receptor specificity of the viral surface coat protein, hemagglutinin. Indeed, the hemagglutinin subtypes from the last two human influenza pandemics of the 20th Century (H2 in 1957 and H3 1968) both adapted successfully to human-type receptor specificity through only two amino acid mutations in the receptor binding pocket (Glutamine226→Leucine and Glycine228→Serine). The recent human infections reported with other avian subtypes such as H5, H7 and H9 have raised public health concerns and focused efforts on identifying potential subtypes from which a future pandemic strain may emerge. Since 1996, H7 viruses of the North American lineage have been circulating in regional live bird markets, containing an eight amino acid deletion in the receptor-binding site of HA. Here we report a detailed structural analysis of the receptor binding site of a hemagglutinin from the North American lineage of H7N2 viruses, in complex with avian and human receptor analogs, to understand how these viruses have adapted to such a dramatic structural change in the binding site while remaining one of the predominant circulating viral strains.
| Influenza is an acute respiratory virus that infects up to 20% of the population in the United States, resulting in ∼36,000 deaths annually [1], [2]. The two membrane glycoproteins on the surface of influenza A virus, hemagglutinin (HA), which functions as the receptor binding and membrane fusion glycoprotein in cell entry, and neuraminidase (NA), which functions as the receptor destroying enzyme in virus release, form the basis for defining subtypes [3]. To date, 16 HA (H1–H16) and 9 NA (N1–N9) have been identified in avian species [4], while in the last century, only three subtypes, H1N1 in 1918 and 2009, H2N2 in 1957, and H3N2 in 1968 [5], [6], [7], have successfully adapted to humans. Hemagglutinin binds to sialic acid (SA) glycans present on host cell surfaces. The receptors on epithelial cells of the human upper respiratory tract are mainly α2-6-linked SA moieties [8]. Since avian influenza viruses predominately bind α2-3-linked SA, and human influenza viruses preferentially bind to α2-6-linked SA, human infection by avian influenza viruses is rare [9]. However, since 1997 a growing number of human cases of avian influenza infection have been reported [10], including H5N1, H7N2, H7N3, H7N7, and H9N2 strains [11]. Although the current situation with the pandemic H1N1 influenza virus dominates public health efforts, the prospect of a novel pandemic emerging from these isolated cases continues to be a major public health threat around the world.
Early cases of human infection by H7 influenza viruses are reported as far back as 1979 [12], [13]. Since 2002, multiple outbreaks and human infections of H7 subtype viruses; within both Eurasian and North American lineages have been reported. In the Netherlands in 2003, a highly pathogenic avian influenza (HPAI) H7N7 outbreak resulted in more than 80 cases of human infections, including one fatality [14], [15]. In New York in 2003, a single case of human respiratory infection of H7N2 was reported [16] and in British Columbia in 2004, an H7N3 virus caused two cases of conjunctivitis [17], [18]. More recently in 2007, the United Kingdom reported several cases of low pathogenic avian influenza (LPAI) H7N2 virus infections that caused influenza-like illness and conjunctivitis [19].
Since 1996, H7 viruses of the North American lineage have been circulating in regional live bird markets [20], containing a 24-nucleotide deletion resulting in an eight amino acid deletion in the receptor-binding site (RBS) of HA (Figure S1). The recent human infections with H7 in North America have raised public health concerns as to how these viruses adapt to such a dramatic structural change while remaining one of the predominant circulating viral strains. A recent study of H7 viruses isolated from previous outbreaks revealed efficient replication in both mouse and ferret animal models [21]. In particular, ferret studies with A/New York/107/2003 (NY107), an H7N2 virus isolated from a man in New York, not only showed efficient replication in the upper respiratory tract of the ferret but also the capacity for intra-species transmission by direct contact [21], [22]. Interestingly, both an increased preference for α2-6 and decreased preference for α2-3-linked sialosides of this virus compared to the other avian influenza viruses was shown by previous glycan microarray analysis but less so by a competitive solid-phase binding assay [22], [23].
Here we report a detailed molecular analysis of the RBS of the HA from North American lineage H7N2 virus, NY107, including glycan microarray analyses and structural analyses of the HA in complex with an avian receptor analog (3′-Sialyl-N-acetyllactosamine, 3′SLN) and two human receptor analogs (6′-Sialyl-N-acetyllactosamine, 6′SLN; Sialyllacto-N-tetraose b, LSTb). These results provide important insight into the interaction of H7 HAs with both avian and human hosts.
By using x-ray crystallography, the structure of H7 HA from the NY107 virus was determined to 2.6 Å resolution (Table 1). In addition, we also report three H7 HA receptor complex structures, with avian receptor analog (3′SLN) to 2.7 Å resolution and with human receptor analogs (6′SLN and LSTb) to 3.0 Å and 2.6 Å resolution, respectively (Table 1). The overall structure of NY107 is similar to other reported HA structures with a globular head containing the RBS and vestigial esterase domain, and a membrane proximal domain with its distinctive, central helical stalk and HA1/HA2 cleavage site (Figure 1A). Although five asparagine-linked glycosylation sites are predicted in the NY107 HA monomer, interpretable electron density was observed at only two sites, Asn38 in HA1 and Asn82 in HA2 (all residue numbers are based on H3 numbering). At these sites, only one or two N-acetyl glucosamines could be interpreted.
During viral replication, HA is synthesized as a single chain precursor (HA0) and cleaved by specific host proteases into the infectious HA1/HA2 form. In baculovirus expression systems, highly pathogenic HAs, with a polybasic cleavage site, are expressed as an HA1/HA2 form [24], whereas HAs with monobasic cleavage sites (single Arg) from low pathogenic viruses are expressed as the HA0 form [25]. NY107 is regarded as a low pathogenic virus, and as expected, was produced in the HA0 form (Figure S2). However, subsequent digestion with thrombin protease to remove the His-tag resulted in cleavage to a profile on SDS-PAGE comparable to that of an HA1/HA2 form (Figure S2). A comparison of the NY107 cleavage site with the consensus cleavage pattern in the MEROPS database (http://merops.sanger.ac.uk) suggests it to be a possible thrombin cleavage site.
Based on their molecular phylogenies, HAs are divided into two groups and five clades: group 1 includes H8, H9, and H12; H1, H2, H5, and H6; H11, H13 and H16; group 2 includes H3, H4, and H14; H7, H10 and H15 [26]. Among all available HA structures, we selected ten representative HAs from both avian and human subtypes for structural analysis. As expected, NY107 HA is structurally very similar to the Avian-H7 in all comparisons and closely related to H3, the other group 2 members used in the analyses (Tables S1 and S2).
The RBS is at the membrane distal end of each HA monomer and its specificity for sialic acid and the nature of its linkage to a vicinal galactose residue is a major determinant of host range-restriction. The consensus RBS for all current HAs is composed of three major structural elements: a 190-helix (residues 188–194), a 220-loop (residues 221–228), and a 130-loop (residues 134–138). In addition, highly conserved residues (Tyr98, Trp153, His183, and Tyr195) form the base of the pocket.
Although the NY107 RBS is similar to other subtypes (H1, H2, H3, H5, and H9), a previously observed specific feature of H7 HAs, is also observed in the NY107 150-loop region: two residues inserted at position 158 result in this loop protruding more than 6Å towards the binding site compared to other subtype HAs (Figure 1B and Table S2) [27]. More interestingly, the eight amino acid deletion, only found in the North American lineage H7s, from position 221 to 228 (Figure S1), resulted in a complete loss of the 220-loop (Figure 1B). Sequence alignment shows that Arg220 and Arg229 are conserved in all influenza A HA subtypes (Figure S1), but structural alignment of NY107 HA shows Arg220 occupying the Gly228 position, and the much shorter loop turns at residue Pro217 (Figure 1C). The Cα distance between NY107 Arg220 and its homolog in the Av-H7 structure (PDB: 1TI8) [27] is 5.8Å, and they point in opposite directions (Figure 1C). The side chain direction of Av-H7 Arg220 is almost parallel with the beta sheet after Arg229, whereas the NY107 Arg220 points downward to the binding pocket. The Cα position of Arg229 in both H7 structures remains the same, except the side chain in the NY107 swings away by about 5.9Å (Figure 1C) and could help to stabilize this region by forming a hydrogen bond to the mainchain carbonyl of Gln210 in the neighboring monomer. In the absence of the 220-loop in NY107 HA, upon glycan binding the long side chain of Arg220 compensates for its loss and is displaced 4Å upward to form hydrogen bonds with receptor analogs inside the binding pocket (Figure 1D).
Previously, mutations in the HA receptor binding domains of H1N1 (Glu190Asp/Gly225Asp) and H2N2/H3N2 (Gln226Leu and Gly228Ser) subtypes were responsible for adaptation of these viruses to pandemic strains [24], [28], [29], [30]. Due to missing residues 221–228 in the NY107 HA RBS, neither mechanism for adaptation is possible. Thus, in order to look more closely at the role of the missing loop and its effect on receptor specificity, we first subjected the recombinant HA (recHA) to glycan microarray analyses and compared it to a reverse genetics-derived NY107 virus, and a co-circulating Eurasian virus and recHA, A/Netherlands/219/2003 (NL219), that has the consensus avian sequence in the 220-loop and it also infected a human [15].
Glycan microarray analysis of recombinant NY107 (Figure 2A and Table 2) revealed a highly restricted binding profile with strong binding to only α2-3 sulfated (#4–8), α2-3 branched (#9–11) and mixed α2-3/α2-6 branched sialosides (#60–64) as well as to the long linear sialyl di- and tri-lactosamines (#22, 24). Weak binding was also observed (above background) to other α2-3 glycans on the array. The recombinant NY107 also revealed a strict glycan binding preference to only one α2-6 glycan, the internal structure, Galβ1-3(Neu5Acα2-6)GlcNAcβ1-3Galβ1-4Glc (#58; LSTb) (Figure 2A), a glycan highlighted in a previous study [22]. The virus with higher valency and avidity revealed stronger binding to all α2-3 groups, in addition to the branched di-sialyl α2-6 biantennary structures (#46–48) as well the LSTb (#58) (Figure 2B and Table 2). In contrast, the NL219 recHA (Figure 2C and Table 2) bound well to only the avian α2-3 containing sialyl-glycans (sulfated, branched, linear and fucosylated). Its corresponding virus also reflected this specificity although it also revealed strong binding to α2-3 N-glycolylneuraminic acid (Neu5Gc) containing glycans (#66–70) (Figure 2D and Table 2).
To further assess the effect of the missing 220 loop on HA structural stability and receptor specificity it was essential to evaluate these functions on the ancestral HA containing the full length 220-loop. To this end, we engineered an HA with an avian H7 consensus (PQVNGQSG) 220-loop re-introduced (NY107-220ins) into the NY107 HA and recovered this virus by reverse genetics. Compared to the NY107 virus (Figure 2A) glycan microarray analyses of the resulting NY107-220ins virus (Figure 3A and Table 2) revealed a decrease in binding to branched (#9–11) and linear (#12–27) α2-3 sialosides and a loss of binding to the branched di-sialyl α2-6 biantennary structures (#46–48), LSTb (#58) as well as the mixed α2-3/α2-6 branched sialosides (#60–64). In addition, sequence analysis of the NY107-220ins HA revealed the presence of quasispecies in the second position of the inserted loop, P(Q/K)VNGQSG, suggesting that re-introduction of the loop alone is not tolerated and does not create an avian-type binding profile. Thus other amino acid substitutions in the HA might have co-evolved with the deletion of the 220 loop to help stabilize the RBS/HA to maintain functionality.
When viruses containing this 220-loop deletion emerged in North America in the mid 90's, four additional amino acid substitutions, Gly114Arg, Asp119Gly, Gly186Glu and Gly205Arg, in the HA1 as well as an Asp19Asn in the HA2 chain were also introduced to most of the circulating isolates. Of these, Gly186Glu and Gly205Arg in the HA1 are close to the RBS, at the monomer interface, and could potentially modulate its structure and/or function. NY107 viruses with a restored consensus 220-loop and a single Glu186Gly (NY107-ins-186) or Arg205Gly (NY107-ins-205) substitution as well as the Glu186Gly/Arg205Gly double substitution (NY107-ins-186/205) were derived by reverse genetics and evaluated. Glycan microarray analysis for the three resulting viruses revealed similar glycan binding profiles with increased binding to α2-3 sialosides, including mixed α2-3/α2-6 branched sialosides (#60–64), α2-3 Neu5Gc (#66–70), but limited binding to the α2,6 sialosides (Figures 3B, 3C, 3D), resulting in a binding profile virtually identical to that of the NL219 virus and other avian influenza viruses (Figure 2D) [30]. Sequence analysis of the three reverse genetics derived viruses revealed no mutations/quasispecies in the HAs of either the NY107-ins-186 or the NY107-ins-186/205 virus stocks, indicative of replication fitness. For the NY107-ins-205 virus however, a Glu186Gly substitution emerged in the HA after only two passages in eggs following recovery from DNA transfection, indicating the importance of the co-variant position 186 with respect to HA functionality/glycan specificity. Altogether, the data indicates that the H7 subtype avian influenza viruses that were circulating in aquatic birds and poultry in North America before 1996 exhibited a classic avian α2-3 sialoside binding preference. In order for the 220-loop deletion to be tolerated, concurrent Gly186Glu and Gly205Arg substitutions in the vicinity of RBS of HA emerged to achieve a restricted α2-3 binding profile and only a moderate/limited increase in binding to branched di-sialyl α2-6 biantennary structures (#46–48) as well the α2,6 internal sialoside, LSTb (#58).
To understand from a structural perspective how NY107 interacts with host receptors, we solved the structure of NY107 in complex with an avian and two human receptor analogs. For the avian receptor analog, 3′SLN, the electron density maps revealed well-ordered features for the Sia-1, Gal-2, and GlcNAc-3 in the NY107 HA complex structure (Figure 4A). Structural comparison of NY107 HA binding to other, H1, H2, H3, H5, and H9 subtypes (Figure S2A) revealed that 3′SLN binding to NY107 resembled binding of the other published HAs. Indeed, the terminal Sia-1 moiety is positioned almost identically in all structures, and forms the majority of hydrogen bonds and contacts with residues in the RBS (Figure 4A and Table S3).
Published avian HA structures with an intact 220-loop form very close interactions with Gal-2 of 3′SLN via residue Gln226 which is important in receptor specificity and host adaptation. For example, in the avian H7/3′SLN HA structure it interacts with Gal-2 O4 [31]. In the NY107 HA structure, although Gln226 is absent and no other residue occupies the same space as Gln226 (Figure 1B), Arg220 does forms a hydrogen bond between Arg220 NH2 and Gal-2 O4 (Figure 4A). Interestingly, although there was interpretable density for the GlcNAc-3 (Figure 4A and Figure S4B), no hydrogen bonding was apparent between the HA and the GlcNAc-3, which is consistent with other reported structures [32]. Thus, for binding to avian receptors, the trans conformation of α2-3 linkages is essential and perhaps only the first two saccharides are required. Indeed, due to the absence of 220-loop in the NY107 HA structure, the “aperture” of the RBS formed by 220-loop and 130-loop in regular HAs is increased by ∼10 Å, so that the branched, internal, and perhaps more complicated glycans might be accommodated more efficiently.
In the NY107/6′SLN complex, only Sia-1 and Gal-2 are ordered (Figure 4B). The Sia-1 remains in the same position as previously analyzed glycan/HA complexes from H1, H2, H3, H5, and H9 (Figure S3B), whereas the Av-H7 complex structure with Sialyllacto-N-tetraose c (LSTc) did not reveal any density for the Sia-1 in the receptor binding site [31]. The Gal-2 position varies significantly among different subtypes. Compared to the human-adapted H1 HA [32], Gal-2 in the NY107 HA is 3Å higher, and thus is further from the protein (Figure S3B). In NY107, the Gal-2 only forms an intramolecular, saccharide-saccharide interaction with Sia-1. The poor electron density map and fewer interactions with protein residues suggest that the cis conformation of α2-6 linkages in 6′SLN trisaccharides show a reduced binding affinity with NY107.
Glycan array results with NY107 revealed a strong binding signal for the internal α2-6 sialoside, LSTb. To further investigate this interaction, we solved the structure of the NY107/LSTb complex. The final model contained Sia-1, NAG-2, Gal-3, and Gal-5 in the RBS. Although glycan microarray data indicated NY107 to have a specific affinity for LSTb, few interactions were apparent from the crystal structure. Sia-1 still forms multiple hydrogen bonds with residues in the RBS (Table S3 & Figure 4C). The branched Gal-5 interacts with Ser137, to help stabilize the LSTb binding. However, Arg220 and Lys193, the two residues showing close binding with 3′SLN, did not form any hydrogen bonds with LSTb. In the structure, Gal-5 also interacts with a crystal packing symmetry mate and thus the flexibility of whole LSTb may be restricted. In solution, with more freedom, the LSTb should be able to tilt closer to the RBS, and thus Glc-4 may have more interactions with the 190-helix than seen in the crystal structure.
Human infections by avian influenza viruses, including H7 subtypes, continue to pose a major public health threat. Although the species barrier prevents avian influenza viruses from widespread infection of the human population, the molecular determinants of efficient interspecies transmission and pathogenicity are still poorly understood. The viral coat protein HA however, is perhaps a critical molecule since previous pandemic viruses modified their receptor specificity and overcame the interspecies barrier to spread in the human population. Although HA structures alone and in complex with receptor analogs provide considerable insight into receptor binding, it is clear that HAs from different species and subtypes have significant structural variation. Indeed, low-pathogenic H7N2 avian influenza viruses with an 8 amino acid deletion within its RBS started to circulate in live-bird markets in the northeast United States in 1996. Despite what one would consider a debilitating mutation, these viruses have been reported as the predominant isolate [33]. Whether such a deletion contributed to their evolutionary success and how are an important questions, especially in light of NY107's ability to produce respiratory illness in humans [16], as well as its reported increased affinity for human-type receptors and ability for contact transmission in ferrets [21]. To try to help answer these questions, we have analyzed the molecular structures of NY107 and its complexes with receptor analogs to explain receptor specificity at the molecular level.
The crystal structures of NY107 and its complexes with both avian and human receptor analogs describe a mechanism as to how an influenza virus might adapt by dramatically altering its RBS, and still be functional. Arg220 of the HA1 chain of NY107 compensates for the loss of the 220-loop, by forming hydrogen bonds with Gal-2 from the avian analog (binding was not observed in either of the structures complexes with the human analogs). However, in the LSTb complex, branched Gal-5 forms extra interactions with the 130-loop, thus improving the binding preference for this particular glycan. Consistent with the structural evidence, glycan microarray analyses of NY107 revealed a strong binding preference for the branched α2-6 sialoside, LSTb. Except for the absence of the 220-loop, other key residues within the RBS are conserved in NY107 and thus, direct interactions with sialic acid are maintained.
The 220-loop is recognized as one of the three crucial structural elements in the RBS. Aside from the North American lineage H7N2 viruses, which have been circulating with a deletion (221–228) in this loop, there has been one other report describing a seven amino acid deletion (224–230) in a laboratory generated H3N2 escape mutant which was reported to have a slightly increased affinity for α2-3-linked glycans by hemagglutination assay [34]. Meanwhile, the equivalent region in the hemagglutinin-esterase-fusion (HEF) protein of influenza C virus reveals a rearrangement resulting in a truncated 260-loop in its RBS (Figure S5) [35]. However, without structural data with appropriate receptor analogs, it is not possible to compare the role of these loop variants in receptor binding to the H7 HA structure described here.
When compared to NL219, another co-circulating H7 avian virus HA (Figure 2C and D), overall binding to α2-3-linked glycans was markedly reduced, while increased binding to α2-6-linked receptors was only marginal. However, these results focus attention on only 2 sub-classes of human-type receptors that may be important for infection (and transmission in ferrets). The NY107 virus interaction with biantennary glycans (Figure 2B), although weak (not seen in Figure 2A with recHA), is a possible route for virus entry as biantennary structures are common on tissues, i.e. glycan profiling data from human lung tissue on the Consortium for Functional Glycomics (CFG) web site. In addition, the internal sialoside, LSTb, was observed in both virus and recHA microarray data, suggesting this type of glycan has good affinity for this HA. The significance of this is unknown since LSTb has only been described in human milk [36].
Interestingly, NY107 and NL219 virus receptor binding and specificity has been addressed previously using glycan microarray analysis that reported a significantly increased preference for α2-6 and decreased preference for α2-3-linked sialosides [22]. In addition, the same viruses were also included in a recent study from Gambaryan et al. using a competitive solid-phase binding assay [23]. Our findings confirm and extend the receptor binding specificity reported by these authors in that they reported both viruses binding to sulfated sialylglycans with a lactosamine (Galβ1-4GlcNAc core and reported only a moderate binding affinity for α2-6-sialyllactosamine, the human-type receptor analog used in their assay.
The 220-loop is an integral feature of the receptor binding site, and thus one would predict that such a deletion might have compromised this strain to be deleted from the population of circulating viruses. However, this was not the case [33] and its existence appears to be in part due to the additional mutations at positions 186 and 205. Restoration of the loop with either or both residues mutated back to the pre-1994 consensus sequence resulted in a classic avian influenza virus binding profile. The emergence of the Glu186Gly mutation in the HA of the NY107-ins-205 mutant after only two passages of the rescued virus in eggs, also indicates the importance of these positions for HA functionality/glycan specificity. Analysis of the structural data reveals that positions 186 and 205 are on opposite sides of a monomer but are both close to the 220-loop deletion region in the trimeric form. The Glu at position 186 is close to Arg220 and may interact with Arg220 when binding avian receptors. Position 205 in the neighboring monomer may be important in trimer stability and maintaining RBS functionality. If one models the pre-1996 220-loop restored into the NY107 structure, Arg205, Glu186 and the loop all clash, thus explaining the Glu186Gly mutation that emerged in the NY107-ins-205 virus HA after limited egg passage.
The NY107 RBS with its more restricted α2-3 glycan binding preference and weak/moderate increase in α2-6 binding may have given the virus a selective advantage to be maintained in poultry at live bird markets and supplying farms. Certain terrestrial birds, such as quails and chickens, have recently been shown to present both human and avian types of receptors in the trachea and intestine [37], [38], [39]. Although it is not known what specific glycans are presented in these animals, it is conceivable that a virus with mixed specificity might have a distinct advantage over avian viruses that have specific avian receptor requirements, particularly in bird markets where multiple species coalesce. Previous results with H7N2, H9N2 and H5N1 viruses all highlight the fact that an increase in α2-6-binding preference is not sufficient for efficient transmission of avian influenza viruses to humans [22], [40], [41]. Although it remains to be seen whether prolonged circulation of viruses in terrestrial birds, such as domestic chickens, can provide a possible route for viruses to adapt for efficient human infection [11], continued surveillance of influenza viruses from avian and other animal reservoirs is urgently needed to define their zoonotic potential.
Based on H3 numbering [42], cDNA corresponding to residues 11–329 (HA1) and 1–176 (HA2) of the ectodomain of the hemagglutinin (HA) from A/New York/107/2003 (H7N2; Genbank:ACC55270) and A/Netherlands/219/2003 (H7N7; Genebank: AAR02640) was cloned into the baculovirus transfer vector, pAcGP67-A (BD Biosciences), incorporating a C-terminal thrombin cleavage site, a “foldon” sequence [43] and a His-tag at the extreme C-terminus of the construct to enable protein purification [25], [44]. Transfection and virus amplification were carried out according to the baculovirus expression system manual (BD Biosciences Pharmingen).
Soluble NY107 was recovered from the cell supernatant by metal affinity chromatography using Ni-NTA resin (Qiagen Inc.). Fractions containing NY107 were pooled and dialyzed against 10 mM Tris-HCl, 50 mM NaCl, pH 8.0, then subjected to ion-exchange chromatography (IEX) using a Mono-Q HR 10/10 column (GE Healthcare). IEX purified NY107 was subjected to thrombin digest (3 units/mg protein; overnight at 4°C) and purified by gel filtration chromatography using a Superdex-200 16/60 column (GE Healthcare) and 50 mM Tris-HCl, 100 mM NaCl, pH 8.0 as running buffer. Protein eluting as a trimer was buffer exchanged into 10 mM Tris-HCl, 50 mM NaCl, pH 8.0 and concentrated to 14.5 mg/ml for crystallization trials. At this stage, the protein sample still contained the additional plasmid-encoded residues at both the N (ADPG) and C terminus (SGRLVPR).
Initial crystallization trials were set up using a Topaz Free Interface Diffusion (FID) Crystallizer system (Fluidigm Corporation, San Francisco, CA). Crystals were observed in several conditions containing PEG 3350 or PEG 4000. Following optimization, diffraction quality crystals for NY107 were obtained at room temperature using a modified method for microbath under oil [45], by mixing the protein with reservoir solution containing 20% PEG 3350, 0.2 M magnesium chloride at pH 7.2. For receptor analog complexes, crystals were soaked for 3 hours in the crystallization buffer containing 10 mM 3′SLN or 6′SLN (V-labs Inc., Covington, LA), or overnight in 10mM LSTb (Sigma, St. Louis, MO). All crystals were flash-cooled at 100K using 20% glycerol as the cryo-protectant. Datasets were collected at Advanced Photon Source (APS) beamlines 22 ID and BM at 100K. Data were processed with the DENZO-SACLEPACK suite [46]. Statistics for data collection are presented in Table 1.
The structure of NY107 was determined by molecular replacement with Phaser [47] using the structure of the avian H7 (Av-H7) from A/turkey/Italy/2002, pdb:1TI8 (HA1, 78% identity; HA2, 90% identity) as the searching model. One HA trimer occupies the asymmetric unit with an estimated solvent content of 58% based on a Matthews' coefficient (Vm) of 2.9 Å3/Da. Rigid body refinement of the trimer led to an overall R/Rfree of 28.6%/37.4%. The model was then “mutated” to the correct sequence and rebuilt by Coot [48], then the protein structures were refined with REFMAC [49] using TLS refinement [50]. The final models were assessed using MolProbity [51]. The three complex structures were refined and evaluated using the same strategy. All statistics for data processing and refinement are presented in Table 1. Electron density maps (2fo-fc) were generated in Refmac [49] while simulated annealing omit maps were generated by sa-omit-map, a part of the Crystallography and NMR System (CNS) software [52].
Wild type and mutant viruses of NY107 (H7N2) and A/Netherland/219/2003 (H7N7) were generated from plasmids by a reverse genetics approach [53]. To generate viruses with amino acid insertion or substitution in the HA, mutations were introduced into plasmid DNA with an overlap extension PCR approach [54]. Viruses derived by plasmid transfection of HK293 cells were propagated in eggs. The genomes of resulting virus stocks were sequenced to detect the emergence of possible variants during amplification.
Glycan microarray printing and recHA analyses have been described previously [24], [30], [44], [55] (see Table 2 for glycans used for analyses in these experiments). Virus were analyzed on the microarray as described previously [30].
The atomic coordinates and structure factors of NY107 are available from the RCSB PDB under accession codes 3M5G for the unliganded NY107, 3M5H for the NY107 with 3′-SLN and 3M5I and 3M5J for NY107 with 6′SLN and LSTb, respectively.
A/New York/107/03 (H7N2), Genbank: ACC55270; A/Netherlands/219/03 (H7N7), Genbank: AAR02640; A/Hong Kong/1-9/68 (H3N2), 2HMG; A/Duck/Ukraine/1/63 (H3N8), PDB: 1MQL; A/South Carolina/1/18 (H1N1), PDB: 1RD8; A/Puerto Rico/8/34 (H1N1), PDB: 1RU7; A/Swine/Iowa/15/30 (H1N1), PDB: 1RUY; A/Singapore/1/1957 (H2N2), PDB: 2WRC; A/Viet Nam/1203/04 (H5N1), PDB: 2FK0; A/Duck/Singapore/3/97 (H5N3), PDB: 1JSM; A/Swine/Hong Kong/9/98 (H9N2), PDB: 1JSD; A/Turkey/Italy/8000/02 (H7N3), PDB: 1TI8; C/Johannesburg/1/66, 1FLC.
|
10.1371/journal.ppat.1003121 | A Pathogen Type III Effector with a Novel E3 Ubiquitin Ligase Architecture | Type III effectors are virulence factors of Gram-negative bacterial pathogens delivered directly into host cells by the type III secretion nanomachine where they manipulate host cell processes such as the innate immunity and gene expression. Here, we show that the novel type III effector XopL from the model plant pathogen Xanthomonas campestris pv. vesicatoria exhibits E3 ubiquitin ligase activity in vitro and in planta, induces plant cell death and subverts plant immunity. E3 ligase activity is associated with the C-terminal region of XopL, which specifically interacts with plant E2 ubiquitin conjugating enzymes and mediates formation of predominantly K11-linked polyubiquitin chains. The crystal structure of the XopL C-terminal domain revealed a single domain with a novel fold, termed XL-box, not present in any previously characterized E3 ligase. Mutation of amino acids in the central cavity of the XL-box disrupts E3 ligase activity and prevents XopL-induced plant cell death. The lack of cysteine residues in the XL-box suggests the absence of thioester-linked ubiquitin-E3 ligase intermediates and a non-catalytic mechanism for XopL-mediated ubiquitination. The crystal structure of the N-terminal region of XopL confirmed the presence of a leucine-rich repeat (LRR) domain, which may serve as a protein-protein interaction module for ubiquitination target recognition. While the E3 ligase activity is required to provoke plant cell death, suppression of PAMP responses solely depends on the N-terminal LRR domain. Taken together, the unique structural fold of the E3 ubiquitin ligase domain within the Xanthomonas XopL is unprecedented and highlights the variation in bacterial pathogen effectors mimicking this eukaryote-specific activity.
| Numerous bacterial pathogens infecting plants, animals and humans use a common strategy of host colonization, which involves injection of specific proteins termed effectors into the host cell. Identification of effector proteins and elucidation of their individual functions is essential for our understanding of the pathogenesis process. Here, we identify a novel effector, XopL, from Xanthomonas campestris pv. vesicatoria, which causes disease in tomato and pepper plants. We show that XopL suppresses PAMP-related defense gene expression and further characterize XopL as an E3 ubiquitin ligase. This eukaryote-specific function involves attachment of ubiquitin molecule(s) to a particular protein targeted for degradation or localisation to specific cell compartments. Ubiquitination processes play a central role in cell-cycle regulation, DNA repair, cell growth and immune responses. In the case of XopL this activity triggers plant cell death. Through structural and functional analysis we demonstrate that XopL contains two distinct domains, one of which demonstrates a novel fold never previously observed in E3 ubiquitin ligases. This novel domain specifically interacts with plant ubiquitination system components. Our findings provide the first insights into the function of a previously unknown XopL effector and identify a new member of the growing family of bacterial pathogenic factors hijacking the host ubiquitination system.
| Most Gram-negative pathogenic bacteria implement the type III secretion system (T3SS) that injects a set of proteins, termed effectors (T3E), directly into the eukaryotic host cell. The effectors' combined function is to subvert the host immune system and to promote bacterial colonization [1], [2]. Plant immunity relies on recognition of conserved pathogen-associated molecular patterns (PAMPs) [3], such as flagellin or bacterial elongation factor Tu [4], [5]. This defense barrier is termed PAMP-triggered immunity (PTI), is activated upon PAMP recognition at the cell surface by specific receptors, followed by a network of cellular signaling events, such as mitogen-activated protein kinase (MAPK) cascades, that ultimately lead to changes in gene expression [3], [6], [7]. In contrast, type III effectors manipulate plant cell processes, often leading to subversion of plant immune responses [1], [8]. T3Es interfere with key eukaryotic cell functions, such as the cytoskeleton rearrangement [9], transcriptional regulation [10], [11] or ubiquitination [12], [13]. However, the biochemical function of the majority of T3Es remains elusive.
Ubiquitination is a highly conserved eukaryote-specific post-translational protein modification involving attachment of ubiquitin to the epsilon amine of a lysine residue in the target protein. This modification alters protein activity, protein localization or targets the protein for 26S-proteasome-mediated degradation [14]. Ubiquitination of target proteins involves coupling of ubiquitin to an ubiquitin activating enzyme (E1), transfer to a conjugating enzyme (E2), before an ubiquitin ligase (E3) mediates ubiquitin transfer from an E2 to a target protein [15]. E3 enzymes exhibit high target specificity and differ in the subset of E2s they interact with. Eukaryotic E3s fall into two major classes according to the mechanism of ubiquitin transfer: RING/U-box and HECT domain proteins [16]. RING finger/U-box proteins transfer ubiquitin directly from the E2 to the target protein, whereas HECT proteins first form a thioester intermediate with ubiquitin before ligating it to the target. While ubiquitination is absent in prokaryotes, it emerges as a prime eukaryotic host target for bacterial pathogens, which have evolved diverse T3Es to mimic ubiquitination-related functions. In particular, several bacterial T3Es from animal and plant pathogens function as E3 ubiquitin ligases, represented on one hand by the Pseudomonas syringae T3E AvrPtoB [12], [13] and the NleG family of E. coli T3Es [17], which contain typical U-box folds, and on the other hand by the NEL (novel E3 ligase) domains found in the IpaH and SspH2 T3Es of Shigella and Salmonella spp., respectively [18], [19]. The latter contain a novel thioester-forming E3 ligase domain with no structural homology to the HECT domain. This suggests that during co-evolution with their hosts, pathogenic bacteria have employed different solutions to fulfill the otherwise typical eukaryote-specific function of E3 ubiquitin ligases.
Here, we characterized the T3E XopL (Xanthomonas outer protein L) from the model plant pathogenic bacterium Xanthomonas campestris pv. vesicatoria (Xcv), which causes disease on tomato and pepper plants. Xcv injects a suite of ∼30 T3Es into the host cell including the TAL (transcription activator-like) effector AvrBs3, which manipulates plant transcription [10], and the SUMO (small ubiquitin modifier) protease XopD [20]. XopL is a newly identified T3E from Xcv, and was found to exhibit E3 ubiquitin ligase activity. Crystal structure determination revealed that the protein contains a novel fold and thus represents a new class of E3 ubiquitin ligases.
The analysis of the genome sequence of Xcv strain 85-10 led to the identification of XCV3220 (xopL) as a new T3E candidate gene. XCV3220 is conserved in Xanthomonas spp. (Figure S1) and contains a PIP box (plant inducible promoter) in its promoter (TTCG-N16-TTCG; genome position 3669238-261). The presence of a PIP box in the xopL promoter suggested a co-regulation with the T3S system, which was confirmed by RT-PCR (Figure S2A). The predicted gene product contains leucine-rich repeats (LRRs), which are typically found in eukaryotic proteins and are thus indicative of an effector protein activity. Type III-dependent secretion and translocation of XCV3220 was confirmed by in vitro secretion and in vivo translocation assays (Figure S2B, C). The protein was therefore renamed XopL (for detailed information see Text S1).
To investigate a possible virulence function of XopL, we deleted the gene from the genome and analyzed the corresponding deletion mutants by infection studies in pepper plants. However, under the conditions tested XopL had no discernible effect on virulence (Figure S2D) or bacterial growth of Xcv (data not shown). To further characterize XopL we expressed xopL in different plant species via Agrobacterium-mediated transformation. Expression of XopL induced plant cell death (PCD) in leaves of Nicotiana benthamiana (Figure 1A), but no macroscopic reaction in pepper or tomato plants (data not shown). PCD was confirmed by quantifying ion leakage, which is used to measure dying plant cells (Figure 1B).
To identify the role of XopL during the infection of plants, we tested if it manipulates plant immunity, as shown previously for several T3Es from Pseudomonas and Xanthomonas, which specifically suppress the PAMP-triggered immunity (PTI) [21]–[26]. To analyze this, we performed Arabidopsis leaf protoplast assays, a well-established system for PAMP-signaling analysis [25], [27], [28]. We tested the activity of the A. thaliana NHL10 (NDR1/HIN1-LIKE 10) [29], [30] promoter fused to the firefly luciferase gene (LUC) after application of elicitor-active epitopes of different bacterial PAMPs. The reporter assays showed that the basal activity of pNHL10 was not affected by XopL (Figure 2A). However, the expression of xopL significantly decreased the activation of pNHL10 by flg22 (a bacterial flagellin epitope) [4] as well as that of elf18 (an 18 amino acid peptide derived from the EF-Tu protein) [5] (Figure 2B, C). Induction of pNHL10 by flg22 depends, at least partially, on activity of mitogen- activated protein kinases (MAPKs) [27]. Therefore, the activation of the MAPKs MPK3, MPK4, MPK6 and MPK11, which are involved in plant immune signaling [31], [32], might be affected by XopL. However, immunoblot analysis using an antibody against activated MAPKs revealed no differences in MAPK activity in protoplasts expressing XopL (or its derivatives; data not shown) compared to CFP (cyan fluorescent protein, negative control) (Figure 2D). AvrPto served as a positive control in both assays as it suppresses PTI by intercepting MAPK signaling pathways [33]. Proteins were stably expressed and protoplasts were still viable during the course of the experiment, confirming that the lack of pNHL10 expression was not due to ongoing cell death of the protoplasts (Figure S3A, B).
The N-terminal LRRs of XopL are reminiscent of the domain architecture of the T3E families IpaH and SspH2 from Shigella and Salmonella, respectively, that were recently identified as E3 ubiquitin ligases [18], [19]. We, therefore, tested XopL for E3 ubiquitin ligase activity in vitro. For this, we purified recombinant full-length XopL[aa 1–660] and truncated XopL derivatives XopL[aa 144–660] (lacking the disordered pre-LRR region), XopL[aa 474–660] (lacking the LRRs) and XopL[aa 86–450] (lacking the C-terminal region).
XopL and its derivatives were tested in ubiquitination assays using human E1 and the ubiquitous human E2 (UBE2D2) or the related plant E2s (AtUBC11 or AtUBC28, both with ∼80% sequence identity to UBE2D2) enzymes. In the case of full-length XopL, XopL[aa 144–660] and XopL[aa 474–660], western blot analysis with ubiquitin antibodies revealed a robust time-dependent accumulation of high-molecular-weight polyubiquitinated protein species (Figure 3A, B), which at later time points correlated with consumption of free ubiquitin (Figure 4B). A similar result was also obtained for the more distantly related XopL from X. c. pv. campestris (Table S1 in Text S1).
Western blot analysis using α-His antibodies (Figure 3A) and Coomassie Blue staining of SDS-PAGE gels (Figure 3B), combined with mass spectrometric analysis of the high-molecular weight species (data not shown) demonstrated minimal modification of the XopL fragments, indicating that the principle product of in vitro ubiquitination reactions were unattached ubiquitin chains. In the case of the XopL[aa 86–450] fragment, no polyubiquitinated protein species were detected (Figure 3A), suggesting that polyubiquitination was dependent on the intact XopL C-terminal region. XopL-mediated formation of ubiquitin chains required both E1 and E2 enzymes (Figure 3B), demonstrating that XopL acts similarly to eukaryotic E3 ubiquitin ligases.
Next, we determined the type of ubiquitin linkages preferentially generated by XopL. Ubiquitin contains seven lysine (K) residues (K6, K11, K27, K29, K33, K48 and K63) that can participate in ubiquitin ligation [14]. Therefore, we analyzed the products of the XopL-mediated polyubiquitination reaction using plant AtUBC11, AtUBC28 and human UBE2D2 conjugating enzymes. While the relative amount of distinct ubiquitination linkages detected by this analysis (Table S1 in Text S1) was different depending on which E2 enzyme was used in the reaction, the K11 linkages represented the largest fraction in all cases. More than half of the linkages analyzed in reactions with AtUBC28 and UBE2D2 enzymes were K11, whereas K11 represented ∼45% of the linkages in reactions with AtUBC11. The remaining polyubuitination linkages corresponded primarily to K33, K48 and K63 (Table S1 in Text S1). Interestingly, K63-linked polyubiquitin chains were detected in reactions using AtUBC28 and human UBE2D2 but not in reactions with plant AtUBC11, suggesting that these homologous E2 enzymes may contribute to a different preference in linkages that are formed during E3 catalyzed reactions.
In order to confirm the prevalence of the detected linkages in the XopL-mediated reaction we then performed polyubiquitination assays using ubiquitin variants with each individual lysine residue mutated to arginine (Figure 3C). In accordance with mass spectrometry results, the K11R mutation significantly dampened the XopL-mediated formation of polyubiquitin chains in the reaction using the AtUBC11 enzyme. A similar effect was detected in case of K33R and K48R mutations. Interestingly, the K6R mutation also resulted in significant reduction of polyubiquitination, while no K6 linkages were detected among XopL polyubiquitination products. This result suggested that this mutation might have a general deleterious effect on ubiquitination, potentially due to reduced affinity to E1 or E2 enzymes.
Next, we tested XopL ubiquitin ligase activity with different plant-derived E2s. As stated above, XopL forms ubiquitin chains with AtUBC11 and AtUBC28 (93% sequence identity), which belong to group VI of the 16 E2 classes of this plant [34], and the close human homologue UBE2D2. However, two more distantly related E2s (Table S1 in Text S1), namely AtUBC13 (group V, 34% sequence identity to AtUBC11) and AtUBC19 (group VIII, 43% sequence identity to AtUBC11) did not show any activity in our in vitro assays (Figure 4A), suggesting that XopL discriminates between different classes of E2 enzymes, as was described for other E3 ubiquitin ligases [19], [35]–[37].
Interactions between the human UBE2D2 enzyme and E3 ubiquitin ligases have been studied in detail by mutagenesis [38]. Because mutation of conserved residues in UBE2D2 abrogated ubiquitination in vitro, we purified the R5A, F62A, K63A and A96D variants of the AtUBC28 E2 enzyme and tested them individually in XopL and XopL[aa 474–660] ubiquitination assays.
The F62A and A96D mutations in AtUBC28 completely abrogated both the XopL[aa 474–660]- and XopL-mediated polyubiquitination reactions (Figure 4B; data not shown), suggesting that F62 and A96 are required for the AtUBC28 interaction with XopL. By contrast, the AtUBC28 R5A and K63A mutants were still very active in vitro (Figure 4B).
Taken together, our results demonstrate that XopL is an E3 ubiquitin ligase that selectively recruits plant E2 enzymes.
The XopL C-terminal domain harboring E3 ubiquitin ligase activity lacks significant sequence similarity with previously characterized E3 ligases. To gain further insight into the structural basis of XopL activity, we determined the structure of XopL by X-ray crystallography. While full-length XopL did not crystallize, fragments XopL[aa 144–450] and XopL[aa 474–660] yielded crystals that diffracted to a resolution of 2 Å and 1.8 Å, respectively. In both cases, single-wavelength anomalous dispersion (SAD) data were collected at the selenium peak wavelength from a single selenomethionine-enriched crystal. The final model of XopL[aa 144–450] contained a single molecule in the asymmetric unit corresponding to residues 145 to 437 plus four additional residues from the N-terminal polyhistidine tag. For the XopL[aa 474–660] fragment, three polypeptide chains were found in the asymmetric unit corresponding to residues 474–642 plus up to six residues from the N-terminal polyhistidine tag. Data collection and refinement statistics for both structures are presented in Table 1.
The structure of the XopL[aa 144–450] fragment follows a canonical LRR architecture with ten β-strands and nine complete repeats each folding into an α-helix (single turn)-turn-β-strand motif (Figure 5A). Three α-helices (α1, α2 and α3) and one α-helix (α 4) cap the LRRs at the N- and C-terminus, respectively. This structure is similar to the LRR domain of IpaH3 (PDB 3CVR [39], Figure 5A). Based on the sequence conservation at specific positions in individual repeats, a consensus sequence for the XopL LRRs can be derived that is similar to that of plant derived LRR-containing proteins (Figure 5B).
The structure of the XopL C-terminal region [aa 474–660] represents a four-helix bundle, which can be subdivided into two uneven lobes almost perpendicular to each other (Figure 6A). The smaller lobe contains the N-terminus, α2b and α3 helices and a region C-terminal to the α2b helix (residues 554–562), which adopts a conformation intermediate between a poly-proline type II helix and a β-strand. The two lobes give the XopL[aa 474–660] molecule an “L”-shape, and a large cleft with a net negative charge is formed at the intersection of the two lobes (Figure 6B, C). A search for structural homology using the DALI server (http://ekhidna.biocenter.helsinki.fi/dali_server/, 2012) did not reveal any significant similarity between the XopL[aa 474–660] structure and other structurally characterized proteins including E3 ubiquitin ligases. This analysis clearly demonstrates that the XopL C-terminal domain represents a novel fold, which we termed XL-box (XopL E3 ligase box). The XL-box lacks cysteine residues. Therefore, XopL E3 ubiquitin ligase activity appears not to involve the formation of thioester intermediates with ubiquitin as was shown in the case of eukaryotic (HECT-type) and effector (IpaH and SopA) catalytic E3 ubiquitin ligases.
Given that structural analysis defined the presence of two distinct domains in XopL (LRR and XL-box), we tested their individual role in suppressing PAMP-induced gene expression and inducing PCD (see above; Figure 1A). When the N-terminal [aa 1–449] and the C-terminal [aa 450–660] XopL regions were expressed individually or co-expressed in N. benthamiana, no PCD was observed (Figure 1A, B) demonstrating that an intact XopL protein is required to provoke PCD, which is consistent with the suggested function of the LRRs in recognition of a plant target protein ubiquitinated by the XL-box. Next, we tested the effect of mutations in the XL-box domain on the ability of XopL to provoke PCD (Figure 1A, B; Figure S5A, B; Table S3 in Text S1). Residues D502, K578, A579, Q612 and L619 co-localize on the surface of the major cleft of the XL-box (Figure 6C), and are highly conserved (Figure S1). Each of the aforementioned residues was substituted by alanine, except for A579, which was mutated to tryptophan. Transient expression of these XopL variants in N. benthamiana revealed that the XopL mutant derivatives were stably synthesized (Figure 1C) and D502A, K578A, A579W or Q612A exchanges abolished the ability of XopL to induce PCD. By contrast, the XopLL619A variant was still active (Figure 1A, B). We then investigated if E3 ubiquitin ligase activity of XopL can be demonstrated in the plant. N. benthamiana leaves expressing full-length XopL, XopL[aa 1–450], XopL[aa 450–660] or GFP (green fluorescent protein; control) were analyzed by western blotting using ubiquitin-specific antibodies. Expression of full-length XopL and XopL[aa 450–660], but not XopL[aa 1–450] led to the presence of additional high molecular mass ubiquitinated protein species, that were not detected upon expression of gfp (Figure 1C).
Notably, the D502A, K578A, A579W and Q612A mutations that abrogated the ability of XopL to cause PCD also dampened the formation of polyubiquitin chains in vivo (Figure 1C, Figure S5C). On the other hand, XopLL619A caused PCD and retained the ability to mediate formation of polyubiquitin chains in vivo similarly to the wild type. A similar result was found performing in vitro polyubiquitination reactions using the AtUBC11 conjugating enzyme and XopL[aa 474–660] (Figure S6). Taken together these results suggested that PCD is caused by XopL E3 ligase activity, manifested by formation of polyubiquitin products in vivo and in vitro.
We also tested the effect of the individual domains on suppression of PAMP-induced gene expression relative to full-length XopL (Figure 2, Figure S3). Unexpectedly, the PAMP-suppression activities of XopL are mediated by the N-terminal (residues 1–450) fragment corresponding to the LRR-containing region, which suppressed PAMP-induced gene expression to a similar extent as the full-length XopL. In addition, full-length XopL with a Q612A mutation in the XL-box, which both strongly hinders the ability of XopL to promote PCD and to polyubiquitinate in vivo and in vitro, retained the ability to inhibit the expression of the reporter gene in the presence of either PAMP elicitor peptides. Finally, the expression of the XopLCTD did not suppress, but rather elevated, the expression of the reporter even in the absence of the PAMP elicitor peptides (Figure 2A–C).
In this study, we identified XopL as a new T3E in Xcv that induces cell death in N. benthamiana and inhibits PTI-related defense gene expression. According to our data, XopL exhibits a robust E3 ubiquitin ligase activity. This activity is associated with its C-terminal region and is required for induction of plant cell death. All ubiquitin ligases known to date including bacterial T3Es with E3 ligase activity belong to the RING/U-box or catalytic (HECT-like) class [16]. RING/U-box proteins act by transferring ubiquitin from E2 directly onto the target protein. T3Es of this class include AvrPtoB from the plant pathogen P. syringae [13], and E. coli NleG [17]. Both T3Es lack significant sequence similarity with RING/U-box proteins but adapt a protein fold similar to that of U-box proteins. On the other hand, the catalytic HECT E3 ligases first attach ubiquitin from the E2 to a catalytic cysteine residue via a thioester intermediate before ligating it to the target protein. A similar mechanism has been adopted by effector proteins of the IpaH and SopA families of animal pathogens [40], [41]. The IpaH and SopA crystal structures are distinct from HECT proteins except for the presence of a catalytic cysteine and certain features of the active site. As XopL lacks cysteine residues in its C-terminal domain, termed XL-box, we hypothesize that it acts by directly transferring ubiquitin from E2 onto a target protein. This is reminiscent of RING/U-box proteins; however, XopL lacks any structural similarity to these E3 ligases.
We found that XopL interacts in vitro through its XL-box with a specific family of E2 enzymes, represented by human UBE2D2 and Arabidopsis AtUBC11 and AtUBC28. In Arabidopsis thaliana, AtUBC11 and AtUBC28 are members of the group VI family of E2 enzymes [34]. Many of the 8 family members are ubiquitously expressed in Arabidopsis (including AtUBC28 and AtUBC11) and the three most highly expressed members of this family (AtUBC8, AtUBC10 and AtUBC28; www.genevestigator.com) share 97% sequence similarity with each other. Homologues to these proteins are also found in tomato (S. lycopersicum gi|350536447|; 97% identical to AtUBC28) and pepper (C. annuum gi|40287554|; 96% identical to AtUBC28). Mutation analyses of AtUBC28 revealed amino acid residues F62 and A96 to be critical for the interaction with the XopL E3 ligase. It is worth noting that residue F62 is essential for E2 interactions with HECT E3 ligases [42], but not for interactions with specific RING/U-box proteins [43]. On the other hand, residue A96 in E2 enzymes was shown to contribute to interactions with both HECT- and RING-type ligases plus the bacterial effector SspH2 [44]. While this data reveals some molecular details of the XopL interaction with E2 enzymes it cannot be modeled according to previously characterized E3-E2 pairs and requires further structural analysis.
XopL-mediated polyubiquitin chains with preponderance of K11 linkages were detected using both Arabidopsis group VI E2 enzymes and the human UBE2D2 enzyme. Ubiquitin contains seven lysine residues that can participate in target protein ubiquitination. Which specific lysine is used is dictated by different E3-E2 enzyme combinations and may trigger different outcomes for a given target protein. Linkage at K48 usually directs target proteins to the proteasome [45], whereas K63-ubiquitination can play a role in signal transduction [46]. The importance of other ubiquitin linkages for cell processes came to light only recently and their physiological role remain largely unknown [47]. A recent report suggested that mixed K11- and K63-linked chains are a virus-internalization signal [48]. In addition, K11-linked ubiquitin chains have been connected to degradation of substrates of the anaphase-promoting complex in cell cycle regulation [49], [50]. The Salmonella T3E E3 ubiquitin ligase SspH2, which similarly to XopL selectively interacts with the human UBE2D2 enzyme, mediates the formation of primarily K48-linked polyubiquitin chains [44]. Considering the predominance of K11-linked polyubiquitination in the case of the interaction between XopL and UBE2D2 or its plant homolog we speculate that K11-linked ubiquitin chains may play an important role in plant-pathogen interactions. However, this remains to be elucidated.
Our structural data confirmed that XopL harbors a bona fide LRR domain. The LRR domain is a common feature between XopL and the IpaH- and SspH2- effector E3 ubiquitin ligases mentioned above. While the LRR domain in IpaH plays a regulatory role by inhibiting the E3 activity in the absence of the substrate [18], [19], [51], there is no indication for this kind of mechanism in the case of XopL, as E3 ligase activity is robust in the presence or absence of the LRR domain. However, we were surprised to find that the LRR is involved in suppression of PAMP-elicited gene expression, which we performed using the well-established Arabidopsis protoplast system. According to our data the expression of pNHL10 following elicitation of protoplasts with either flg22 or elf18 peptides was suppressed by the LRR domain, similarly to full-length XopL. This argues for an adaptor function of the LRR domain in which the LRR domain binds a target downstream of PAMP-receptor binding and either downstream or independent of MAPK cascade-signaling, leading to altered gene expression. These results are reminiscent to those reported for the Pseudomonas type III effector AvrPtoB, where suppression of plant immunity by blocking downstream signaling through BAK1-kinase are due exclusively to the two binding domains localized to residues 121–205 and 270–359 [52].
As shown by our in planta ubiquitination profiles, the presence of both the LRR and XL-box domains is essential for XopL-induced reactions. While expression of the XL-box domain in planta resulted in formation of additional polyubiquitin chains, in line with its in vitro activity, only full-length XopL with an intact LRR domain triggered cell death. In addition, expression of the individual XL-box and LRR domain had the opposite effect on expression of the NHL10 promoter, even in the absence of PAMP-response elicitor. This suggests that the LRR domain functions as a protein-protein interaction module necessary for both the suppression of PAMP-elicited gene expression and the cell death phenotype we observed. Thus, we hypothesize that XopL fulfills multiple functions in planta by (i) suppressing PTI via its LRR-region and (ii) ubiquitinating a yet unknown plant substrate(s) whose initial recognition may also require the LRR-region.
In conclusion, characterization of the bacterial pathogen effector XopL uncovered a novel E3 ubiquitin ligase fold that is part of the pathogen repertoire to mimic an otherwise strictly eukaryotic function such as ubiquitination. This underlines the variety of E3 ligases evolved in pathogenic bacteria for subverting host biology. The next challenge is the identification of host targets of XopL that are involved in suppression of plant defenses, as well as determination of the mechanism of action of this unusual E3 ligase.
Escherichia coli cells were cultivated in lysogeny broth medium (LB) at 37°C. Agrobacterium tumefaciens was grown at 30°C in YEB (yeast extract broth) medium and Xcv at 30°C in NYG (nutrient yeast glycerol, [53]) or secretion medium (minimal medium A, [54]) supplemented with 10 mM sucrose and 0.3% casamino acids. Plasmids were introduced into E. coli and A. tumefaciens by electroporation and into Xcv by conjugation, using helper plasmid pRK2013 in triparental matings [55].
The near-isogenic pepper (Capsicum annuum) cultivars ECW, ECW-10R and ECW-30R [56] were grown at 23°C with 60% relative humidity and 16 h light and Nicotiana benthamiana plants were grown at 22°C with 60% relative humidity and 16 h light. Xcv strains were inoculated with a needleless syringe into leaves at 108 colony-forming units (cfu)/ml in 10 mM MgCl2. For in planta transient expression studies, A. tumefaciens strain GV3101 [57] was incubated in inoculation medium (10 mM MgCl2, 5 mM MES, pH 5.3, 150 µM acetosyringone) and inoculated into leaves at 8×108 cfu/ml.
Xanthomonas in vitro secretion experiments were performed as described [58]. Equal amounts of total bacterial cell extracts and culture supernatants were analyzed by SDS-polyacrylamide gel electrophoresis (PAGE) and immunoblotting following standard protocols. To exclude bacterial lysis, blots were routinely reacted with an antibody specific for the inner membrane lipoprotein HrcJ [59]. To analyze Agrobacterium-mediated protein expression, two leaf discs (0.9 cm in diameter) were frozen and ground in liquid nitrogen, resuspended in 100 µl 8 M urea and 50 µl 5× Laemmli buffer, and boiled for 10 min. Proteins were separated by SDS-PAGE and analyzed by immunoblotting. We used polyclonal antibodies for detection of AvrBs3 [60] and ubiquitin (Abcam, Cambridge, U.K.), and a monoclonal Strep-tag antibody (IBA GmbH, Göttingen, Germany). Horseradish peroxidase-labeled α-rabbit and α-mouse antibodies (Amersham Pharmacia Biotech, Piscataway, N.J., U.S.A.) were used as secondary antibodies. Antibody reactions were visualized by enhanced chemiluminescence (Amersham Pharmacia Biotech).
RNA extraction from Xanthomonas, cDNA synthesis and reverse transcription polymerase chain reaction (RT-PCR) experiments were performed as described [61].
To generate a genomic deletion of xopL 2 kb and 1.1 kb fragments upstream and downstream of xopL were amplified by PCR from genomic DNA of Xcv 85-10 using oligonucleotides harboring appropriate restriction sites. PCR-fragments were cloned into the suicide vector pK18mobsac [62]. The resulting constructs were conjugated into Xcv strain 85-10 and xopL deletion mutants were selected by PCR.
To generate binary expression constructs, the coding sequence of xopL was amplified by PCR, fused to a Strep-tag-coding sequence, cloned into pENTR/D-TOPO (Invitrogen GmbH, Karlsruhe, Germany) and recombined into pGWB2 [63] using GATEWAY technology (Invitrogen). XopL-derivatives listed in Table S3 (in Text S1) were generated using the Phusion Site-Directed Mutagenesis Kit (Fisher Scientific GmbH, Schwerte, Germany). To generate avrBs3Δ2-fusions, the promoter and 5′coding sequence of xopL were amplified by PCR, cloned into pENTR/D-TOPO and recombined into pL6GW356 [64]. Sequences of oligonucleotides are available upon request.
Triplicates of five leaf discs each (0.9 cm in diameter) were harvested 2 dpi and 4 dpi. Measurements were carried out as described [65]. Values (n = 3) for XopL and each of its derivatives were compared to GFP (control) using unpaired Student's t-test.
In vitro E3 ligase assays were performed as described [17], [19]. Arabidopsis E2s used in this study were amplified from the CD4-16 cDNA library from the Arabidopsis Biological Resource Centre (ABRC, www.arabidopsis.org/abrc) and cloned into expression plasmid p15Tv-L (gi |134105575|). Plasmids encoding AtUBC28-variants R5A, F62A, K63A and A96D were generated using the Quick Change Site-Directed Mutagenesis II kit (Agilent Technologies Canada, Inc., Mississauga, Canada). The E1-enzyme, ubiquitin and ubiquitin mutants were purchased from Boston Biochem (Cambridge, USA). Ubiquitin- and His antibodies were purchased from EMD Millipore (Billerica, USA) and Qiagen (Toronto, Canada), respectively. His-tagged UBE2D2 was prepared as described [19], and Arabidopsis wild-type and mutant His-tagged E2s were purified accordingly. Sequences of oligonucleotides are available upon request.
In vitro ubiquitination reactions were analyzed by LC-MS/MS on OrbitrapVelos as described [66]. Briefly, 20 µl reactions containing 0.029 µM E1, 3 µM E2, 6 µM E3, 25 µM ubiquitin and 10 mM ATP (in 50 mM Tris pH 7.5 buffer, with 0.1 M NaCl, 10 mM MgCl2 and 0.5 mM DTT) were incubated at 25°C for 3 hours. Reactions were stopped by the addition of 2× Laemmli buffer and incubation for 5 minutes at 95°C. Proteins were separated by SDS-PAGE, and the gel band corresponding to >100 kDa excised and trypsinized. 1/10th of each band was analyzed in duplicates.
Fragments of XcvXopL (XCV3220, gi 28872465) and Xanthomonas campestris pv. campestris str. ATCC 33913 XopL (XCC4186, gi 21233603) were cloned into expression plasmid p15Tv-L, followed by transformation of E. coli BL21(DE3)-RIPL (Agilent Technologies Canada, Inc., Mississauga, Canada). After optimizing solubility, E. coli cells expressing XopL fragments were cultured in 1 l LB at 37°C to an optical density (600 nm) of approximately 1.2, before IPTG was added to induce protein expression. Selenomethionine-enriched protein was produced following growth and induction of cells in SeMet high-yield media (Shanghai Medicilon, Shanghai, China). After induction, bacteria were incubated overnight on a shaker at 25°C. Cells were harvested by centrifugation, disrupted by sonication, and the insoluble material was removed by centrifugation. XopL fragments were purified using Ni-NTA affinity chromatography and dialyzed at 4°C in 10 mM HEPES (pH 7.5), 500 mM NaCl and 0.5 mM TCEP, concentrated to >15 mg/ml and stored at −70° C.
Crystallization trials were performed at room temperature using hanging-drop vapor diffusion with an optimized sparse matrix crystallization screen [67], with or without limiting amounts of proteases [68] including TEV. XopL[aa 144–450] crystals were grown at 25 mg/ml. The XopL[aa 144–450] crystal used for data collection (see Table 1) was grown from a crystallization liquor containing 0.2 M Potassium Sulfate and 20% PEG3350 monodisperse (Hampton Research, Aliso Viejo, USA) and cryoprotected in a similar buffer containing 10% glycerol and flash-frozen in liquid nitrogen, while the XopL[aa 474–660] crystal was grown using a protein concentration of 26 mg/ml from a crystallization liquor containing 0.1 M Tris pH 8.5, 0.2 M Sodium Acetate, 30% PEG4K and 4% ethylene glycol, cryoprotected using Paratone-N oil (Hampton Research) and flash-frozen in liquid nitrogen.
The structure of XopL[aa 144–450] was determined by a crystal derived from selenomethionine-enriched protein with SAD phasing using a peak wavelength of λ = 0.97937 Å. Diffraction data were collected at 100° K at APS beamline 19-BM. Diffraction data were integrated and scaled at the beamline using HKL3000 [69]. Positions of heavy atoms were found using SHELXD [70], followed by solvent flattening using SHELXE [71], which was in turn used to automatically build an initial model using ArpWARP [72], all used within the CCP4 program suite [73]. The model was improved by alternate cycles of manual building and water-picking using COOT [74] and restrained refinement against a maximum-likelihood target with 5% of the reflections randomly excluded as an Rfree test set. These refinement steps were performed using REFMAC in the CCP4 program suite. In addition we refined using Phenix.refine from the PHENIX crystallography suite [75], [76]. The final model contained a nearly complete chain containing 4 residues in the Ni-affinity tag and residues 145–437, in which the C-terminal Gly residue from the tag, residues 144, 297 and 438–450 were omitted due to protein disorder, and was refined to an Rwork and Rfree of 17.1 and 22.6%, respectively, including TLS parameterization [77], [78]. The structure of XopL[aa 474–660] was also solved by SAD phasing at peak wavelength (λ = 0.97921 Å) using a selenomethionine-enriched crystal. Structure solution, model building and refinement followed a similar protocol as for XopL[aa 144–450]. However, during refinement, phenix.xtriage, as part of the PHENIX crystallography suite, we detected merohedral twinning with twin law h, -h-k, -l and a twinning fraction of 0.273. Refinement then proceeded with a newly derived Rfree set to take the twinning into consideration. As stated above, the final model contained three molecules in the asymmetric unit. Molecule A contains a complete chain involving the 5 most C-terminal residues from the Ni-affinity tag followed by residues 474–639. No electron density for residues 641–660 was observed due to protein disorder. Molecules B and C contained a very similar chain. In addition, in molecule B, the 6 most C-terminal residues of the Ni-affinity tag were modeled as well as residues 640–642. In molecule C, residues 474–476 were not modeled due to protein disorder, but residue 640 was. The final model (to 1.8 Å) was refined to an Rwork and Rfree of 15.2 and 19.4%, respectively.
Data collection, phasing and structure refinement statistics for both structures are summarized in Table 1. The Ramachandran plot generated by PROCHECK [79] showed very good stereochemistry overall with 99.6 and 100% of the residues in the most favored and additional allowed regions for XopL[aa 144–450] and XopL[aa 474–660], respectively.
Transient expression experiments with A. thaliana (Col-0)-derived protoplasts were carried out as described [28]. Protoplast samples were co-transformed with the NHL10 promoter-luciferase construct [27], [28], pUBQ10-GUS [80] and either p35S-effector gene constructs (xopL, xopLQ612A, xopLCTD[aa450–660]) or p35S-cfp as control (10 µg total DNA per 100 µl protoplasts; ratio 1∶1∶1). Activity of MAPKs was determined by protein extraction and immunoblotting using a specific pTepY-antibody as described previously [25]. GUS-activity was determined by measuring the turnover of 4-MUG (4-Methylumbelliferyl-β-D-glucuronide) with a Cytofluor II Platereader (Millipore Corp.; excitation 380 nm, emission 460 nm).
Coordinates for the XopL LRR domain (XopL[aa 144–450]) and the C-terminal domain (XopL[aa 474–660]) structures were deposited at the Protein Data Bank with accession codes 4FCG and 4FC9, respectively. XCV3220 (XopL) and XCC4186 (XccXopL) are targets APC108260 and APC105826 of the Midwest Center for Structural Genomics, respectively.
|
10.1371/journal.pntd.0001177 | Comparison of Argentinean Saint Louis Encephalitis Virus Non-Epidemic and Epidemic Strain Infections in an Avian Model | St. Louis encephalitis virus (SLEV, Flavivirus, Flaviviridae) is an emerging mosquito-borne pathogen in South America, with human SLEV encephalitis cases reported in Argentina and Brazil. Genotype III strains of SLEV were isolated from Culex quinquefasciatus mosquitoes in Cordoba, Argentina in 2005, during the largest SLEV outbreak ever reported in South America. The present study tested the hypothesis that the recent, epidemic SLEV strain exhibits greater virulence in birds as compared with a non-epidemic genotype III strain isolated from mosquitoes in Santa Fe Province 27 years earlier. The observed differences in infection parameters between adult House sparrows (Passer domesticus) that were needle-inoculated with either the epidemic or historic SLEV strain were not statistically significant. However, only the House sparrows that were infected with the epidemic strain achieved infectious-level viremia titers sufficient to infect Cx. spp. mosquitoes vectors. Furthermore, the vertebrate reservoir competence index values indicated an approximately 3-fold increase in amplification potential of House sparrows infected with the epidemic strain when pre-existing flavivirus-reactive antibodies were present, suggesting the possibility that antibody-dependent enhancement may increase the risk of avian-amplified transmission of SLEV in South America.
| St. Louis encephalitis virus (SLEV, Flavivirus, Flaviviridae) is an emerging arbovirus in South America, with human SLEV encephalitis cases reported in Argentina and Brazil. Genotype III strains of SLEV were isolated from mosquitoes during the largest SLEV outbreak ever reported in South America (Córdoba, Argentina, 2005). These strains are related to a non-epidemic genotype III SLEV strain isolated in 1979 in Santa Fe Province, Argentina. There is currently no clear explanation for the reemergence of SLEV in Argentina. This study tested the hypothesis that the epidemic strain exhibited greater virulence compared to a non-epidemic genotype III strain in an avian model, the House sparrow (Passer domesticus). House sparrows were susceptible to infection with Argentinean SLEV strains; however, the proportion of birds that became detectably viremic was low for both strains. Although no significant difference was detected between both strains, House sparrows inoculated with epidemic strain developed higher and longer viremias than those inoculated with non-epidemic strain. The virus amplification role of House sparrows was apparently enhanced when they had previous flavivirus immunity. The evolutionary/introduction process of a more viremogenic SLEV strain and the immunological interactions among antigenically-related flaviviruses will undoubtedly affect the continued reemergence of SLEV in Argentina.
| The geographic distribution of St. Louis encephalitis virus (SLEV, Flavivirus, Flaviviridae) encompasses tropical, sub-tropical and much of the temperate-tropical zones of the Western Hemisphere and therefore, most of the populated land masses of North and South America [1]. In the United States of America (USA), this virus is known to be naturally maintained by transmission cycles between several Culex (Cx.) mosquito species and a variety of bird species, including the House sparrow (Passer domesticus) [2]. SLEV is an emerging arbovirus in South America, with febrile illness and encephalitis cases reported in Argentina in 2002 and 2005, and in Brazil in 2004 and 2006 [3]–[6].
In Argentina, SLEV reemerged in the central region (i.e., Córdoba and Santa Fe Provinces) in 2002, when two cases of encephalitis and three fever cases were reported in humans (Morales MA and Enria D, unpublished data) [5]. In 2005, 47 laboratory-confirmed clinical cases of SLEV infection, including nine fatalities, were reported from Córdoba Province [7]. Two genotype III SLEV isolates were obtained from Cx. quinquefasciatus mosquitoes during this outbreak [4]. Genotype III SLEV was previously isolated from mosquitoes collected in Santa Fe Province 27 years earlier with no human encephalitis cases reported [8]. The cause of the 2005 outbreak remains unknown, but may have derived from virological factors, changes in populations of vectors and/or avian amplifying hosts, and/or environmental conditions [9]–[11]. These were the first reported outbreaks of SLEV-induced encephalitis in South America. By March 2010, the Health Ministry of Buenos Aires Province reported a total of five confirmed and five probable human cases of SLEV [12]. In Brazil, SLEV was identified as the etiologic agent of a small meningoencephalitis outbreak among humans in Sao Paulo State in 2006 [3].
We propose that the recently isolated genotype III strain has pathogenic properties in an avian model, and more specifically, House sparrows. These properties presumably favor epidemic activity relative to the historical genotype III strain that was thought to be enzootic. To test this hypothesis, we evaluated the viremogenic and pathogenic capacities of both recent and historic genotype III isolates from Córdoba in House sparrows. In addition, we evaluated cross-protection conferred by heterologous flavivirus-neutralizing antibodies in a small number of SLEV-challenged House sparrows.
House sparrows (hatch-year or older; i.e., adults) were captured in mist nets during January 2007 in Larimer County, Colorado and housed in commercial cages (Safeguard, Inc., New Holland, PA). Mixed bird seed and water were provided ad libitum. The maintenance and care of experimental animals in this study complied with institutional guidelines and the National Institutes of Health guidelines for the humane use of laboratory animals. All animal use was conducted at Colorado State University under approval from the Institutional Animal Care and Use Committee (approval 09-137A).
The CbaAr-4005 (epidemic) and 79V-2533 (non-epidemic) SLEV strains were isolated from pools of adult female Cx. quinquefasciatus collected in 2005 in Córdoba Province and Cx. (Culex) spp. collected in 1978 in Santa Fe Province, Argentina, respectively [4], [8]. The CbaAr-4005 strain had previously been passaged four times and the 79V-2533 strain two times, and all passages were in African Green monkey kidney (Vero) cells.
House sparrows were needle-inoculated subcutaneously over the breast with 3,000 plaque-forming units (PFU) with one of the two SLEV strains (or mock BA-1 inoculation for the negative control group) in 0.1 mL (milliliter) of BA-1 diluent (Hanks M-199 salts, 0.05 M Tris, pH 7.6, 1% bovine serum albumin, 0.35 g/L of sodium bicarbonate, 100 units/mL of penicillin, 100 µg/mL of streptomycin, 1 µg/mL of Fungizone). Seven seronegative birds were inoculated with 79V-2533 (Group A), 8 seronegative (Group B) and 4 flavivirus-seropositive birds (Group C) were inoculated with CbaAr-4005, and 10 birds (Group D) were mock-inoculated with BA-1 to serve as a non-inoculated, morbidity/mortality control group. Following inoculation, birds were monitored for clinical signs (e.g., lethargy, fluffed feathers, decreased activity, and emaciation) every 12 h. From 1–7 days post-inoculation (DPI), 0.1 mL of blood was collected by jugular venipuncture from each bird (including controls) and diluted in 0.45 mL of BA-1 in 2-mL cryovials. The samples were centrifuged for separation of serum (diluted approximately 1∶10) and stored at −80°C until assayed for infectious viral particles.
The detection and titration of viruses in blood samples was performed using a double-overlay Vero cell plaque assay [13]. The second overlay contained neutral red dye and was added on 5 DPI; plaques were counted on 6 and 7 DPI. Each sample was titrated in duplicate using serial 10-fold dilutions in BA-1 diluent. The detection threshold for SLEV in serum was 101.7 PFU/mL.
Pre-inoculation status for flavivirus-reactive antibodies was determined by blocking ELISA (enzyme-linked immunosorbent assay) using the flavivirus group-reactive monoclonal antibody 6B6C-1 and sonicated suspension of CbaAr-4005 antigen [14]. Serum samples collected between 10–14 DPI were assayed using the plaque reduction neutralization test (PRNT) using CbaAr- 4005 on Vero cell monolayer prepared in six-well cell culture plates (Costar Inc, Cambridge, MA) [13]. Between 10 and 14 DPI, all surviving House sparrows were bled (0.6 mL) and whole blood was placed in Microtainer serum separator tubes (Becton Dickinson, Franklin Lakes, NJ), centrifuged for separation of serum, stored at −20°C, and heat inactivated at 56°C for 30 min prior to testing. Serum samples were diluted 1∶10 in BA-1 for antibody screening, and serial 2-fold dilutions in duplicate were used to determine reciprocal endpoint 80% SLEV antibody titers (PRNT80). Because West Nile virus (WNV) is endemic in Colorado, flavivirus-reactive antibodies were presumed to be due to previous infection with WNV, which is commonly detected in local House sparrows in northern Colorado (N. Komar, unpublished data).
Values for vertebrate reservoir competence index (C) were calculated according to the formula:where s is susceptibility to infection (a proportion of viremic birds), i is the extrapolated mean daily infectiousness (the proportion of feeding Cx. quinquefasciatus that are expected to become infected after a viremic blood meal and surviving the extrinsic incubation period; [i = 0.5475*log viremia (PFU/ml) – 1.6526], and d is mean duration of infectious viremia (in days) [15]. C indicates the relative inherent potential for a vertebrate host to amplify a pathogen to sufficient levels to infect vectors. Infectiousness was extrapolated using data published by Mitchell et al. on oral infectivity of Cx. quinquefasciatus for 78V-6507 SLEV Argentinean strain. These data indicated an approximate threshold of 103.02 PFU/mL for infectious viremia titers [16].
Non-SLEV immune sparrows that were inoculated were included in the analyses if they had evidence of infection (i.e., detectable SLEV-viremia between 1–7 DPI and/or seroconverted by 10–14 DPI). Flavivirus-seropositive birds that were inoculated were included in the analyses if they had detectable viremia between 1–7 DPI. To compare susceptibility to infection and mortality among experimental groups, Fisher's exact test was used. Mean durations of viremia and log-transformed mean peak viremia measures were compared using a Poisson generalized linear model and ANOVA, respectively, with significance threshold α = 0.05. Individuals were considered immune if they had detectable anti-SLEV antibodies (PRNT80≥10). Individuals were considered refractory (i.e., not susceptible) to infection if they did not show evidence of infection either by detectable viremia titers and/or seroconversion.
Four of seven House sparrows inoculated with 79V-2533 (Group A) had evidence of infection and were therefore included in the mathematical analyses. Two had detectable viremia (mean peak titer 103.1 PFU/mL serum, mean duration 1.5 days; Table 1) and two were refractory to infection. Six of eight House sparrows inoculated with CbaAr-4005 (Group B) were similarly included in the analyses. Four of these birds had detectable viremia (mean peak titer 105.3 PFU/mL serum, mean duration 2.75 days) and two were refractory. Two of four anti-flavivirus antibody positive birds that had been inoculated with CbaAr-4005 (Group C) were also included in the analyses; these two birds developed detectable viremia (mean peak titer was ≥107.1 PFU/mL serum, mean duration ≥3.5 days (Table 1). The uncertainty in the upper limit of the means for Group C is due to the peak viremia being in one of House sparrows occurring on the last day of sampling (i.e., 7 DPI; Figure 1).
Mortality was observed in all four of the study groups, including the non-inoculated control group. Through 7 DPI, mortality occurred in 3 of 7 (Group A), 4 of 8 (Group B), 2 of 4 (Group C), and 3 of 10 (Control Group) birds in the four groups. The rates did not differ significantly among any of the groups (P>0.5).
Infection parameters (i.e., susceptibility to infection, mean duration of viremia, and mean peak viremia titers) were compared between House sparrows in Groups A and B, A and C, and B and C. Sample sizes were insufficient to detect statistically significant differences in any of these pair-wise comparisons. However, considering that a threshold viremia titer for infection of vector mosquitoes with the 79V-2533 SLEV strain could be estimated from published data, and that the viremia titers observed in the experimental groups appeared to be skewed in relation to this threshold, we used these data to assess the biological significance of the data. A rough estimate of the relative reservoir competence of House sparrows in each experimental group indicated that adult House sparrows were not competent amplifying hosts for the non-epidemic SLEV strain (Group A). On the other hand, House sparrows inoculated with the epidemic SLEV strain (Group B) were theoretically able to produce 10 times more infectious mosquitoes than those inoculated with the non-epidemic strain (Table 2). House sparrows with prior flavivirus infection (Group C) would infect three mosquitoes for every one mosquito infected by flavivirus-naïve House sparrows (Group B) inoculated with the epidemic SLEV strain (CbaAr-4005) (Table 2).
Variation in biological characteristics (i.e., viremia profiles, neuroinvasiveness, virulence, and pathogenicity) among SLEV strains has been previously described [17], [18], [19]. Bowen et al. [18] compared infection parameters for 44 strains of SLEV, including 14 from South America, in adult and juvenile House sparrows, and found that strains could be categorized as low, intermediate, or high viremogenic capacity [18]. In the present study, these same infection parameters in adult House sparrows were compared for a non-epidemic strain (presumably not pathogenic in humans) from Argentina and a novel, epidemic strain isolated during the first SLEV outbreak ever recognized in South America. The two strains were isolated from the same subtropical region of northern Argentina, but 27 years apart. The recently discovered strain (CbaAr-4005) is associated with an outbreak of human encephalitis resulting in nine fatalities in Córdoba, Argentina. Despite alignment of the genomes of these two strains within the same genotype, the epidemic strain appears to be more viremogenic than the non-epidemic strain in our model avian host. Comparison of the complete genome sequences of the two strains revealed an amino acid difference at position 249 in the NS3 protein. This is the same position as described for bird-virulent strains of WNV (T249P) [20].
Low or undetectable viremia titers among House sparrows in the present study may have lessened the likelihood of detection of statistically significant differences. The sample sizes were similar to those used by Bowen et al. [18], yet applying their criteria for categories of viremogenic capacity failed to differentiate the two Argentinean strains into either the high or low viremogenic categories. However, using previously published data [16] for vector competence of Cx. quinquefasciatus mosquitoes, data obtained in the present study indicated that adult House sparrows infected with the non-epidemic strain were incompetent amplifiers of SLEV. House sparrows inoculated with the epidemic strain were theoretically competent, based on a biologically significant difference in the observed data for the two strains.
Viremia profiles varied among individuals, including among those inoculated with the same SLEV strain. Intrinsic avian factors such as genetics, age, and immunocompetence are likely to variably-affect host responses to infection. The inverse effect of age over resulting viremia profiles after arbovirus infection (i.e., older individuals develop lower viremia titers than younger individuals) has been well documented [18], [21], [22], [23]. Although all House sparrows in the present study were considered adults, we could not further specify age beyond >1 year of age, so that actual ages may have varied widely and therefore influenced viremic responses. In addition, while we observed no visible health effects in these birds during the pre-inoculation period, underlying health conditions may also have affected their responses to infection. In addition, previous studies have identified a specific gene (i.e., the Oas1b gene; 2′-5′ oligoadenylate synthetases) as a determining factor for resistance to infection in animals (i.e., humans, mice, chickens and horses) [24], [25], [26]. Genetic variability in this gene and possibly others among inoculated House sparrows in this study could have also caused differences in responses to infection.
Mortality was observed in each of the four study groups, one of which was the non-inoculated control group. No statistical difference was detected in mortality among groups. Therefore, we believe that the observed mortality was attributable to captivity, handling stress, and possibly the aforementioned factors, and not attributable to SLEV infection. Further, SLEV is not historically believed to cause morbidity and mortality in birds either experimentally or naturally infected [2], [22], [27], [28].
The relatively small proportion of House sparrows with detectable viremia titers, which were generally low, could indicate co-adaptation of South American SLEV strains among the resident avifauna. House sparrows have a broad geographic range and are considered the main amplifying host of SLEV in south and central USA, and House finches (Carpodacus mexicanus) the main host for SLEV strains in the western coast of the USA (e.g., California) [2]. Argentinean SLEV strains are not well amplified by resident House sparrows in Argentina. [LA Diaz, unpublished data] However, Picui ground doves (Columbina picui) and Eared doves (Zenaida auriculata) are amplifying hosts for SLEV strains in Argentina [Diaz LA, unpublished data], further supporting the idea that SLEV strains have become adapted to their respective resident bird populations in both the USA and Argentina.
In the present study, House sparrows with evidence of previous (i.e., natural) flavivirus infection that were subsequently inoculated with SLEV strain CbaAr-4005 developed higher viremia titers of longer duration than naïve House sparrows, as evidenced by a 3-fold greater reservoir competence in the former. While this assessment is based on a small number of available sparrows with previous flavivirus immunity and more data are needed, it suggests that SLEV activity may be enhanced by previous circulation of WNV or other flaviviruses among avian host populations. Pre-existing flavivirus-reactive antibodies in House sparrows (perhaps homologous anti-SLEV or heterologous anti-WNV) may potentiate subsequent SLEV infection, as indicated by the higher resulting viremia titers in these birds following challenge. Ludwig et al. reported a similar finding in a laboratory controlled experimental SLEV infection of House sparrow chicks circulating various levels of homologous maternal antibodies [29]. Viremia titers of greater magnitude and duration were observed in nestling House sparrows from SLEV-inoculated mothers versus nestlings from naïve mothers. As maternal SLEV-neutralizing antibody titers wane in nestling birds, antibody-mediated amplification of serum virus titers may result following SLEV infection. Antibody-dependent enhancement of Flavivirus infections is a well-known phenomenon observed among humans with secondary dengue virus infection [30]. However, House finches (Carpodacus mexicanus) with pre-existing antibodies to WNV were protected from experimental challenge with a North American strain of SLEV [31].
With many closely-related flaviviruses circulating in South America, the possibility of antibody-dependent enhancement of SLEV infections in birds requires further investigation, especially if an association with emergence of SLEV epidemic activity is to be corroborated. In Brazil, for example, there are at least ten circulating flaviviruses, including Bussuquara, Cacipacoré, Dengue (serotypes 1 to 4), Igaupe, Ilheus, Rocío, SLE, and yellow fever viruses [32]. In addition, WNV and SLEV are now sympatric throughout the Americas from southern Canada to central Argentina, and co-circulation has been observed in some locations with active SLEV surveillance programs such as Florida, Texas and California [33], [34]. WNV has been active in Argentina since at least 2004, as indicated by detection of specific WNV-reactive neutralizing antibodies in sera collected from a Rufous hornero (Furnarius rufus), a resident passerine, sampled on 5 January 2005 [35]. The appearance of WNV in central Argentina shortly before an unprecedented encephalitis epidemic caused by SLEV may not have been a coincidence, but rather potentially the consequence of the same antibody-dependent enhancement that we observed in our small study.
Currently, there is no definitive explanation for the reemergence of SLEV in Argentina. Two possible explanations gain some support from data in the present study. The first is the evolution or introduction of a SLEV strain with increased viremogenicity. The higher viremia titers generated by infections with the CbaAr-4005 strain in local amplifier species such as the Eared dove [36] will theoretically lead to increased numbers of infected vectors. Furthermore, this epidemic strain appears to broaden the number of avian species that are likely to be competent amplifying hosts relative to the non-epidemic 79V-2533 strain. The identification of a limited number of specific amino acid substitutions between the two genotype III strains used in the present study helps direct future research to identify molecular virulence factors [37]. The second possible explanation is that recent introduction of WNV into the region has boosted the reservoir competence of local avian reservoir hosts for SLEV through antibody-mediated enhancement. WNV activity in the USA had a sobering impact on wild bird populations, resulting in the deaths of millions of birds, while SLEV does not cause avian mortality. However, since the introduction of WNV to Argentina in 2004, there have been no reports of associated avian mortalities. These explanations are not mutually exclusive, and other factors may be involved. In order to provide more support for our findings, further studies should focus on the immunological interactions among antigenically-related flaviviruses in birds and other potential amplifying hosts.
The possibility that this newly discovered epidemic SLEV strain may spread within Argentina as well as to other regions of Central and South America represents an important public health threat. In early 2010, SLEV-associated encephalitis cases in humans were reported in Buenos Aires Province [12]. During this outbreak, molecular detections of SLEV were made. BLAST analyses revealed that the nucleotide sequence of the 232 pb NS5 polymerase amplified fragment had 100% homology with that of the epidemic CbaAr-4005 SLEV strain (GenBank accession # FJ753286.1) (L. Valinotto, unpublished data), indicating an epidemic association with a genotype III SLEV strain. Surveillance for vector-borne pathogens remains an unattended civic priority across the globe. In the absence of early detection through environmental surveillance, clinicians should be on alert for neurologic syndromes in human patients attributed to this novel strain of SLEV.
|
10.1371/journal.pgen.1002915 | Role of Mex67-Mtr2 in the Nuclear Export of 40S Pre-Ribosomes | Nuclear export of mRNAs and pre-ribosomal subunits (pre40S and pre60S) is fundamental to all eukaryotes. While genetic approaches in budding yeast have identified bona fide export factors for mRNAs and pre60S subunits, little is known regarding nuclear export of pre40S subunits. The yeast heterodimeric transport receptor Mex67-Mtr2 (TAP-p15 in humans) binds mRNAs and pre60S subunits in the nucleus and facilitates their passage through the nuclear pore complex (NPC) into the cytoplasm by interacting with Phe-Gly (FG)-rich nucleoporins that line its transport channel. By exploiting a combination of genetic, cell-biological, and biochemical approaches, we uncovered an unanticipated role of Mex67-Mtr2 in the nuclear export of 40S pre-ribosomes. We show that recruitment of Mex67-Mtr2 to pre40S subunits requires loops emanating from its NTF2-like domains and that the C-terminal FG-rich nucleoporin interacting UBA-like domain within Mex67 contributes to the transport of pre40S subunits to the cytoplasm. Remarkably, the same loops also recruit Mex67-Mtr2 to pre60S subunits and to the Nup84 complex, the respective interactions crucial for nuclear export of pre60S subunits and mRNAs. Thus Mex67-Mtr2 is a unique transport receptor that employs a common interaction surface to participate in the nuclear export of both pre-ribosomal subunits and mRNAs. Mex67-Mtr2 could engage a regulatory crosstalk among the three major export pathways for optimal cellular growth and proliferation.
| Eukaryotic pre-ribosomal subunits (pre40S and pre60S) are one of the largest RNA containing cargos that are transported from the nucleus through the nuclear pore complex (NPC) into the cytoplasm. While genetic approaches in budding yeast have identified bona fide export factors for pre60S subunits, little is known regarding nuclear export of pre40S subunits. The conserved transport receptor Mex67-Mtr2 (TAP-p15 in humans) binds mRNAs and pre60S subunits in the nucleus, and facilitates their passage through the NPC, by interacting with Phe-Gly (FG)-rich nucleoporins that line its transport channel. Here, we report an additional role of Mex67-Mtr2. We show that Mex67-Mtr2 binds pre40S subunits via loops present on its NTF2-like domains and that the C-terminal FG-rich nucleoporin interacting UBA-like domain within Mex67 contributes to the transport of pre40S subunits to the cytoplasm. Remarkably, the same loops of Mex67-Mtr2 also contribute to the nuclear export of pre60S subunits and mRNAs. Thus, the transport receptor Mex67-Mtr2 could employ a common interaction surface to engage a regulatory crosstalk among the three major export pathways.
| All living cells expend a significant proportion of their cellular energy to manufacture ribosomal subunits [1]. To construct ribosomal subunits, eukaryotic cells assemble >70 ribosomal proteins (r-proteins) with four different ribosomal RNA (rRNA) species [2], [3]. Transcription machineries (RNA polymerases I, II and III) are co-ordinated to ensure high efficiency and accuracy of ribosome production; Pol-I and Pol-III synthesize the rRNAs destined for 60S (25S, 5.8S, 5S rRNA) and 40S (18S rRNA) subunits, whereas Pol-II transcribes the mRNAs for the ribosomal proteins. >200 non-ribosomal factors, also termed trans-acting factors, aid the assembly and maturation of eukaryotic pre-ribosomal subunits [2], [3].
Our current understanding of eukaryotic ribosome assembly and intracellular transport has been shaped mainly by a combination of genetic, cell-biological and proteomic approaches applied to the model organism budding yeast. The precursor 35S rRNA produced by RNA Pol I transcription of rDNA repeats in the nucleolus is co-transcriptionally modified and associates with small subunit r-proteins and trans-acting factors to form the 90S particle [4]–[6]. The 90S contains mostly small subunit r-proteins but not large subunit r-proteins or trans-acting factors involved in 60S biogenesis [7]. Cleavage of the precursor 35S rRNA releases the pre40S particle and permits the remaining pre-rRNA to associate with large subunit r-proteins and pre60S biogenesis factors to form pre60S particles [7]. The pre40S and pre60S particles hereafter follow independent biogenesis and transport pathways. Pre40S particles undergo few compositional changes as they travel through the nucleoplasm [7], [8]. In contrast, pre60S subunits associate with ∼100 trans-acting factors along their biogenesis pathway and therefore undergo dynamic compositional changes as they travel through the nucleoplasm towards the NPC [9]. The stripping of trans-acting factors by diverse energy-consuming enzymes (ATP-dependent RNA helicases, AAA-ATPases, ABC-ATPases, GTPases) is thought to induce sequential reduction of compositional complexity resulting in export competence [10], [11].
Export competent pre40S and pre60S subunits are passaged separately through NPCs by shuttling export receptors. This is accomplished by transient interactions between export factors bound to pre-ribosomal subunits and Phe-Gly (FG)-repeat nucleoporins that line the transport channel of the NPC [12]–[14]. Transport factors include Ran-dependent exportins that interact with adaptor proteins on the pre-ribosomal subunits and trans-acting factors that can directly interact with FG-rich nucleoporins. The essential exportin Xpo1 (Crm1 in humans) recognizes leucine-rich nuclear export sequences (NESs) present in diverse export cargos [15] including both pre60S and pre40S subunits and mediates their nuclear export [9], [16]–[21]. Nmd3 is the only known essential NES-containing adaptor for pre60S particles that recruits Xpo1 in a RanGTP-dependent manner [16], [17]. Three Xpo1/Crm1 interacting trans-acting factors (Ltv1, Dim2 and hRio2) have been suggested to act as export adaptors for pre40S subunits [22]–[24], but whether they function as bona fide export adaptors in vivo is unclear [12], [24], [25]. An essential adaptor protein that recruits Xpo1 to pre40S particles has yet to be identified. Genetic studies in budding yeast have uncovered trans-acting factors that serve as export factors (Arx1 and Ecm1) for pre60S subunits [26]–[28]. The shuttling HEAT-repeat containing trans-acting factor Rrp12 was shown to promote nuclear export of both pre40S and pre60S subunits [29]. These karyopherin-like factors directly interact with the FG-rich hydrophobic meshwork of the transport channel, thereby allowing pre-ribosomal subunits to efficiently overcome the permeability barrier of the NPC [14], [26], .
Recently, the yeast heterodimeric mRNA transport factor Mex67-Mtr2 (TAP-p15 in humans) was shown to function as an export receptor for pre60S subunits [14]. Mex67-Mtr2 is structurally unrelated to karyopherins and does not rely directly on the RanGTP gradient. However, like karyopherins, Mex67-Mtr2 binds FG-rich nucleoporins and promotes nuclear export of mRNAs and pre60S subunits [14], [30]. Mex67, the large subunit of the heterodimer contains an amino-terminal (N) domain, a leucine-rich repeat (LRR) domain, a nuclear transport factor 2 (NTF2)-like middle domain and a carboxy-terminal ubiquitin associated (UBA-like) domain [31], [32]. The N and LRR domains can directly bind mRNAs or recruit mRNA binding adaptor proteins such as Yra1 [33]–[35]. Mtr2 is structurally related to NTF2, an import factor for RanGDP [36] and forms a functional heterodimer with the NTF2-like middle domain of Mex67 [34], [37]. Structural analysis revealed loops on the NTF2-like domains of Mex67-Mtr2 [38], [39] that contribute to pre60S binding in vivo and in vitro [14]. The loops also contribute to the recruitment of Mex67-Mtr2 to the Nup84 complex, an important structural unit of the NPC. This interaction is crucial for nuclear export of mRNAs, but not pre60S and pre40S subunit export [30]. Both the NTF2-like domains of Mex67-Mtr2 and the C-terminal UBA-like domain of Mex67 can directly bind FG-rich nucleoporins and therefore promote translocation of bound cargos through the NPC [14], [30], [34], [40]–[42].
Large-scale tandem affinity purification (TAP) combined with sensitive mass spectrometry in budding yeast revealed a co-enrichment of Slx9 with Enp1-TAP and Tsr1-TAP that purify both 90S and pre40S particles [43]. Subsequent studies implicated Slx9 in early rRNA processing steps, however its role in early pre40S maturation was not explored [44]. A rationally directed screen to identify novel factors involved in ribosome biogenesis/export in budding yeast showed that the slx9Δ mutant accumulates the 40S reporter S2-GFP in the nucleus [45]. Northern hybridization performed to investigate defects in rRNA processing in the slx9Δ mutant revealed an accumulation of 20S rRNA in the slx9Δ mutant, but not the early 35S precursor rRNA [45]. These observations have implicated Slx9 in late maturation steps in the 40S biogenesis pathway. Here, a genetic screen aimed at uncovering the role of Slx9 in the 40S maturation pathway unexpectedly revealed a role for Mex67-Mtr2 in pre40S subunit export. Together with previous studies, this study identifies Mex67-Mtr2 as a unique transport receptor that functions in the nuclear export of both pre-ribosomal subunits and mRNAs.
Large-scale proteomic approaches in budding yeast revealed co-enrichment of Slx9 with bait proteins that purify both 90S and pre40S pre-ribosomal particles [43]. Affinity purified ProteinA-Slx9 co-enriches mainly 20S rRNA and low levels of precursor 35S rRNA [44]. Sucrose gradient sedimentation showed that Slx9 co-peaks with the 40S pool, and to lesser extent with heavier fractions [45]. To directly investigate association of the nucleolar localized Slx9 (Figure 1A) [46] with pre-ribosomal particles, we purified Noc4-TAP that purifies the early 90S, Enp1-TAP that purifies both 90S and early pre40S subunits, and two late pre40S subunits Hrr25-TAP and Rio2-TAP [7], [8]. Co-enrichment of Slx9 with the purified pre-ribosomes was investigated by Western blotting using α-Slx9 antibody. These analyses revealed that Slx9 co-enriches mainly with Enp1-TAP, that purifies both 90S and pre40S particles (Figure 1B). The Western signal for Slx9 is specific, since it was absent in Enp1-TAP isolated from an slx9Δ mutant (Figure S1A). Only a weak co-enrichment of Slx9 was seen in the 90S (Noc4-TAP) and the late pre40S particle purified using Rio2-TAP (Figure 1B). No association of Slx9 was detected with early to late pre-ribosomal particles in the 60S maturation pathway (Figure 1C). We conclude that Slx9 transiently associates with early 40S pre-ribosomes.
To investigate the role of Slx9 in the 40S maturation pathway, we generated a yeast mutant strain that is deficient for SLX9 (slx9Δ) by disrupting the endogenous SLX9 gene encoded by the open reading frame in wild-type (WT) diploid cells. Tetrad analysis yielded two spores with WT growth rates and two spores with a slow-growth phenotype at 25°C, which carry the SLX9 deletion. We found that growth of the slx9Δ mutant is impaired at 20°C, 25°C and 30°C, as determined by the size of single colonies. At 37°C, the slx9Δ mutant grew nearly like WT cells (Figure 2A).
Next, we investigated the localization of the 40S reporter S2-GFP in the slx9Δ mutant at different temperatures. At 20°C, where growth of slx9Δ cells is significantly impaired (Figure 2A), we found that 89% of slx9Δ cells showed nuclear accumulation of S2-GFP (Figure 2B). At 37°C, where the slx9Δ mutant grows nearly like WT cells (Figure 2A), no nuclear accumulation of S2-GFP was observed (Figure 2B). At intermediate temperatures 25°C and 30°C, 44% and 12% of the slx9Δ cells showed nucleoplasmic accumulation of S2-GFP, respectively (Figure 2B). We wondered whether the nucleoplasmic localization of S2-GFP in the slx9Δ mutant is a direct consequence of accumulating pre40S subunits. To address this, in vivo localization of the 5′ portion of the internal transcribed spacer 1 (ITS1), present within 20S rRNA, was monitored by fluorescence in situ hybridization (FISH). In WT cells, due to rapid nuclear export of pre40S subunits, Cy3-ITS1 (red) is seen in the nucleolus, but not in the nucleoplasm (DAPI, blue) (Figure 3A) [47]. Upon nuclear exit of pre40S subunits, the 5′ portion of ITS1 is efficiently degraded by the nuclease Xrn1 after cytoplasmic cleavage of the 20S rRNA into mature 18S rRNA [47], [48]. At 20°C, where 89% of the slx9Δ cells exhibit nuclear accumulation of S2-GFP, a strong increase in nucleoplasmic signal of Cy3-ITS1 was observed from the merge of DAPI and Cy3-ITS1 fluorescence as compared to SLX9 cells (Figure 3A). The nucleoplasmic accumulation of Cy3-ITS1 in slx9Δ cells is similar to that observed in the xpo1-1 strain, in which the nuclear export of pre40S subunits is impaired (Figure 3B) [47]. At 37°C, where slx9Δ cells did not show nuclear accumulation of S2-GFP (Figure 2B), Cy3-ITS1 was not detected in the nucleoplasm (Figure 3A).
Previously, a large-scale visual screen reported that the slx9Δ mutant is impaired in poly-(A)+RNA export [49]. We sought to analyse the extent of inhibition of poly-(A)+RNA export in the slx9Δ mutant, in comparison to the temperature sensitive mex67-5 cells [31]. The mex67-5 mutant cells after 30 min incubation at non-permissive temperature (37°C) revealed a massive nuclear accumulation of poly-(A)+ RNA in >95% of cells, whereas the slx9Δ mutant did not show such a dramatic phenotype (Figure S2A). Contrary to the previous report, only <1% of slx9Δ cells showed nuclear accumulation of poly-(A)+ RNA in the temperature range between 20–37°C (Figure S2B). Further, the slx9Δ mutant constructed in the W303 and RS453 backgrounds also showed nuclear accumulation of poly-(A)+ RNA in <1% cells, in the temperature range between 20–37°C (data not shown), suggesting that the discrepancy with the previous report might not be due to strain background differences. However, in agreement with earlier findings, the slx9Δ mutant was impaired only in pre40S (Figure 2B and Figure 3), but not pre60S subunit export (Figure S3A) [45].
Thus, the slow growth phenotype of the slx9Δ mutant between 20–30°C co-relates with impaired pre40S subunit export.
Nucleoplasmic accumulation of pre40S subunits as judged by the small subunit reporters S2-GFP and Cy3-ITS1 indicated that Slx9 contributes to late pre40S subunit maturation/export step(s). To gain insight into the role of Slx9 in the 40S maturation/export, we resorted to a genetic screening approach. A high-copy suppressor screen was performed with the aim of identifying genes that suppress the impaired growth of the slx9Δ mutant. The slx9Δ strain was transformed with a multi-copy (2μ) plasmid library and grown at 25°C. Plasmid recovery was performed from fast growing suppressor colonies. Sequence analysis revealed that, in addition to Slx9, the mRNA and pre60S export receptor Mex67 partially rescued the slow growth of slx9Δ mutant at 25°C, as determined by the size of single colonies (Figure 4, left panel). Notably, Mex67 expressed from a single copy (CEN) plasmid was sufficient to partially rescue the impaired growth of slx9Δ mutant (Figure 4, left panel). However, Mtr2, the functional partner of Mex67, expressed alone from either CEN or 2μ plasmids did not rescue the slow growth of the slx9Δ mutant (Figure 4, right panel). As determined by the slightly bigger size of single colonies compared to Mex67 alone, expression of Mex67-Mtr2 from CEN plasmids in the slx9Δ mutant, resulted in a better rescue of the slow growth of the slx9Δ strain at 25°C (Figure 4, right panel). Over-expression of other pre60S export factors (Arx1, Ecm1, Nmd3 and Rrp12) from either CEN or 2μ plasmids did not rescue the slow growth phenotype of slx9Δ cells (Figure 4, right panel and data not shown), suggesting a specific rescue of slx9Δ slow growth by the export factor Mex67-Mtr2.
Next, we investigated whether over-expression of Mex67 could rescue the pre40S subunit export defect seen in the slx9Δ mutant. Over-expression of Mex67 alone (from 2μ and CEN plasmids), or Mex67-Mtr2 (from CEN plasmids), but not Mtr2 alone (from 2μ and CEN plasmids), rescued the nucleoplasmic accumulation of both S2-GFP and Cy3-ITS1 seen in the slx9Δ mutant at 20°C, 25°C and 30°C (Figure 5 and data not shown). Previously, sucrose gradient sedimentation revealed a decreased abundance of 40S subunits in the slx9Δ mutant as compared to the WT, leading to an imbalance between free 40S and 60S subunits [45]. We quantified the 60S/40S ratio in the slx9Δ mutant by performing sucrose gradient sedimentation under conditions that dissociate 40S from 60S subunits [50], [51]. At 25°C, the slx9Δ mutant showed a 60S/40S ratio of 3.5, indicating a ∼40% reduction in the level of 40S subunits, as compared to the WT (Figure 6). We investigated whether the imbalance between 60S and 40S subunits in the slx9Δ mutant could be rescued by the over-expression of Mex67-Mtr2. Over-expression of Mex67 (from 2μ and CEN plasmids), that partially rescued impaired growth of SLX9 deficient cells (Figure 4, left panel), also partially rescued the imbalance between 60S and 40S subunits observed in the slx9Δ mutant (Figure 6). Expression of both Mex67 and Mtr2 from CEN plasmids that resulted in a better rescue of growth of the slx9Δ mutant (Figure 4, right panel) restored the 60S/40S ratio closer to the WT level (Figure 6).
Together, these studies show that expression of an additional copy of Mex67-Mtr2 can compensate the requirement of Slx9 in pre40S subunit export.
Mex67 is a modular protein that contains multiple interaction domains (Figure 7A). The N and LRR domains can directly bind mRNAs or recruit mRNA binding adaptor proteins [33]–[35]. The middle NTF2-like domain of Mex67 forms a functional heterodimer with the NTF2-like domain of Mtr2 [34], [37]. Both the middle NTF2-like domains of Mex67-Mtr2 and the C-terminal UBA-like domain of Mex67 directly bind FG-rich nucleoporins and promote translocation of bound cargoes through the NPC (Figure 7A) [34], [40]–[42]. Over-expression of Mex67-Mtr2 was reported to rescue the pre60S subunit export defect of the nmd3ΔNES1 and ecm1Δarx1Δ mutants [14], [28]. Consistent with the contribution of the loop on the NTF2-like domain of Mex67 in pre60S subunit binding, over-expression of the mex67Δloop allele (deletion of residues 409–435 within the Mex67 loop; Figure 7A) did not rescue the pre60S export of the nmd3ΔNES1 and ecm1Δarx1Δ mutant [14], [28].
We wondered whether alleles of Mex67 and Mtr2 harbouring deletions within the different interaction domains (mex67Δloop, mex67ΔC1 and mtr2Δloop116-137; Figure 7A) could rescue the impaired growth and the pre40S subunit export defect of the slx9Δ mutant. The mtr2Δloop116-137 allele is a shorter deletion within the loop emanating from the NTF2-like domain of Mtr2 (Figure 7A). The mex67Δloop and mtr2Δloop116-137 alleles do not exhibit growth defects and, are not defective in the nuclear export of mRNAs, pre60S and pre40S subunits (Figure S4 and Figure S5) [14], [26], [30]. Both Mex67Δloop and Mtr2Δloop116-137 do not bind pre60S subunits and the Nup84 complex [14], [30], the interactions being important for nuclear export of pre60S subunits and mRNAs, respectively (Table 1 summarizes phenotypes and genetic interactions of mex67Δloop and mtr2Δloop116-137 alleles). We included in our analysis the mex67ΔC1 allele, a deletion of the C-terminal UBA-like domain of Mex67 (deletion of residues 525–599) that directly binds FG-rich nucleoporins and also contributes to efficient translocation of bound cargos through the NPC (Figure 7A) [34], [40]–[42]. To test whether the various alleles could rescue the impaired growth and pre40S subunit export defect of the slx9Δ mutant, the slx9Δ strain was transformed with CEN plasmids carrying mex67Δloop, mex67ΔC1, and mtr2Δloop116-137 alleles. Note that the Mex67Δloop, Mex67ΔC1 and Mtr2Δloop116-137 mutant proteins are expressed similar to WT levels (Figure S1B). While expression of Mex67 partially rescued the impaired growth of the slx9Δ mutant, we found that expression of mex67Δloop, mex67ΔC1 and mtr2Δloop116-137 alleles did not rescue the impaired growth of the slx9Δ mutant. Curiously, we noticed that expression of the mex67Δloop, mex67ΔC1 and mtr2Δloop116-137 alleles hampered the growth of slx9Δ mutant (compare single colony sizes in Figure 7B, left panel), but not SLX9 cells (Figure 7B, right panel). Thus the mex67Δloop, mex67ΔC1 and mtr2Δloop116-137 alleles exhibit a dominant negative behaviour in the slx9Δ mutant, but not in the WT background.
We investigated the reason for the dominant negative nature of these alleles in the slx9Δ mutant. The mex67Δloop and mtr2Δloop116-137 alleles are functionally linked to both mRNA and pre60S subunit export (Table 1). Whereas, nuclear export of pre60S subunits in the single mex67Δloop and arx1Δ mutants remains unaffected, the arx1Δmex67Δloop double mutant is strongly impaired in pre60S subunit export, but not mRNA export [26]. Notably, the mex67Δloop and the mtr2Δloop116-137 alleles, when combined with nup85ΔN133 allele (mutant of Nup85, a component of the Nup84 complex), are impaired in mRNA export, but not pre60S subunit export (Table 1) [30]. These data led us to hypothesize that expression of mex67Δloop, mex67ΔC1 and mtr2Δloop116-137 alleles in the slx9Δ mutant induces defects in pre60S subunit and/or mRNA export, thereby further impairing growth. Surprisingly, expression of these alleles neither induced a pre60S subunit export defect in slx9Δ cells nor did the alleles aggravate the poly-(A)+RNA accumulation seen in <1% of the slx9Δ cells (Figure S3B and Figure S6A). Instead, the pre40S subunit export defect of slx9Δ cells was strongly exacerbated (Figure 7B). At 25°C, ∼44% of slx9Δ cells show nuclear accumulation of S2-GFP (Figure 7B). Expression of the mex67Δloop, mex67ΔC1 and mtr2Δloop116-137 alleles from CEN plasmids strongly aggravated the pre40S subunit defect of the slx9Δ mutant, as judged by strong nuclear accumulation of S2-GFP seen in >80% cells (Figure 7B).
We conclude that the loops on the NTF2-like domains of Mex67-Mtr2 and the C-terminal FG-rich nucleoporin interacting UBA-like domain within Mex67 play crucial roles to compensate the requirement of Slx9 in pre40S subunit export.
The above data led us to test genetic interactions between Mex67-Mtr2, and Slx9. We found that the mex67Δloop and mex67ΔC1 alleles when combined with the slx9Δ mutant were synthetic lethal (Figure 8A). Remarkably, the mtr2Δloop116-137 allele when combined with the slx9Δ mutant was strongly synthetically enhanced (Figure 8B). We investigated whether the slx9Δmtr2Δloop116-137 strain was impaired in mRNA, pre40S and pre60S subunit export. We found that >90% of the slx9Δmtr2Δloop116-137 double mutant exhibited strong nucleoplasmic accumulation of S2-GFP and Cy3-ITS1 in the temperature range between 25–30°C (data not shown). Strikingly, at 37°C, where the slx9Δ and the mtr2Δloop116-137 mutants alone were not impaired in pre40S subunit export, >93% of the slx9Δmtr2Δloop116-137 double mutant showed strong nucleoplasmic accumulation of S2-GFP and Cy3-ITS1 (Figure 9). No defect in the nuclear export of pre60S subunits was observed in the slx9Δmtr2Δloop116-137 strain in the temperature range between 25–37°C (Figure S3C and data not shown). Nuclear accumulation of poly-(A)+RNA found in <1% of the slx9Δ cells was not exacerbated in the double mutant slx9Δmtr2Δloop116-137 strain in the temperature range between 25–37°C (Figure S6B and data not shown). No genetic interaction was observed between Slx9 and other pre60S subunit export factors such as Nmd3, Ecm1 and Arx1 (Figure S3D).
Next, we tested further genetic interactions between Mex67-Mtr2 and factors involved in late maturation and nuclear export of pre40S subunits. The Ran binding protein Yrb2 was reported to be specifically required for proper pre40S subunit export. The yrb2Δ mutant exhibits nuclear accumulation of S2-GFP and reduced abundance of 40S subunits [21], [45]. We found that the mex67Δloop, mex67ΔC1 and mtr2Δloop116-137 alleles were synthetically lethal when combined with the yrb2Δ mutant (Figure 8). Ltv1 has been suggested as a potential adaptor for the export receptor Xpo1 [22], [52]. We found that the ltv1Δ mutant when combined with the mex67ΔC1 allele was synthetically enhanced (Figure 8A). Growth of the ltv1Δ mutant was further impaired, when combined with the mex67Δloop allele, and was unaffected when combined with the mtr2Δloop116-137 allele (Figure 8). The conserved S-adenosylmethionine methyl transferase Bud23 was reported to be required for efficient pre40S subunit export [53]. The bud23Δmutant exhibits nucleoplasmic accumulation of S2-GFP and Cy3-ITS1, suggesting that Bud23 acts in a late pre40S maturation/export step [53]. Notably, the enzymatic activity of Bud23 appears to be dispensable for pre40S subunit export [53]. The bud23Δ mutant when combined with the mex67Δloop and mex67ΔC1 alleles were synthetically enhanced, in particular the bud23Δmex67ΔC1 double mutant showed strong growth impairment (Figure 8A). Growth of the bud23Δ mutant was unaffected when combined with the mtr2Δloop116-137 allele (Figure 8B). The ribosomal protein Rps15 has been implicated in pre40S subunit export, although its precise contribution to export step remains unclear [54], [55]. Rps15 was shown to genetically interact with Slx9 and the pre40S associated factor Bud23 [56]. Here we found that rps15-1, a previously reported impaired allele of Rps15 [56], when combined with the mex67ΔC1 allele is synthetically lethal (Figure 8A). However, growth of rps15-1 mutant remained unaffected when combined with the mex67Δloop and the mtr2Δloop116-137 alleles (Figure 8A and Figure 8B). Large-scale synthetic genetic array (SGA) analysis revealed a genetic interaction between Mex67 and the small subunit ribosomal protein Rps6a [57]. For this analysis, the Guthrie and Krogan laboratories exploited a DAmP (Decreased Abundance by mRNA Perturbation) allele of Mex67. We found that the rps6aΔ mutant combined with the mex67Δloop and mex67ΔC1 was synthetically enhanced (Figure 8A). Growth of the rps6aΔ mutant was not affected when combined with the mtr2Δloop116-137 allele (Figure 8B).
At 37°C, the synthetically enhanced slx9Δmtr2Δloop116-137 double mutant showed strong nucleoplasmic accumulation of Cy3-ITS1 indicating that the severe growth defect stems from impaired nuclear export of pre40S subunits, but not early rRNA processing/biogenesis defects (Figure 9). In order to directly assess whether the growth defects observed in the different synthetically enhanced mutant strains arise from either early rRNA processing/biogenesis defects or impaired nuclear export, we monitored the in vivo localization of ITS1 within 20S rRNA using FISH (Figure S7). Localization of ITS1 would be restricted to the nucleolus, if there were blockage in early rRNA processing/maturation steps upstream of nuclear export. The bud22Δ strain, that is defective in early rRNA processing steps and accumulates 35S rRNA [45], served as a control for our analyses. The Cy3-ITS1 signal is restricted to the nucleolus in the bud22 Δmutant (Figure S7). Notably, all synthetically enhanced double mutant strains analysed exhibited a nucleoplasmic accumulation of ITS1 (Figure S7), in a manner similar to the xpo1-1 strain (Figure 3B). Moreover, the cytoplasmic Cy3-ITS1 signal seen in the ltv1Δ mutant was substantially reduced in the ltv1Δmex67Δloop and ltv1Δmex67ΔC1 double mutants (Figure S7). Thus late nucleoplasmic pre40S maturation steps and/or nuclear export, not early rRNA processing, appear to be inhibited in the synthetically enhanced strains (Figure S7).
Collectively, these data point to a role of Mex67-Mtr2 in the late assembly and/or transport of pre40S subunits.
The rescue of impaired growth and nucleoplasmic accumulation of pre40S subunits of the slx9Δ mutant upon Mex67-Mtr2 over-expression and, in particular the strong synthetic interactions between Mex67-Mtr2 and the Ran-binding protein Yrb2, and the pre40S subunit associated Slx9, led us to test whether Mex67-Mtr2 co-enriches pre40S subunits. We purified early 90S (Noc4-TAP), early (Enp1-TAP) and late (Hrr25-TAP, Rio2-TAP) pre40S particles [7], [8]. As additional controls, we purified several pre60S particles at different stages of maturation in parallel. These include an early (Ssf1-TAP), an intermediate (Rix1-TAP), a late (Arx1-TAP) and a cytoplasmic pre60S particle (Kre35-TAP) [9]. The purified pre-ribosomal subunits were analysed by SDS-PAGE and Western blotting was performed using α-Mex67 and α-Mtr2 antibodies. As previously shown in [14], Mex67 and Mtr2 co-enrich with late pre60S particles (Figure 10A, left panel). Consistent with our genetic and cell-biological observations, we found that Mex67-Mtr2 co-enriches with pre40S subunits and to a lesser extent with a 90S particle (Noc4-TAP; Figure 10A, right panel). These biochemical studies show that Mex67-Mtr2 co-enriches with pre40S subunits in vivo. Next, we examined the nature of the interaction between Mex67-Mtr2 and pre40S subunits, by isolating the Rio2-TAP particle in different NaCl concentrations. These analyses showed that the Mex67-Mtr2 remained stably bound to pre40S subunits at 50 mM and 75 mM NaCl. Association of Mex67 and Mtr2 with Rio2-TAP was lost at 100 mM NaCl (Figure 10B). These data suggest that the interactions between Mex67-Mtr2 and pre40S subunits are likely to be driven by electrostatic interactions.
We investigated how Mex67-Mtr2 binds pre40S subunits. Two observations raised the possibility that the loops present on the NTF2-like domains of Mex67-Mtr2 contribute to pre40S subunit binding: (1) expression of the mex67Δloop and mtr2Δloop116-137 alleles exacerbated the pre40S subunit export defect of the slx9Δ mutant and (2) mex67Δloop and mtr2Δloop116-137 alleles genetically interact with factors required for efficient pre40S subunit export. To address whether the loops of Mex67-Mtr2 contribute to pre40S subunit binding, we purified Enp1-TAP from mex67Δloop and mtr2Δloop116-137 strains and examined the co-enrichment of Mex67Δloop and Mtr2Δloop116-137 mutant proteins by Western blotting using α-Mex67 and α-Mtr2 antibodies (Figure 11, left panel). Note that antibodies against Mex67 and Mtr2 recognize both Mex67Δloop and Mtr2Δloop116-137 mutant proteins in whole cell extract (WCE) samples, respectively (Figure 11, right panel). These analyses revealed that the Mex67Δloop and the Mtr2Δloop116-137 mutant proteins fail to co-enrich with pre40S particles (Figure 11, left panel). Next, we investigated whether the association of Mex67 with pre40S subunits was dependent on the loop of Mtr2 and vice versa. Consistent with our genetic and cell-biological analysis, pre40S subunits affinity purified via Enp1-TAP from mtr2Δloop116-137 and mex67Δloop strains fail to co-enrich Mex67 and Mtr2, respectively (Figure 11, left panel). These analyses revealed that Mex67Δloop and the Mtr2Δloop116-137 mutant proteins fail to co-enrich with pre40S particles. Similar observations were made when Rio2-TAP was purified from mtr2Δloop116-137 and mex67Δloop strains (Figure S1C). We conclude that, like in the case of pre60S subunits and the Nup84 complex, both loops on the NTF2-like domains of Mex67-Mtr2 contribute to pre40S subunit binding.
How pre-ribosomal subunits efficiently overcome the permeability barrier of the NPC is poorly understood. To achieve brisk export rates, pre-ribosomal subunits recruit several transport factors at distinct sites on their surface [13], [14], [29]. Therefore, uncovering export receptors for pre-ribosomal subunits will contribute towards our current understanding to this highly conserved transport process. While genetic approaches in budding yeast have greatly aided the identification factors that participate in pre60S subunit transport, little is known regarding nuclear export of pre40S subunits.
Here, we have uncovered an unanticipated role for Mex67-Mtr2 in 40S pre-ribosome export. A genetic screen revealed that expression of an additional copy of Mex67-Mtr2 rescues the nuclear accumulation of S2-GFP and Cy3-ITS1 of the slx9Δ mutant (Figure 5). Mex67 and Mtr2 genetically interact with trans-acting factors associated with pre40S subunits and involved in their nuclear export. Different alleles of Mex67 and Mtr2 (mex67Δloop, mex67ΔC1 and mtr2Δloop116-137) when combined with the slx9Δ, yrb2Δ, bud23Δ and rps6aΔ mutants were either synthetically lethal or enhanced (Figure 8). Moreover, at 37°C though the slx9Δ and mtr2Δloop116-137 mutants are not impaired in pre40S subunit export, the slx9Δmtr2Δloop116-137 double mutant exhibits strong nucleoplasmic accumulation of S2-GFP and ITS1 (Figure 9). These data are analogous to genetic interactions observed between Mex67 and factors (Nmd3, Arx1 and Ecm1) that directly participate in the nuclear export of pre60S subunits [14], [28]. Over-expression of Mex67, not the mex67Δloop allele, was shown to rescue the pre60S subunit export defect of the nmd3ΔNES1 and ecm1Δarx1Δ mutants [14], [28]. Whereas the single mutant arx1Δ and mex67Δloop are unaffected in pre60S subunit export, the arx1Δmex67Δloop double mutant is strongly impaired in pre60S subunit export [26].
The NTF2 fold is a functionally versatile domain in nuclear transport. The transport factor NTF2, that contains the NTF2 domain and functions as a homodimer, binds RanGDP in the cytoplasm and simultaneously engages in interactions with the NPC for importing RanGDP into the nucleus [36]. The NTF2-like domains of Mex67-Mtr2 exhibit loops on their surface (Figure 7A) that contribute to the interaction with double stranded regions of 5S rRNA in vitro [14], suggesting that Mex67-Mtr2 could interact with a structured rRNA segment on the surface of pre60S subunits. The Hurt laboratory has exploited the mex67Δloop and mtr2Δloop116-137 alleles to uncover another physical interaction of Mex67-Mtr2 loops: the Nup84 complex [30]. Genetic and cell-biological studies revealed that this interaction is crucial for mRNA export, but not pre60S subunit export [30]. In the present study, these alleles have helped to reveal a role for Mex67-Mtr2 loops in pre40S export. Whereas an additional copy of Mex67-Mtr2 rescued the pre40S export defect of the slx9Δ mutant, expression of the mex67Δloop and mtr2Δloop116-137 alleles strongly aggravated this phenotype in a dominant negative manner (Figure 7). Consistent with these genetic and cell-biological data, biochemical studies show that binding of Mex67-Mtr2 to pre40S subunits is mediated by both loops (Figure 11). Whether the loops of Mex67-Mtr2 in vivo interact with an exposed structured rRNA or a protein factor(s) on pre40S subunits remains to be determined. In addition to the loops, other regions of the NTF2-like fold of Mex67-Mtr2 contribute to pre60S subunit binding. Only the mex67ΔloopK343E mutant is impaired in nuclear export of pre60S subunits [14]. These data prompted us to construct alleles of Mex67 that are specifically defective in pre40S subunit export. To this end, several positively charged residues on the NTF2-like fold were mutated in mex67Δloop (R353E, K366E, K370E, K372E, K439E, K442E and K443E). Unfortunately, these mutants did not complement the lethality of the MEX67 null strain (data not shown), and hence we were unable to perform further phenotypic and functional analysis. Information regarding the precise molecular contacts between the NTF2-like folds of Mex67-Mtr2 and pre40S subunits should aid the rational design of mutants of Mex67 and Mtr2 that are specifically impaired in pre40S export.
In the mRNA export pathway, co-transcriptional recruitment of Mex67-Mtr2 to the growing nascent pre-mRNA takes place during early steps of mRNP biogenesis [58], [59]. In the 60S pathway, Mex67-Mtr2 is recruited to late export competent pre60S subunits (Figure 10A, left panel) [14]. Here, we found that Mex67-Mtr2 maximally co-enriches with early 40S pre-ribosomes isolated via Enp1-TAP that purifies both the 90S and early pre40S subunits (Figure 10A, right panel). Thus pre40S subunits might be competent to load the Mex67-Mtr2, perhaps after separation of the 90S into 40S and 60S precursors. Biochemical analysis revealed that the loops of Mex67-Mtr2 play a role in stable incorporation of Slx9 into early pre40S subunits (Figure 11). Notably, strong genetic interactions were observed between Mex67-Mtr2 and Slx9, which also co-enriches mainly with early pre40S subunits (Enp1-TAP; Figure 1B). These data raise the possibility that recruitment of Mex67-Mtr2 via the loops to pre40S subunits might be required for maturation steps that lead to the formation of export competent pre40S subunits.
The C-terminal UBA-like domain of Mex67 interacts with FG-rich nucleoporins and directly contributes to the transport of bound cargos through the NPC [34], [40]–[42]. We found that the mex67ΔC1 allele did not rescue the pre40S subunit export defect of the slx9Δ mutant; instead, this allele exacerbated the pre40S export defect of slx9Δ cells (Figure 7B). Notably, growth of the mex67ΔC1 allele when combined with the slx9Δ, yrb2Δ, rps15-1, ltv1Δ, rps6aΔ and bud23Δ were either synthetic lethal or strongly synthetically enhanced (Figure 8). These data indicate that the FG-rich nucleoporin interacting UBA-like domain contributes to the function of Mex67 in pre40S subunit export. How Mex67-Mtr2 is released from pre40s subunits remains to be determined. Yet unknown energy consuming factors could trigger the release of Mex67-Mtr2 from pre40S subunits [60], [61].
What could be the reason for the dominant negative nature of the mex67Δloop, mtr2Δloop116-137 and mex67ΔC1 alleles in the slx9Δ mutant? A functional full-length Mex67-Mtr2 becomes limiting and essential for pre40S subunit nuclear export in the slx9Δ mutant (Figure 5 and Figure 7). Thus one plausible explanation could be that expression of the mex67Δloop, mtr2Δloop116-137 and mex67ΔC1 alleles poisons the WT Mex67-Mtr2 by forming either Mex67-Mtr2Δloop116-137 or Mex67Δloop-Mtr2 heterodimers, that cannot bind pre40S subunits (Figure 11). Alternatively, in the case of the mex67ΔC1 allele, a mex67ΔC1-Mtr2 heterodimer could be formed that cannot efficiently interact with FG-rich nucleoporins. Expression of mex67Δloop, mtr2Δloop116-137 and mex67ΔC1 alleles might therefore aggravate the pre40S subunit export defect in the slx9Δ mutant (not in SLX9) by creating non-functional Mex67-Mtr2 heterodimers that either do not bind pre40S subunits or inefficiently transport pre40S subunits through the NPC.
What could be the role of Slx9 in pre40S subunit maturation/export pathway? Pre40S subunits are exported out of the nucleus containing 20S rRNA precursor. Upon reaching the cytoplasm, the 20S rRNA is cleaved to mature 18S rRNA releasing the ITS1 fragment for degradation by the nuclease Xrn1 [47], [48]. In this study, we found a strong nuclear accumulation of pre40S subunits in the slx9Δ mutant as judged by the localisation of S2-GFP and the 5′ portion of ITS1 (Figure 2 and Figure 3A). The nucleoplasmic accumulation of ITS1 seen in the slx9Δ mutant is similar to that observed in the xpo1-1 strain that is impaired in nuclear export of pre40S subunits (Figure 3B). Because a pre40S subunit containing 20S pre-rRNA is exported into the cytoplasm, a blockage in subunit export is expected to result in increased levels of 20S rRNA [21]. This has been observed for the slx9Δ mutant [45]. These data suggest a requirement of Slx9 for efficient nuclear export of the pre40S subunits. Large-scale synthetic genetic array (SGA) screens from the Krogan and Boone laboratories revealed strong genetic interactions between Slx9 and several nucleoporins (Nup57, Nup120, Nup2, Nup53, Asm4, Nup42), and integral nuclear membrane proteins that are required for NPC biogenesis (Apq12, Pom34) [57], [62]. These results indicate that Slx9 is embedded in a network of functional interactions involving the NPC. However, Slx9 did not bind the export receptor Xpo1 in presence of RanGTP in vitro, suggesting that it does not contain a nuclear export signal and therefore is unlikely to be an export adapter (data not shown). Slx9 may be necessary for recruitment of an export adaptor on maturing pre40S subunits. However, Mex67-Mtr2 levels on Enp1-TAP were not altered in the slx9Δ mutant suggesting that Slx9 might facilitate incorporation of a yet unknown export factor onto pre40S subunits (Figure S1A). While the recruitment of Mex67-Mtr2 to pre40S subunits remained unaffected in the slx9Δ mutant, we found that efficient recruitment of Slx9 to pre40S subunits requires Mex67-Mtr2 loops (Figure 11). These data indicate that recruitment of Mex67-Mtr2 to pre40S subunits precedes stable incorporation of Slx9. Another possibility could be that Slx9 participates in a yet unknown maturation step that renders pre40S subunits export competent. In line with this possibility, Slx9 briefly visits early pre-ribosomal particles in the 40S maturation pathway (Figure 1B). The precise role of Slx9 in pre40S subunit maturation/export pathway remains to be determined.
Previously, a high-throughput visual screen reported that the slx9Δ mutant is impaired in nuclear export of poly-(A)+ RNAs [49]. However, in our hands, <1% of the slx9Δ cells showed nuclear accumulation of poly-(A)+ RNA in the temperature range between 20–37°C (Figure S2A). This discrepancy appears not to be due to differences in strain backgrounds, since the slx9Δ mutant constructed in three different backgrounds (SC228c, W303 and RS453) exhibited the same phenotype (data not shown). Notably, the mex67Δloop allele, which in combination with the nup85ΔN133 mutant showed an mRNA export defect (Table 1) [30], did not further exacerbate the nuclear poly-(A)+RNA accumulation of the slx9Δ mutant (Figure S6A). Only the pre40S export defect of the slx9Δ mutant was strongly aggravated upon the expression of the mex67Δloop allele (Figure 7B). Further, the nuclear accumulation of poly-(A)+ RNA observed in <1% of the slx9Δ cells was not aggravated when combined with the mtr2Δloop116-137 allele (Figure S6B). The synthetically enhanced slx9Δmtr2Δloop116-137 double mutant strain was strongly impaired only in the nuclear export of pre40S subunits (Figure 9). Finally, Slx9 does not genetically interact with factors (Sub2 and Yra1) that are specifically involved in mRNA export (Figure S2B). Altogether, these results do not support a function for Slx9 in the mRNA export pathway.
Ribosome production is one of most energy consuming processes that needs to be quickly repressed or induced in response to nutrient availability. An extensive regulatory crosstalk must exist between the mRNA and ribosome biogenesis/export pathways that ensure correct levels/stoichiometry of mRNAs and ribosomes reach the cytoplasm. Mex67-Mtr2 could co-ordinate nuclear export of pre-ribosomal particles and mRNAs. Moreover, given the vital need to rapidly transport mRNAs and pre-ribosomes to the cytoplasm, Mex67-Mtr2 could step in to perform the duty of pre40S subunit export under circumstances when either adaptors for Xpo1 or yet unknown karyopherin-like factors are not recruited to pre40S subunits. Mex67-Mtr2 could export some of the pre40S subunits albeit with less efficiency to keep a certain cytoplasmic pool of mature ribosomes, and thus translation, sustained in slx9Δ cells. How is the cellular pool of Mex67-Mtr2 fractionated to participate in the nuclear export of mRNAs, pre40S and pre60S ribosomal subunits? Unravelling the mechanism(s) by which the three transport pathways cooperate to ensure the arrival of appropriate levels of pre-ribosomal subunits and mRNAs in the cytoplasm is a challenge for the future.
The Saccharomyces cerevisiae strains used in this study are listed in Table S1. Genomic disruptions and C-terminal tagging at the genomic loci were performed as described previously [63]–[65]. Preparation of media, yeast transformations, mating, sporulation of diploids, tetrad analysis, genetic manipulations and the slx9Δ high-copy suppressor screen were performed according to established procedures.
Plasmids used in this study are listed in Table S2. Details of plasmid construction will be provided upon request. All recombinant DNA techniques were performed according to established procedures using E. coli XL1 blue cells for cloning and plasmid propagation. All cloned DNA fragments generated by PCR amplification were verified by sequencing.
Genetic interactions were tested as described previously [26]. (a) Example for a synthetic lethal interaction: the mex67Δslx9Δ strain containing pURA3-MEX67 was transformed with pairs of plasmids and grown on –LeuTrp (SD) plates (media that does not select for the pURA3-MEX67 plasmid): (1) pLEU2-MEX67/pTRP1-SLX9; (2) pLEU2-mex67Δloop/pTRP1-SLX9; (3) pLEU2-MEX67/pTRP1-empty; (4) pLEU2-mex67Δloop/pTRP1-empty. To score for a genetic interaction, the transformants were spotted on 5-FOA (SD). (b) Example for synthetic enhancement: the mtr2Δslx9Δ strain containing the pURA3-MTR2 was transformed with the following plasmids: (1) pLEU2-MTR2/pTRP1-SLX9; (2) pLEU2-mtr2Δloop116-137/pTRP1-SLX9; (3) pLEU2-MEX67/pTRP1-empty; (4) pLEU2-mtr2Δloop116-137/pTRP1-empty. The mtr2Δslx9Δ strain containing the pURA3-MTR2, containing the plasmids: pLEU2-mtr2Δloop116-137 and the pTRP1-empty, grew very slowly on 5-FOA (SD) plates (as compared to the single mtr2Δloop116-137 and slx9Δ mutants), indicating a synthetic enhancement. The strains that grew on 5-FOA (SD) plates were subsequently spotted on YPD plates at different temperatures for analysis.
Tandem affinity purification (TAP) of pre-ribosomal particles were carried out as described previously [51], [65]–[67]. All purifications were performed using the TAP lysis buffer (50 mM Tris-HCl, pH 7.5, 75 mM NaCl, 1.5 mM MgCl2, and 0.15% NP-40). Eluates of all TAP purifications were analysed by NuPAGE 4–12% Bis-Tris gel (Invitrogen) followed by silver staining or Western blotting.
The L25-GFP and S2-GFP reporter assays to analyse pre-ribosomal subunit export were performed as previously described [51], [67]. Cells were observed by fluorescence microscopy (see below). Percentages of cells exhibiting nuclear export defects (mRNA, pre40S and pre60S) reported in this study were averaged from three independently performed experiments. >200 cells were analysed for each strain indicated.
Fluorescence in situ hybridization (FISH) to monitor nuclear accumulation of poly-(A)+RNA in different strains was carried out using Cy3-oligo (dT)30 as previously described [68]. Nucleoplasmic accumulation of 20S rRNA in the different strains was carried using a Cy3-labeled oligonucleotide probe (5′-Cy3-ATG CTC TTG CCA AAA CAA AAA AAT CCA TTT TCA AAA TTA TTA AAT TTC TT-3′) that is complementary to the 5′ portion of ITS1 as previously described [69].
Sucrose gradient sedimentation to determine the subunit stoichiometry was performed as described previously [50], [51]. The indicated strains were grown to OD600 0.8 and lysed in lysis buffer (50 mM Tris-HCl, pH 7.5, 50 mM KCl, and 1 mM DTT). The lysate was clarified by centrifugation and loaded onto 7–50% sucrose density gradient containing 50 mM Tris-HCl, pH 7.5, 50 mM KCl, and 1 mM DTT. The gradient was centrifuged at 39,000 rpm for 165 min (SW41 rotor; Beckman Coulter). Run-off polysome profiles were recorded by measuring rRNA at A254 using a density gradient fractionator (Teledyne).
Cells were visualized using DM6000B microscope (Leica) equipped with HCX PL Fluotar 63× and 100× 1.25 NA oil immersion objective (Leica). Images were acquired with a fitted digital camera (ORCA-ER; Hamamatsu Photonics) and Openlab software (PerkinElmer).
Western blot analysis was performed according to standard protocols. The following primary antibodies were used in this study: α-Slx9 (1∶1000; this study), α-Mex67 (1∶5,000; C. Dargemont, Institut Jacques Monod, France), α-Mtr2 (1∶1000; E. Hurt, University of Heidelberg, Germany), α-Nmd3 (1∶5,000, A. Johnson; University of Texas-Austin, USA), α-S3 (1∶2,000; Proteintech), α-S5 (1∶4,000; this study), α-L3 (1∶10,000; J. Warner, Albert Einstein College of Medicine, USA), α-L35 (1∶4,000, this study), and α-TAP (1∶4,000; Thermo Scientific). The secondary HRP-conjugated α-rabbit and α-mouse antibodies (Sigma-Aldrich) were used at 1∶1,000 to 1∶5,000 dilutions. Proteins were visualized using Immun-Star HRP chemiluminescence kit (Bio-Rad). Whole cell lysates were prepared by alkaline lysis of yeast cells as previously described [51].
|
10.1371/journal.pgen.1006178 | High Resolution Genomic Scans Reveal Genetic Architecture Controlling Alcohol Preference in Bidirectionally Selected Rat Model | Investigations on the influence of nature vs. nurture on Alcoholism (Alcohol Use Disorder) in human have yet to provide a clear view on potential genomic etiologies. To address this issue, we sequenced a replicated animal model system bidirectionally-selected for alcohol preference (AP). This model is uniquely suited to map genetic effects with high reproducibility, and resolution. The origin of the rat lines (an 8-way cross) resulted in small haplotype blocks (HB) with a corresponding high level of resolution. We sequenced DNAs from 40 samples (10 per line of each replicate) to determine allele frequencies and HB. We achieved ~46X coverage per line and replicate. Excessive differentiation in the genomic architecture between lines, across replicates, termed signatures of selection (SS), were classified according to gene and region. We identified SS in 930 genes associated with AP. The majority (50%) of the SS were confined to single gene regions, the greatest numbers of which were in promoters (284) and intronic regions (169) with the least in exon's (4), suggesting that differences in AP were primarily due to alterations in regulatory regions. We confirmed previously identified genes and found many new genes associated with AP. Of those newly identified genes, several demonstrated neuronal function involved in synaptic memory and reward behavior, e.g. ion channels (Kcnf1, Kcnn3, Scn5a), excitatory receptors (Grin2a, Gria3, Grip1), neurotransmitters (Pomc), and synapses (Snap29). This study not only reveals the polygenic architecture of AP, but also emphasizes the importance of regulatory elements, consistent with other complex traits.
| Alcohol Used Disorder (AUD) or Alcoholism extracts a great societal cost in terms of human suffering. Understanding the genetic basis is critical to comprehend, treat and prevent this disease, but difficult in humans, as choice is influenced by nature and nurture. To discover its genetic basis, we used an animal model system that controlled for genetic and non-genetic factors through randomization, study replication, long-term divergent selection, and a controlled environment. We conducted whole genome sequencing in breeds that were either compulsive excessive drinkers or completely abstinent. We discovered consistent alterations in several genes and neurological pathways previously unassociated with alcoholism. These results strengthened our understanding of the genetic basis of alcoholism and revealed potential genetic- and neurological-based treatments.
| The quest to discover the underlying genetic etiology that contributes to alcoholism (a.k.a. Alcohol Use Disorder, AUD) is an ultimate goal for understanding, preventing and treating alcoholism caused by inherited risk factors. Genomic investigations in humans have yet to determine the genetic causes of this disease due to a number of challenges, including partial or complete confounding of family history with familial drinking behaviors, drinking variability and non-genetic (e.g. social, economical and cultural) factors. Genomic analyses in humans, including genome-wide association studies (GWAS), account for a small proportion of the total genetic variance associated with AUD [1–4]. In addition, reproducibility of results has become a major concern [5]. To address these issues, an animal model with heterogeneous origins [6], combined with replication and randomization (see below), was utilized with the expectation that the model will gain insights into the human condition.
The high (HAD) and low (LAD) alcohol drinking rat lines constitute one of the most extensively characterized rodent models of human AUD [7]. These bi-directionally selected (>40 generations) and replicated lines [8] exhibit a wide range of alcohol preferences (AP) and other alcohol-related endophenotypes that recapitulate human AUD [7, 9]. The HAD/LAD lines were derived from the NIH heterogeneous stock (NIH-HS) [10] and encompass genetic variation from eight inbred lines. The HAD/LAD demonstrate a major heritable component for AP, providing an attractive model to expose the genetic and neurobiological bases underlying AUDs.
The AP trait is characterized by heightened alcohol seeking behavior driven by the reinforcing action and other effects of ethanol on the brain. HAD rats show the reinforcing effects through performing an operant response for access to alcohol and attain pharmacologically relevant blood alcohol concentrations (BACs; 50 to 150 mg/dL) in under 24 hours [7]. HAD animals display an alcohol deprivation effect after repeated cycles of access to a single concentration of alcohol and an increased response when multiple concentrations of ethanol are used. In contrast, the LAD lines exhibit none of these phenotypes [7]. Although previous studies from our group identified several candidate genes associated with AP on Chrs 5, 10, 12 and 16 in HAD rats [11–13], the QTL resolving power precluded fine mapping of the causative genes.
In this study we employed whole genome sequencing to identify regions of excessive differentiation associated with AP using the population stratification index (Fst). The Fst identifies genes and gene regions under selection by finding genomic regions with an excessive degree of differentiation [14–17]. While this approach has been tried in a number of studies [18–20], most were inconclusive due to issues related to demography, lack of replication [21] or large LD blocks. We eliminated most of these issues by using a heterogeneous, bi-directionally-selected and replicated animal model that enabled us to detect genomic signatures of selection (SS) for AP in the HAD/LAD lines including HAD1/LAD1 lines and replicate HAD2/LAD2 lines. We identified specific genes and in many cases regions within those genes under selection, allowing insight into the importance of regulatory vs. coding regions in the evolution of phenotypes associated with AP. Our findings show SS in genes previously predicted in the human population and identify new genes that are functionally relevant to the plasticity of the reward mechanism indicative of AP.
Following whole genome sequencing, variants (SNPs) were called using GATK as described in Methods-Sequence assembly and variant calls. There were 6,585,606 SNPs identified when combined across replicates. Allele frequencies and Fst statistics were determined from these SNPs. While sequencing errors were filtered, any remaining are expected to have only minor effects on the metric we used for finding SS (see Discussion-Sequencing Errors, Window Size, LD and Haplotype Blocks). From these SNPs, we found that the LD_25 (the genomic distance for which the R2 < .25, see Methods-Detecting Signatures of selection (SS)), combined across replicates, was approximately 1.4kb and included 7 SNPs on average. Consequently, the sliding window used to calculate the running average of the Fst statistic was based on 7 adjacent SNPs, 3 on either side of the central or focal position. Remarkably, from the cumulative frequency distribution of R2 and for pairs of loci 7 SNPs apart (Fig 1A), approximately 55.6% had an R2 less than .25 and 81.4% had an R2 less than 0.50, indicating that the majority of variation between loci at this distance is independent of linkage.
Similarly, the average haplotype block (HB) length was small (Fig 1B). We defined HB using the Four Gamete Rule (FGR) [23] (see Methods-Genotypes, Allele frequencies, Linkage Disequilibrium, and Haplotype Blocks). For comparison with the base NIH-HS-S population, we used a restriction that the least frequent gamete (Minor Gamete Frequency, MGF) be > 6.25% and resulted in a median HB size of 870bp in the HAD/LAD lines. Using the same MGF, the median HB size for the NIH-HS-S base population, was 1730bp (Fig 1B), showing that initial HB length was small and 60+ generations of random mating reduced the interval by ~50%. However, for identification of SS associated with gene regions, we used a more conservative MGF = 9%. For this value of MGF, we observed a near continuous spine of HB (Haploview), and the median HB length was 1.48kb, slightly larger than the LD_25 based on R2.
The genome wide average Fst estimated within or across replicates is given in Table 1. These statistics estimate the overall level of genetic differentiation between and within replicates, which in the absence of selection would be due to genetic drift. Signatures of selection (SS) were identified based on the intra-class correlation (θ) method of estimating Fst and calculated from a running average over 7 adjacent loci, the Genome Wide significance (GWp) of which was determined from 100,000 genome wide permutations (see Methods-Detecting Signatures of selection (SS)). There were 930 genes with θ values where GWp < .00001 (shown in S1 Table). A Manhattan plot of θ values with critical cut off and top-scored genes is illustrated in Fig 2.
Without accounting for size, the occurrence of SS in the defined gene regions (Methods-Annotation of Data by Gene and Region, illustrated in Fig 3), indicates that the greatest number of SS occurred in promoters and intronic regions, which were 7 and 15 times more abundant than those in exons, respectively. When adjusted for the size of the gene region (Fig 3), the greatest densities (SS/Mb) were observed in the intron-exon junctions, UTRs and promoters.
Among the SS shown in Fig 4, the majority (50%) were located in only one region of a gene, indicating that selection operated on units within the gene (e.g. promoter, intron-exon junction), as expected based on alternative modes of gene function and expression, see also examples illustrated in Fig 5. Of the SS that mapped uniquely to a single region, the greatest numbers were in the promoters (284) and intronic regions (169) with the least in exons (4) and none in UTRs. For genes in which selection operated on more than one region, correlations between numbers of SS in those regions are given in Table 2. As expected, the correlations were generally small, except between the UTR and exon, which showed a strong positive correlation. In contrast and unexpected, a strong negative correlation was observed between number of SS in the promoter and intron regions, suggesting that selection of polymorphisms in these regions was antagonistic.
We identified 127 exonic SS that contained missense/non-synonymous SNPs (see SIFT analysis in Methods-Putative Functional Impacts of SS) in 85 genes shown in S2 Table, including Cyp2ab1 (cytochrome P450, family 2, subfamily ab, polypeptide 1), Scn5a (Sodium Channel, Voltage Gated, Type V Alpha Subunit), Ccdc13 (coiled-coil domain containing 13). Some SS in promoters (S1 Table) overlapped putative transcription factor binding (TFB) sites. Examples for neuronal genes (See TFB analysis in Methods-Putative Functional Impacts of SS) are shown in S3 Table. In these neural-related genes, 8 out of 22 SS were associated with putative TFB and occurred at CpG dinucleotides, indicating that DNA methylation could contribute to selection. These include Kcnf1 (Potassium voltage-gated channel; subfamily F; member 1), Syt1 (Synaptotagmin I), and Cck (Cholecystokinin) (S3 Table). We detected 1575 SS in intron-exon junction regions. Ten SS are predicted to alter loci that potentially contain elements for RNA-binding proteins (RBPs; S4 Table). These include Scn5a (Sodium channel; voltage-gated; type V; alpha subunit), Syt1 (Synaptotagmin I), and Ehmt2 (Euchromatic histone lysine N-methyltransferase 2).
Among the 930 genes identified, several are involved in pathways known to be important for the regulation of alcohol consumption based on prior functional analysis and genes detected in our previous QTL mapping studies of the same animal model [11–13]. One of the most intriguing pathways is the Glutamate Receptor Signaling Pathway, where changes in glutamate transmission in the nucleus accumbens and amygdala have been implicated in AUD [26, 27]. Not only does this pathway contain the Grin2a gene (NMDA receptor subunit) in the chr10 QTL region reported by Carr et al. [11] in the HAD1/LAD1 lines, but also we identified five additional genes in this pathway with SS: Gria3 (AMPA receptor subunit), Grm1, Grm8 (metabotropic glutamate receptor), Grip1 (a glutamate receptor interaction protein), and Slc17a8 (a vesicular glutamate transporter). The proteins encoded by these genes are involved in the efficiency of excitatory communication and synaptic memory (Grin2a, Gria3, Slc17a8), modulation of glutamate reception (Grm1 and Grm8) and excitatory synaptic protein trafficking (Grip1). Their collaborative function in the pathway is illustrated in Fig 6 using IPA analysis.
Besides the glutamate system, we also found SS in a number of genes related to other transmitter systems that have been implicated or have the potential to be involved in AUD, e.g. peptidergic transmitters (e.g. Cck, Pomc), catecholamine metabolic enzyme (e.g. Comt), serotonin receptors (e.g. Htr4) and acetylcholine receptors (Chrna10). Another interesting finding is that a number of genes associated with ion channels were detected. These genes serve important functions in electrical activity mediating neuronal transmission, sensitivity, and effectiveness, including Na+ channels (e.g. Scn5a, Scn10a), K+ channels (e.g. Kcnn3, Kcnf1, Kcnc2), the cation channel (Trpm8), and calcium-activated chloride channels (Ano10). At the synaptic level, the genes encoding synaptic proteins e.g. Shank2, Snap29, Syt1, Syt11 that mediate neuronal plasticity were also found.
We identified SS in 29 olfactory receptor genes (e.g. Olr37, Olr41, Olr72) that control complex olfaction as well as 24 solute carrier genes (e.g. Slc17a5, Slc22a14, Slc35d1) that transport essential molecules across cellular membranes, including neurotransmitters, ions, and sugar molecules. Other genes relevant to gene regulation include epigenetic modulators (e.g. Hira, Ehmt2, Yeats2), miRNAs (e.g. Mir138-1, Mir760, Mir301b) and 16 zinc finger transcription factor genes (e.g. Zfp105, Zbtb8a, Zbtb8b, Zdhhc17).
The HAD/LAD bi-directionally selected lines are uniquely suited for fine mapping for a number of reasons: a) They are one of the few replicated animal models that were initiated from the same resource population in the same generation, allowing the ability to separate signal (fixation due to selection) from noise (fixation due to random drift); b) They result from a MAGIC cross that resulted in small HB that pinpoints SS to discrete portions of a gene; c) They are derived from a long-term, bi-directionally selected population, that magnified the phenotypic and genomic differences between lines for increased power; and d) Within family selection was used to develop the lines, which maximized the effective population size and minimized the rate of inbreeding because all families contributed at least one individual to the next generation [28]. Maximizing the effective population size also minimizes the buildup of new HB [29].
Finally, while use of the Fst as a metric to detect SS has a number of limitations in natural populations [19, 20] due to demography, expanding populations, migration, random genetic drift, and time frame (ecological vs. evolutionary), in our experimental situation all of these variables were controlled. With the low mutation rates associated with SNPs, the Fst statistic is a good metric for enrichment of genes under selection, as proposed by Akey et al. [30]. However, the population specific Fst of (36) offers an alternative method of finding SS that may have more power than the average Fst used in our analysis and needs to be explored further.
Fixation of alleles can result from the combined effects of random genetic drift and selection for AP. Since alleles lost due to random genetic drift are proportional to the genome wide average Fst [31] (the values for which are given in Table 1), it is possible to obtain a crude estimate of the proportion of SNPs lost due to random genetic drift and also the number fixed due to selection. The model for estimating the number of SNPs fixed due to selection and lost due to drift is given by the following:
Lost=b0+b1Fst+e
where
Lost = the total number of alleles fixed or lost,
b0 = the intercept which is the number of alleles fixed due to selection
b1 = rate of loss of alleles due to random genetic drift
Fst = Wright's between population fixation index
e = residual error
In the absence of selection and drift (Fst = 0), the intercept (b0) is expected to be 0 because allele frequencies would remain polymorphic and constant [31]. The intercept therefore estimates the number of SNPs fixed due to selection. The least squares fit of the data to the model gives b0 = 255,278 ± 94,000. By comparing this number (b0) to the total number fixed or lost (N) within and across replicates (Table 1), one can estimate the false discovery rate [FDR = (N- b0)/N]. The FDR here is defined as the proportion of SNPs fixed or lost between lines that might be attributed to selection. The FDR estimated within replicates ranged from 57% (N = 592,328) and 63% (N = 688,080), while the FDR = 39% (N = 418,177) across replicates. Thus within replicates, random genetic drift accounts for the majority (>50%) of SNPs fixed and lost. When averaged over replicates, the reverse was true, the majority of fixed or lost alleles were due to selection. These rough approximations assume all loci exhibited the same initial allele frequency and were going to fixation or loss at the same rate. Nevertheless, these results illustrate the importance of replication for detecting loci under selection and minimization of false positives [21].
These results also directly address the concerns raised by Collins and Tabak [5] for lack of reproducability. Without replication, the majority of allele fixations were due to random genetic drift. This problem cannot be fixed statistically by increasing stringency for testing. For example, when analyzed by replicate, and GWp <.001, 3,698 genes were detected in Replicate-1 and 3876 genes were detected in Replicate-2, yet only 1,060 genes were verified across replicates at this GWp. Only ~29% of genes identified by a single replicate were reproducible across replicates (71% failure rate). Further, increasing the stringency by decreasing GWp <.0001, resulted in 3,100 genes detected in Replicate-1 and 3,041 genes detected in Replicate-2; yet only 761 verified across replicates at this Gwp, i.e. only ~24% were reproducible, which is less than that at the lower stringency GWp. In the extreme, if only the top 1000 SS in the first and second replicates were chosen, these were associated with, respectively, 177 and 184 genes, but only 9 of these genes were verified across repliactes, which translates to a >95% failure rate. The lack of reproducibility between replicates was due to the confouncing effects of random genetic drift, not to incorrectly setting the critical Type 1 error rate.
Although lack of replication increases Type 1 error rate because the error term without replication is the mean square for variation among individuals within line, which is not appropriate as it does not include drift variance. The correct error term is the replicate x line interaction [32]. Similarly, Weir and Cockerham cautioned that even if "we were to census the entire population, our results are still affected by 'sampling' in that the particular population sampled is but one of the many possible replicates that could have arisen under the same conditions". However, even if the error variance were corrected to include drift variance, this would not correct for lack of replication. Because genetic drift is random, it is expected that a high proportion of alleles with neutral effects will drift in opposite directions [31] and it is likely that some SS could be significant within each replicate, but fixed in opposite directions in different replicates. With replication, such SS are averaged over before the Fst values are calculated and the SS would disappear. Thus without replication these types of SS would largely give non-reproducible results even at high stringencies, as seen here.
The neuroimmunoendocrine actions of alcohol in the gut-liver-brain network are increasing in importance as a potential mechanism underlying AUD [48–51]. Although a few of these genes have been investigated for their contribution to AUD, the scope and the major players are largely unknown. We identified SS in several genes and pathways involved in acquired immunity, including six Major Histocompatibility Class II (MHC II) genes (e.g. Hla-doa, Hla-dob, Hla-dqa1, Hla-dqb1, Hla-drb5, H2-eb2). The gene NF-kappaB signaling gene (Nfkb1) is involved in a wide range of immune and stress response pathways. We also identified genes with SS in the Corticotropin Releasing Hormone Signaling pathway which, taken together, support and provide new candidates in stress-peptide related AUD [52–54].
Excitatory and inhibitory neuronal communication pathways are directly involved in neuronal plasticity and brain adaptation, and are key to developing AUD [55, 56]. Among these, the Glutamate Receptor Signaling Pathway is pivotal not only in glutamate transmission, but is also critical in long term potentiation (LTP) and synaptic plasticity. It has been implicated in alcohol drinking or addictive behaviors [56–58]. Among the 6 genes in the pathway, the NMDA receptor subunit gene Grin2a has been associated with susceptibility of drug abuse [59, 60]. Grin2a contains 88 SS within the intronic regions. In a primate study [57], alternative splicing of AMPA receptor subunits Gria3 in the prefrontal cortex was found following chronic alcohol self-administration [57]. Additionally, expression levels of GRIA3 flip and flop mRNAs were positively correlated with daily ethanol intake and blood ethanol concentrations [58]. These flip and flop variants alter the kinetics of this glutamate channel [57, 58]. Gria3 contained 37 SS within the intronic regions. Possible effects of SNPs within the intronic regions include: alternative splicing, differential binding of transcription factors to intronic regulatory regions and epigenetic gene regulation of the AMPA gene subunit. The mGluR1 (Grm1) receptor has been shown to function in the brain reward system in the nucleus accumbens (NAc) regulation of alcohol intake [61]. In addition, we found 7 SS in the glutamate receptor interacting protein 1 gene (Grip1). Grip1 mediates LTP by strategically positioning the AMPA receptor towards the dendritic spine postsynaptic position. Identification of the Grip1 receptor as a putative gene in AP is unique and has not been previously studied in this context.
Nervous system ion channels directly influence the electrical activity and efficiency of neuronal transmission. Among the potassium channel genes identified, Kcnn3 encodes the KCa2.3 channel, which controls neuronal excitability and synaptic plasticity. This gene has been implicated in alcohol, nicotine, cocaine, and heroin abuse [62, 63]. In a mouse study [63] using an interspecific cross between C57BL/6 and DBA/2 mice with diverse drinking patterns, Kcnn3 activity negatively modulated voluntary and excessive alcohol consumption. Low transcription levels of Kcnn3 in the NAc were associated with higher alcohol intake. Inhibition of the KCa2 channels in the NAc of these mice may also increase alcohol consumption [63]. Our results support Kcnn3 as a gene involved in AP.
Synaptic plasticity underlies brain adaptation and, has been shown to be disrupted in AUD [56]. We found intronic SS in the synaptic gene, Syt1 encoding synaptotagmin, a protein which regulates presynaptic vesicular release. DNA methylation of the related gene Syt2 in the medial prefrontal cortex (mPFC) has been linked to escalation of alcohol drinking [64] while inhibition of Syt2 expression in the mPFC increased aversion-resistant alcohol drinking [64]. Another synaptic protein gene found in that group, Shank2, is a member of the Shank family that functions as a molecular scaffold in the postsynaptic density (PSD). Shank2 attaches metabotropic glutamate receptors (mGluRs) to an existing pool of NMDA receptors critical in modulating excitatory transmission. These genes encode proteins that interact with the Glutamate Receptor Signaling Pathways discussed earlier and can form a sophisticated network with the potential to address AP behavior.
We also found SS in the Cck gene (Cholecystokinin), which is a peptide hormone that stimulates digestion of fat and protein in the gastrointestinal system. This hunger suppressant has been shown to induce tolerance to opioids and plays a role in withdrawal. Chronic alcohol intake has been associated with increased sensitivity to CCK8 [65].
Signatures of selection were found across all regions of the Catechol-O-methyltransferase gene (Comt), which encodes an enzyme involved in the metabolism of dopamine, adrenaline and noradrenaline. Elevated expression of COMT has been associated with AUD in a male Czechoslovakian population, which is linked to a Val158Met amino acid substitution in COMT [66]. In a Finnish population, the high (H) and low (L) activity COMT allele frequencies were compared and the L frequency was found to be significantly higher among alcoholics when compared with controls [67]. These results indicate that the COMT polymorphism contributes significantly to the development of late-onset AUD.
The neurotensin receptor 2 gene (Ntsr2) has two SS in intron 2 and has been implicated in alcohol dependence with conduct disorder and suicide attempts in humans [68]. In Dick et al’s study, two significant SNPs in intron1 of Ntsr2, rs12612207 and rs4669765, showed evidence for association with alcohol dependence [68]. Although the direct biological function of NTSR2 in alcohol dependence is still unclear, previous studies reported the potential function of the neurotensin-containing pathway in linking the hippocampal and mesolimbic dopamine systems in response to drug addiction [68].
A number of genes that have been associated with human AUD, e.g. GABA-A receptor GABRA2 [69], were not found in our analysis, although some suggestive differences were present in one HAD/LAD replicate.
In conclusion, this study revealed that AP exhibits a polygenic architecture, consistent with a complex trait and indicates that efforts to concentrate on single genes, or a few genes with large effects, are likely to result in only a small fraction of the genetic variation in the trait [2, 69, 70].
Two replicate HAD and LAD lines were initiated from the N/NIH HS heterogeneous stock [71] (i.e., with different parents in two different colonies) and selectively bred using within family selection and a rotational breeding program [10]. During selection both lines were given free access to food, water and a 10% (v/v) ethanol solution. The selection criteria for the HAD lines were consumption of at least 5.0 g of ethanol/kg body weight/day, with an ethanol to water ratio of at least 2:1, while LAD rats were required to drink less than 1.5 g/kg/day with an ethanol to water ratio of less than 0.5:1. Bi-directional selection for AP was repeated continuously for 30 generations, followed by generations of selection that were interspersed with relaxed selection. Sixty generations were completed at the time of sampling, of which 40 were for and against AP in each replicate.
Spleen samples from 10 alcohol naïve rats (5 males and 5 females) of each line and replicate were collected and stored at -80°C. DNA was isolated using Gentra Puregene Kits (Qiagen, Redwood City, CA, USA). About 25 mg of spleen was homogenized in cell lysis solution, followed by proteinase K digestion, RNase A incubation, and DNA precipitation as described in manufacture’s protocol. DNA purity (260/280 ratio) measured 1.79–1.94, with the majority >1.85. The DNA yield was more than 200 μg in most samples. One of the samples was removed from the analysis as genetic testing indicated its ID was incorrectly recorded.
Individual DNA from each rat was sequenced in the IU Center for Medical Genomics using an ABI SOLiD 5500xl platform using a 35/75 bp paired end fragments protocol in the first replicate and 75bp single end protocol in the second. We reached an approximate depth of 5x per sample and 50x across each line within a replicate. Reads were mapped to the reference genome (Rattus_norvegicus.Rnor_5.0.71) using LifeScope (http://www.lifescopecloud.com). Duplicate reads were removed using Picard (http://broadinstitute.github.io/picard/), and reads with mapping quality < 8 were also removed. Variants (SNPs) were called using GATK multi-sample variant calling [8]. Reads with coverage greater than 4 times the expected read depth across all samples were removed as possible copy number variants or other genome duplication events. A 0.3% sequencing error rate was assumed [72] and screened for by combining all samples and if the total number of SNPs at that location was less than 0.3% of total reads, the location was removed.
Allele frequencies in each line and replicate at a given locus were estimated as the total number of non-reference (NR) alleles present across samples within a replicate to the total reads at that locus (N). For estimating LD and HB, genotypes were called for each individual at each polymorphic locus. Gametic phase disequilibrium (or Linkage disequilibrium, LD) was estimated pair wise between all polymorphic loci on the same chromosome as the coefficient of determination (R2) using the iterative method [33]. Only loci with a minor allele frequency (MAF) > 10% were used to calculate LD to avoid issues related to rare alleles. In addition, following the methods of Wang and associates [73], haplotype blocks (HB) were defined as chromosomal regions flanked by at least one historical recombination event, termed the Four Gamete Rule (FGR) [23], and as implemented by Haploview [43]. In brief, blocks [25] are searched from the start of a region by sequential addition of the next locus, the number of unique gametes between the initial and last locus are counted. When all four gametes are observed between the last locus and any of previous loci in the block, the position of the last locus is regarded as the putative starting point of a new block, and the block size is determined as the sequence length between the start and end positions. The only explanation for observing all four gametes between a pair of loci is the occurrence of at least one historical recombination event [23]. While strict use of the FGR would require only one recombination event to infer an HB, for purposes of examination of HB as units of selection, a more conservative Minor Gamete Frequency (MGF) was used. Increasing the MGF results in larger HB. The MGF was therefore increased until a near continuous spine of HB resulted as defined and visualized by Haploview. A continuous spine results when nearly all SNPs are contained in HB and a new HB starts where the previous ends. This process of finding a continuous spine will tend to only find common HB, which for our purposes is the desired outcome. However, for purposes of comparison of HB size to the NIH-HS-S (see Methods-Estimation of allele frequencies in the NIH-HS parent lines) we used a MGF of 6.25% (1/16) because allele frequencies in that population were coarse with the smallest expected non-zero MGF = 1/2N for N = 8 lines.
The initial allele frequency was estimated from the parent lines that composed the NIH-HS. The 8 parent lines of the NIH-HS i.e., ACI, BN, BUF, F344, M520, MR, WKY and WN, were recently sequenced (Rat Genome Sequencing and Mapping Consortium, 2013). We downloaded the genomic data (ERP001923) from the NCBI Sequence Read Archive (SRA) and mapped the sequence to the reference genome used in this study (Rattus_norvegicus.Rnor_5.0.71). Our results were essentially the same of that of the Consortium, except we detected that some SNP loci were still segregating within some lines, although the differences were minor. Allele frequency was estimated within each line and locus as the number of SNP reads at that locus over the read depth at that locus. These were then averaged over the 8 parent lines. We refer to the "in silico" NIH-HS as the NIH-HS-S line.
Signatures of selection (SS) were detected as excessive differentiation measured by Wright's Fst statistic [23], but estimated as the intra-class correlation due to between line differentiation (θ), based on a simple extension of the method presented by Weir and Cockerham [15]. Weir and Cockerham (1984) showed that it is possible to define 'F'-statistics based on the method of moments from the analysis of variance. In a nested analysis, variance components can be estimated for 1) between population, 2) within population between individuals, and 3) between alleles within individuals, and 4) total variation. Based on various ratios of these they defined: F, f, and θ which correspond to Wright's Fit, Fis, and Fst. Because Fst, estimated as θ, was defined as a ratio of the between population variance component to the total variance, we refer to this statistic as an intraclass correlation as defined by Fisher [74]. The Fst statistic is usually found in running windows of arbitrary size [17]. We chose a window size based on the coefficient of determination (R2) between loci as suggested by Weir and colleagues [34]. The quantity (1- R2) is the proportion of the covariation between loci that is independent of linkage. With an R2 < 50%, the majority of covariation is independent. Thus at a minimum, the distance between loci should be greater than R2 = 50%. We chose a more conservative critical value of R2 = 25% (LD_25). The median number of SNPs between loci associated with the LD_25 was used as the size of the sliding window for the Fst statistic. Details of the method are as follows
Following Weir and Cockerham [15], the Fst statistic was estimated based on the following linear model where Y is the presence or absence of the SNP at the qth locus in the mth gamete
Yijkm=μ+Li+R(i)j+S(ij)k+G(ijk)m
(1)
for the ith Line (i = 1,2), of the jth Replicate (j = 1,2), in the kth Sample (k = 1,10), and mth gamete (m = 1,2) The expected mean squares for this model, in terms of drift variance, was given by Muir [32]. From the analysis of variance, variance components for each term were estimated for each locus containing a polymorphic SNP. The Fst statistic was estimated as θ in sliding windows of size w following Equation 10 from Weir and Cockerham [15] for multiple loci as
θ(q)=∑q=1,wσL2q∑q=1,w(σLq2+σR(L)q2+σS(RL)q2+σG(RLI)q2)
Statistical Significance was determined by genome wide permutations [75] whereby the design classifications were randomized onto gametes, following which the statistic was evaluated at all loci across the genome and the largest value recorded. This process was repeated 100,000 times. Among the ordered 100,000 largest values, the genome wide critical θ value (GWp) corresponded to the p% largest ordered θ value. Because all SNP locations across the genome were included in the permutation analysis, the critical value is automatically adjusted for multiple comparisons. As an additional level of caution, permutations were also conducted at the individual level, where the design classifications were randomized onto individuals, with their complete set of genotypes across loci. This method of permutation preserves the covariance structure among the gametes within individuals. These were also permutated 100,000 times and the GWp estimated in the same manner. The larger critical value from the two permutation methods was used as the GWp.
To give a global perspective of the genome-wide magnitude and spatial distribution of SS, we created a Manhattan plot based on the ' θ ' values on Chrs 1–20 and the X. The plot was generated using MATLAB 8.3 (The MathWorks, Inc., Natick, MA, USA).
Significant ' θ ' values in windows of running averages were annotated by gene regions. Gene locations and regions were based on the current Ensemble annotation of the rat genome (http://useast.ensembl.org/info/genome/index.html) version 5.0 and included: exon, untranslated region (UTR) and intron. The promoter region was defined as a 10kb domain upstream of the transcriptional start site in order to capture the proximal promoter, closely-linked enhancer elements and regions subject to epigenetic regulation.
Critical cis-acting splicing elements within the intron include the 5’ splice site, the branch point, the polypyrimidine tract and the 3’ splice site were included in the intron-exon junction fragment. Although the branch point is most commonly found within the last 40 nucleotides (nt) of the intron, its position is highly variable and has been mapped up to 400 nt away from the 3’ splice site [76, 77]. Analysis of several hundred human exons indicates that the majority of branch points are located within 150 bp of the 3’ splice site [78]. To ensure we captured the majority of branch point sequences, we divided introns into two components: a) intron-exon junction regions (JR) 150bp upstream and downstream of each exon for multi-exon genes and b) intron-non junction (INJR) regions, which capture intronic sequences that extend beyond 150 bp from the exon. The JR include some, but not necessarily all alternative splice sites, while the INJR may pick up additional alternative spice sites and other non-coding regulatory factors, such as enhancers and non-coding RNA binding elements.
Effects of SS in the exonal region were determined by the Ensembl Variant Effect Predictor (VEP): http://www.ensembl.org/info/docs/tools/vep/index.html [79], which uses the Sorting Intolerant From Tolerant (SIFT) software (http://sift.jcvi.org) [80] to predict the effect of SNPs in protein function [81]. The current version of Ensembl VEP uses rat genome Rnor_6.0 assembly, we therefore convert our data from Rnor_5.0 to Rnor_6.0 using Ensembl Assembly Converter (http://useast.ensembl.org/common/Tools/AssemblyConverter) before uploading data to VEP. To predict the effects of promoter SS on altering putative transcription factor binding, we retrieved 20 bp of flanking sequence (10bp on each side of the SS; 21bp total) from the RGSC Rnor_5.0 rat genome assembly, and put the sequence into PROMO (http://alggen.lsi.upc.es/) [82]. The HAD and LAD sequences were analyzed separately for putative transcription factor binding sites and additions or loss of CpG dinucleotide. For the prediction of RNA binding factors on junctional SS, similar to promoter SS, 10bp of flanking RNA sequence (converted T to U) on each side of the junctional SS were input to http://cisbp-rna.ccbr.utoronto.ca/TFTools.php [83, 84], using the model of PWMs-Log Odds, and threshold of 6. For genes expressed from the Watson strand, we used the published sequence; for genes expressed on the Crick strand, we used the reverse complement. The HAD and LAD sequences were analyzed separately for putative RNA factor binding sites.
Pathways of biological interest were defined and visualized using QIAGEN’s Ingenuity Pathway Analysis (IPA; Qiagen, Redwood City, CA, USA, www.qiagen.com/ingenuity). We did not do an enrichment or over-representation analysis common in RNA-seq studies because, as opposed to an RNA based expression analysis where all pathway members have the opportunity to be expressed, at the DNA level not all genes in a pathway may have had the opportunity to be selected upon due to lack of new mutation within those genes, or intolerance of those genes to new mutations, i.e. pathway analysis would fail to detect pathways under selection if selection did not have an opportunity to function because no new mutations were present to act upon. As such, pathway analysis is expected to exhibit a large number of false negatives and should not be used to prioritize pathways found via number of genes with SS. But, it is also reasonable to assume that multiple genes from a crucial pathway could be under selection and contribute to collaborative function.
|
10.1371/journal.pcbi.1006431 | Multiscale analysis of autotroph-heterotroph interactions in a high-temperature microbial community | Interactions among microbial community members can lead to emergent properties, such as enhanced productivity, stability, and robustness. Iron-oxide mats in acidic (pH 2–4), high-temperature (> 65 °C) springs of Yellowstone National Park contain relatively simple microbial communities and are well-characterized geochemically. Consequently, these communities are excellent model systems for studying the metabolic activity of individual populations and key microbial interactions. The primary goals of the current study were to integrate data collected in situ with in silico calculations across process-scales encompassing enzymatic activity, cellular metabolism, community interactions, and ecosystem biogeochemistry, as well as to predict and quantify the functional limits of autotroph-heterotroph interactions. Metagenomic and transcriptomic data were used to reconstruct carbon and energy metabolisms of an important autotroph (Metallosphaera yellowstonensis) and heterotroph (Geoarchaeum sp. OSPB) from the studied Fe(III)-oxide mat communities. Standard and hybrid elementary flux mode and flux balance analyses of metabolic models predicted cellular- and community-level metabolic acclimations to simulated environmental stresses, respectively. In situ geochemical analyses, including oxygen depth-profiles, Fe(III)-oxide deposition rates, stable carbon isotopes and mat biomass concentrations, were combined with cellular models to explore autotroph-heterotroph interactions important to community structure-function. Integration of metabolic modeling with in situ measurements, including the relative population abundance of autotrophs to heterotrophs, demonstrated that Fe(III)-oxide mat communities operate at their maximum total community growth rate (i.e. sum of autotroph and heterotroph growth rates), as opposed to net community growth rate (i.e. total community growth rate subtracting autotroph consumed by heterotroph), as predicted from the maximum power principle. Integration of multiscale data with ecological theory provides a basis for predicting autotroph-heterotroph interactions and community-level cellular organization.
| Microbial communities often display emergent properties, such as enhanced productivity, stability, and robustness, compared to their component populations in isolation. However, determining the governing principles of these emergent properties can be elusive due to the complexities of interpreting and integrating genomic and geochemical data sets collected at largely different observational scales. Here, we use multiscale, metagenome-enabled modeling of an Fe(II)-oxidizing community to extract information regarding biomass productivity limitations, relative population abundance, total biomass concentration, and electron acceptor uptake rates. The systematic approach used herein is broadly applicable to any microbial community with modest activity and metagenomic data as well as provides a mechanism to characterize interaction motifs in communities that include uncultivated organisms.
| Microorganisms are the largest component of the biosphere and drive biogeochemical cycles through metabolic activity [1]. Microorganisms commonly exist in biofilms or mats that contain numerous microenvironments due to the interplay between convection-, diffusion-, and chemical concentration gradients induced by microbial activity [2–4]. In addition, most natural microbial communities have diverse microbial populations and an array of nutrient and energy sources, which often precludes detailed analyses of microbial interactions linked to metabolic activity. Natural microbial communities containing well-characterized microbial populations and tractable nutrient inputs are excellent systems to elucidate the principles that organize microbial metabolism and interaction.
The phylogenetic diversity of microorganisms within Fe(III)-oxide microbial mats of acid-sulfate-chloride springs in Yellowstone National Park (YNP) is limited due to high temperature (65–75°C) and low pH (~ 3) [5–7]. These biomineralizing communities are formed and inhabited by a limited number of distinct phylotypes, including crenarchaea from the order Sulfolobales (e.g. Metallosphaera yellowstonensis str. MK1) and candidate phylum Geoarchaeota (e.g. Geoarchaeum str. OSPB) (supplemental material) [6–12]. The aqueous and solid-phase geochemistry of two such environments in Beowulf and One Hundred Springs Plain (OSP) hot springs have been studied in detail [5,12–15], and provide bounding conditions and physicochemical context for modeling microbial community interactions. The primary electron donors that drive chemolithoautotrophy in Fe(III)-oxide microbial mats include Fe(II) (25–40 μM) and possibly reduced forms of sulfur (dissolved sulfide < 10 μM) and As(III) (25–30 μM) [9]. The oxidation of Fe(II) coupled with the reduction of oxygen provides energy necessary for the fixation of carbon dioxide by M. yellowstonensis, a major autotroph in these mats [6,8,14,16,17]. The consumption of oxygen is diffusion-limited in Fe(III)-oxide microbial mats, and results in steep gradients in dissolved oxygen from 60 μM to below 1 μM over 0.5 to 1 mm [14]. The steep oxygen concentration gradients and corresponding relative abundance of community members as a function of mat depth indicate microbial competition for this limiting electron acceptor [12]. Genomic and mRNA data indicate that predominant autotrophs (e.g. M. yellowstonensis) and heterotrophs (e.g. Geoarchaeum str. OSPB) in Fe(III)-oxide mats utilize oxygen as an electron acceptor [7,16,18].
Carbon dioxide fixation by autotrophs in the community contributes 42 to 99% of the total microbial biomass carbon in Fe(III)-oxide mats from Beowulf and OSP hot springs; the remaining carbon originates from exogenous sources that are produced independent of system electron donor and acceptor requirements [17]. Carbon dioxide fixation has been demonstrated in M. yellowstonensis, which is one of the primary autotrophs in the oxic zones of the Fe(III)-oxide mats [17,19]. Metagenome analysis has established Geoarchaeum str. OSPB as a primary aerobic heterotroph, which comprises 30 to 50% of the total microbial community in the oxic zones of Fe(III)-oxide mats found at OSP [7,12]. Chemolithoautotrophic metabolism and the subsequent transfer of nutrients and energy to heterotrophs (e.g. Geoarchaeum str. OSPB) is hypothesized to drive major autotroph-heterotroph interactions along with competition for the primary terminal electron acceptor, oxygen. Sulfolobales viruses are highly represented in the metagenome sequence of these Fe(III)-oxide mats [10], which suggests that viral predation of M. yellowstonensis and other Sulfolobales populations contributes to the turnover of autotrophic biomass and creates reduced carbon sources for heterotrophs.
Genome-enabled stoichiometric modeling is a powerful approach in systems biology for examining metabolic acclimation to environmental stress across size scales from individual cells to communities of interacting populations [20–22]. A summary and graphical representation of stoichiometric modeling can be found in the supplemental information (Figure A in S13 File). Briefly, these approaches construct in silico representations of cellular metabolism inferred from genome sequence analysis [23]. The metabolic models define possible routes of electron transport, cellular energy production, carbon acquisition and central metabolism, and include details of the anabolic processes necessary to synthesize biomass. There are two major types of stoichiometric modeling: elementary flux mode analysis (EFMA) and flux balance analysis (FBA). EFMA identifies all distinct and indecomposable routes through a metabolic network; these routes are termed elementary flux modes (EFMs) [24]. EFMs, and non-negative linear combinations thereof, describe all possible physiologies independent of kinetic parameters, which makes EFMA well-suited for evaluating energetic efficiencies of different electron donors, acceptors, and nutrient sources involved in biomass production. FBA identifies optimal routes through a metabolic network that, for example, maximize growth rate for a given substrate uptake rate [25]. The use of optimization to identify these routes and rates for a given set of nutrient uptake rates is ideal for sampling a large metabolic space bounded by known but flexible kinetic parameters. Stoichiometric modeling has been used to predict optimal genotypes in engineered systems [20,26], interpret physiological behavior [21,27], and evaluate the transfer of mass and energy between distinct populations in a natural phototrophic microbial community [22] and others [28].
A primary goal in environmental microbiology is to understand and predict microbial behavior in communities, where interactions among different populations lead to emergent properties, such as enhanced productivity, stability, and robustness [29]. Consequently, the objectives of this study were to 1) construct metabolic network models for major autotroph and heterotroph populations present in oxic zones of high-temperature Fe(III)-oxide mats using metagenome sequence assemblies from representative sites; 2) analyze the use of electron donors and acceptors for the production of biomass and cellular energy under different nutrient limitations; 3) integrate individual population models to examine possible autotroph-heterotroph interactions that may be fundamental to the ecology of microbial communities (e.g. relative population abundances and oxygen competition); and 4) perform a sensitivity analysis based on parameters measured in situ to determine model limitations and identify future priorities for field measurements. The approach used here integrates data from the nanoscale of electron transport to microscale oxygen depth-profiles to hot spring-scale measurements of biotic Fe(III)-oxide deposition. Novel insights and governing principles of community structure and function were established through the integration of metagenome-enabled in silico approaches and in situ measurements.
Genomic data were used to construct in silico representations of the electron transport network responsible for the oxidation of Fe(II) and sulfur species and the reduction of oxygen in M. yellowstonensis str. MK1 (Fig 1). Genomic and physiological evidence indicated that this chemolithoautotroph oxidizes Fe(II) using proteins encoded by the fox operon, which are also found in other Sulfolobales [16]. M. yellowstonensis can also oxidize a variety of sulfur species, including sulfide, elemental sulfur, sulfite, and thiosulfate, to reduce the quinone pool to quinol [16]. Electrons in the quinol pool can drive cellular energy production via oxidative phosphorylation through the reduction of oxygen using a high affinity heme copper oxidase. Alternatively, electrons in the quinol pool can reduce NAD+, enter central carbon metabolism, and be used to reduce inorganic carbon via the 3-hydroxypropionate / 4-hydroxybutyrate (3-HP / 4-HB) pathway [17]. These electron transport pathways are hypothesized to provide the majority of energy to the studied microbial communities. The modeled central carbon metabolism of M. yellowstonensis included the tricarboxylic acid cycle, gluconeogenesis, the pentose phosphate pathway via ribulose monophosphate [30], and the mevalonate pathway [31].
Electron donor and acceptor pathways for M. yellowstonensis were integrated into a cellular-level metabolism model to quantify the relationship between growth and the consumption of environmental resources (Fig 1, supplemental material). Biological systems often minimize their requirements for growth limiting resources, providing an ecologically relevant basis for the in silico prediction of metabolic phenotypes [32]. A total of 6,337 elementary flux modes (EFMs) were calculated that produce M. yellowstonensis biomass using the inorganic electron donors, Fe(II), sulfide, elemental sulfur, sulfite, or thiosulfate. Each EFM was plotted as a function of moles of electron donor and moles of electron acceptor required to produce one carbon mole (Cmole) of M. yellowstonensis biomass (Fig 2A). The moles of electron donor required to form a Cmole of biomass was lowest for sulfide and highest for Fe(II) (Fig 2A), which follows the available free energy predicted for the oxidation of these electron donors coupled to the reduction of oxygen [33]. The oxidation of Fe(II) also required the most moles of oxygen per Cmole of biomass produced of the electron donors evaluated. These relationships were consistent on both a moles of electron donor and moles of oxidized electrons basis (Figure B in S13 File, Table A in S13 File for relevant degrees of reduction).
Genomic analysis of Geoarchaeum str. OSPB indicated that this organism is an organoheterotroph with the metabolic potential to utilize a wide variety of reduced carbon species, including biomass macromolecules (i.e. lipids, peptides, polysaccharides, and nucleic acids; see the supplemental material or [34] for a description of reactions involved). Details of Geoarchaeum biomass production from components of lysed Metallosphaera cells were elucidated in a prior report [34]. Briefly, the production of heterotroph biomass based on the consumption of autotroph biomass as a carbon source was evaluated with respect to the Cmoles of carbon substrate or moles of oxygen required to produce a Cmole of biomass (Table B in S13 File). The production of a Cmole of Geoarchaeum str. OSPB biomass required 2.4 Cmoles of autotroph biomass to supply the cellular energy and structural components for growth under carbon-limited conditions (Fig 2B, point A). Production of a Cmole of Geoarchaeum str. OSPB biomass under oxygen-limited conditions requires 3.8 Cmoles of autotroph biomass (Fig 2B, point B).
The growth of autotroph required substantial inputs of oxygen and was governed largely by the degree of reduction and redox potential of the electron donor (Fig 2C, Table A in S13 File). The predicted growth of autotroph on elemental sulfur required 3.5 moles of oxygen per Cmole of biomass produced while oxidation of Fe(II) required 38 moles oxygen per Cmole of biomass produced. Both predicted oxygen requirements are within the range of measured values for similar organisms growing on sulfur (3.5 moles of oxygen per Cmole biomass, assuming 1.8 * 10−13 grams per cell) and iron oxidation (35–104 moles oxygen per Cmole biomass) [35,36].
Different scenarios were used to evaluate heterotroph growth in this system: organic carbon was either made available, opportunistically, as a result of viral predation and subsequent lysis of the autotrophic cells, modeled as free monomer pools, or organic carbon was supplied as the result of a biological strategy where metabolites were excreted from the autotroph. Alternatively, organic carbon was provided from the surrounding landscape (hereafter called landscape carbon) which allowed for heterotrophic growth independent of the examined autotroph. The landscape carbon was modeled to have the same macromolecular composition as autotroph biomass for simplicity (supplemental material). The landscape carbon was produced from electron donors and acceptors external to the system boundaries and therefore did not contribute to oxygen consumption in the mat. Heterotroph growth on landscape carbon required 1.6 moles oxygen per Cmole of biomass produced, which is 54 and 96% less than the oxygen required to support the production of autotroph biomass from either elemental sulfur or Fe(II), respectively (Fig 2C, Table B in S13 File). The lower oxygen requirement for organoheterotrophic growth was due to utilization of reduced carbon substrates as both electron donors and anabolic precursors. By comparison, autotrophic metabolism required substantially more energy to reduce carbon dioxide, which resulted in high oxygen requirements, especially when Fe(II) was the primary electron donor. Consequently, the oxygen requirements to produce heterotroph biomass increased sharply when community-level interactions were included. For example, Geoarchaeum str. OSPB requires 2.4 Cmoles of autotroph biomass to produce a Cmole of heterotroph biomass under carbon-limited conditions (Fig 2B, Table B in S13 File). The total, aggregate oxygen requirement to produce heterotroph biomass from the biomass of an Fe(II) oxidizing autotroph included both the oxygen requirements of the heterotroph and the oxygen requirements to produce the autotroph biomass consumed, which resulted in 93 moles of oxygen being consumed per Cmole heterotroph biomass (Fig 2C).
Oxygen is required to produce biomass and cellular energy for both the autotroph and heterotroph, which results in resource competition between these two trophic levels. Additionally, the heterotroph consumed autotroph biomass as an electron donor, creating a competitive interdependence. Mass balances on the autotroph and heterotroph within the oxic system boundaries (0.7 mm depth, Fig 1C) were described as functions of growth rates, resource requirements, and biomass concentrations (Tables 1 and 2). The system was assumed to be at steady-state for the analyzed time scale of days to weeks, which resulted in a relationship between specific growth rates of the autotroph and heterotroph. This relationship and the flux of oxygen into the system were used to solve for steady-state total community growth rate, which sums both the autotroph and heterotroph, and net community growth rate, which subtracts autotroph carbon consumed by heterotroph from the total community growth rate, (Table 2). The rate of Fe(II) oxidation was then calculated using the autotroph growth rate. These relationships (Table 2) provided a mechanism to examine the effects of oxygen flux and the relative population abundance of autotroph and heterotroph on specific and community growth rates (Figs 3 and 4). Three oxygen fluxes into the mat were analyzed: 50, 100, and 200% of the average in situ oxygen flux (420 nmol O2 cm-2 h-1) measured with microelectrodes [12]. In situ values of relative population abundances of autotroph to heterotroph (0.3–0.5) were determined based on metagenome analyses from Fe(III)-oxide mat samples collected from Beowulf and OSP springs (Table C and D in S13 File).
Simulated total biomass concentrations were bounded by two constraints: 1) maximum specific growth rate of the autotroph and 2) oxygen flux into the mat. A microbial community composed of mostly autotroph was constrained to a minimum biomass concentration of 0.04 mg cm-3 by the maximum measured specific growth rate for M. yellowstonensis in vitro (μ = 0.1 h-1 [8,37]) and in situ oxygen flux into the mat (420 nmol O2 cm-2 h-1 [12,14]) (Figure C in S13 File). Increasing total biomass concentration decreased the specific growth rate for an autotroph dominated community and increased the feasible steady-state heterotroph abundance in the community at a given oxygen flux into the mat (Fig 3). Two scenarios were examined for heterotrophic growth: 1) a carbon-limited scenario, which was expected to occur at the top of the mat where oxygen is plentiful, and 2) an oxygen-limited scenario, which was expected at the bottom of the oxic zone (Fig 3). Heterotroph acclimation to intermediary oxygen availabilities would be bounded by these two scenarios. The observed in situ relative population abundance (0.3–0.5 autotroph per heterotroph) and oxygen flux was predicted to be feasible at steady-state with a total biomass concentration of ~0.2 mg cm-3, thereby defining a minimum total biomass concentration for the studied microbial community (Fig 3). Increases in total biomass concentration decreased the specific growth rate for the heterotroph.
The total community growth rate, which was the sum of autotroph and heterotroph produced, was constrained by competition for oxygen and the autotroph biomass requirement of the heterotroph. Increases in flux of oxygen into the mat increased total community growth rate supporting oxygen limitation of the modeled community. Increases in relative population abundance of autotroph to heterotroph decreased the total community growth rate, as it did with the specific rate of the autotroph (Figs 3 and 4A). This trend highlights the available mass and energy in autotroph biomass. This mass and energy can be utilized by heterotrophs with less oxygen required per biomass to further increase the total biomass using the same oxygen flux. Optimized carbon usage by the heterotroph, hypothesized to occur under carbon limitation, increased community growth rates by minimizing the autotroph biomass required to produce heterotroph. Conversely, optimized oxygen usage by the heterotroph increased carbon requirements for the heterotroph and decreased all steady-state growth rates by consuming more autotroph biomass (Figs 3 and 4).
The net community growth rate, which quantified autotroph and heterotroph accumulation as opposed to production, was constrained by competition for oxygen and the heterotrophic requirement for autotroph biomass. The net and total specific rate of an autotroph only community was equal because no biomass was consumed (Fig 4). Net community growth rate increased with oxygen flux, as observed for specific rate, and relative population abundance of autotroph to heterotroph (Fig 4B). The maximum net community growth rate occurs in a community composed of mostly autotroph and is equal to the total growth rate of the autotroph. The observed oxygen flux into the mat (420 nmol O2 cm-2 h-1 [12,14]) and biomass concentrations above 0.7 mg cm-3 resulted in negative net community growth rates, which are not sustainable in a steady-state community, due to the minimum metabolic activity to maintain an active population (Figure C in S13 File). The observed relative population abundances and oxygen flux predicted a maximum feasible, total biomass concentration of ~ 2.0 mg cm-3, above which the net community growth rate was negative. A negative net community growth rate indicated that the total biomass concentration in the mat was not sustainable and would decrease for the oxygen fluxes simulated (Fig 4B).
In situ measurements in natural ecosystems are complicated by numerous variables, such as seasonal variation in weather, wind, precipitation, temperature, and inherent heterogeneity. A sensitivity analysis was performed using sulfide and landscape carbon as alternative electron donors for the autotroph and heterotroph, respectively, using a multiscale, hybrid EFMA and FBA approach. The hybrid methodology first identified cellular-level, metabolic acclimation strategies to simulated environments, such as carbon- or oxygen-limited environments, using EFMA. The analysis hypothesized that organisms maximized the desired product, such as biomass or cellular energy, per limiting resource utilized (discussed above) (Fig 2). The same optima are calculable using FBA, but cellular-level EFMA provides additional insights, such as the number of reactions and distribution of suboptimal yields, which are difficult or prohibitive to determine using FBA [34]. The EFM resource requirements to produce biomass and associated maintenance energy were adjusted for specific growth rate using the relationship discussed in Pirt et al [38]. These cellular-level activities were represented by overall resource transformation reactions, quantified using the model exchange reactions, then incorporated into a model of community-level function, which was analyzed using FBA. FBA provides a convenient tool for assessing the impact of rate constraints and different modeling optimization criteria, such as maximizing net or total community growth rate, on the feasible range of metabolic activities. The hybrid FBA simulations evaluated three different electron donor scenarios (A, B, and C in Fig 5, Figure D in S13 File).
Firstly, Fe(II) and autotroph biomass were the only electron donors for the autotroph and heterotroph, respectively (Scenario A), which addressed the capacity of the community to oxidize Fe(II) based on varying oxygen flux into the mat (Fig 5, Figure E in S13 File). The resource requirements for the heterotroph to produce biomass and cellular energy were bounded by results from the carbon- and oxygen-limited EFMA simulations to approximate the high and low oxygen regions in the mat, respectively. The growth rate constraints used for FBA are described in Figure D in S13 File and Table 1 and included relative population abundance of autotroph to heterotroph, total biomass concentration, and the maximum specific growth rates of autotroph and heterotroph. The maximum specific growth rates of both populations were set to 0.1 h-1 based on in vitro studies of M. yellowstonensis [8,37]. Maximization of net and total community growth rate impacted the simulated rates but not the range of metabolic activities feasible by the system.
Secondly, autotroph biomass and landscape carbon were both evaluated as possible carbon sources for the heterotroph (Scenario B). This scenario examined the effect of 42 to 99% of the biomass carbon in the system having been supplied by the Fe(II)-oxidizing autotroph with the balance from exogenous, landscape carbon (Fig 5, Figure F in S13 File) as determined by in situ isotopic analyses [17]. The biomass carbon origins were simulated by setting the relative population abundance of autotrophy-based populations (i.e. autotroph and autotroph consuming heterotroph) and the landscape carbon consuming heterotroph. Growth of the autotroph-consuming heterotroph resulted in a lower limit for iron oxidation below which sustaining the heterotrophic population would consume more autotroph than produced (Fig 5, Figure F in S13 File).
Finally, sulfide was evaluated as an alternative electron donor for the autotroph (Scenario C). Elemental sulfur and sulfide can be present at low levels in Fe(II) oxidizing mats. The presented analysis only considered sulfide as it was predicted to have the largest impact (Fig 2A, Figure G in S13 File). Autotrophy was simulated to vary between oxidation of sulfide exclusively or Fe(II) exclusively with a total autotroph maximum specific growth rate of 0.1 h-1; therefore, as Fe(II) oxidation rates increased, less oxygen could be directed toward sulfide oxidation (Fig 5, Figure G in S13 File). A scenario where heterotroph was able to consume Fe(II) and/or sulfide oxidizing autotroph but not landscape carbon, predicted a trend similar to scenario C.
The three scenarios were compared to the in situ measurements of Fe(III)-oxide deposition and oxygen flux into the mat. The average in situ measurements for OSP Spring were within the predicted metabolic space for both scenarios B and C in a microbial mat that has a high biomass concentration (~ 2.0 mg cm-3 as determined above). This overlap between in silico and in situ observation suggests that oxidation of landscape carbon or combinations of landscape carbon and reduced sulfur species could account for the additional aerobic activity in OSP Spring communities. The average in situ measurements for Beowulf spring were within the predicted metabolic space for scenario C for the high biomass concentration simulations, which indicated that the community would have to use landscape carbon, as well as Fe(II) and sulfide oxidation to account for the oxygen consumption. At lower biomass concentrations the metabolic space covered by each scenario decreased, excluding the in situ measurements of both Beowulf and OSP Springs, which indicated if these springs have such low biomass concentrations there must be additional oxygen consumption occurring, such as the oxidation of additional electron donors (see Discussion).
Microbial processes span multiple spatial scales from metabolites, enzymes, individual cells, populations, and communities to microbial mats that can ultimately impact planetary biogeochemical cycling (Fig 1). Metagenomic data from Fe(III)-oxide mats in YNP provided a foundation for identifying enzymes responsible for electron transport in a dominant autotroph and heterotroph that occur within the oxic zones of Fe(III)-oxide microbial mats. Biochemical pathways for autotrophic and heterotrophic metabolism were integrated into stoichiometric models to quantify cellular-level resource requirements for biomass and cellular energy production (Fig 2). In situ measurements of relative population abundances (i.e. metagenomes) and fractions of autotrophic- versus landscape-based biomass carbon (i.e. stable carbon isotopes [17]) provided context for modeling microbial interactions within these communities. In addition, in situ measurements of oxygen flux into these mats using microelectrodes [12,14], and Fe(III)-oxide accretion rates from long-term temporal studies [12] (Fig 5) were integrated with the in silico models to quantify possible interactions between a dominant primary producer and secondary consumer, as well as the total contribution of these populations to the biogeochemical activity of the natural Fe(III)-oxide mats. Autotroph-heterotroph interaction modeling indicated that at least 98% of the measured oxygen flux in situ is consumed by the autotroph (e.g. M. yellowstonensis). This is due in part to the relatively low energy content of the electron donor (i.e. Fe(II)) and the high energy demands of carbon dioxide fixation. Conversely, heterotrophic activity had little effect on predicted Fe(II) oxidation rates as a function of oxygen consumed (Scenario A and B in Fig 5), though it represented the major fraction of steady-state biomass. The relatively high energy content of autotroph biomass compared to Fe(II) meant heterotroph growth increased the total biomass produced at the cost of a decreased net community growth (Fig 4).
The turnover of autotroph biomass is important to nutrient and energy flux through the mat community. It has been hypothesized that viral predation and subsequent cell-lysis represents a major mechanism of carbon cycling in microbial communities [39,40]. Numerous Sulfolobales viruses have been identified [41,42], and more specifically, the genome of M. yellowstonensis contains an extremely high abundance of transposases [16] and viral CRISPR spacers, which suggests that viruses are important in the life-cycle of these organisms. Assembled sequences of Sulfolobales viruses have been obtained from Fe(III)-oxide microbial mats [10], and the lysis of archaeal cells by viruses has been observed in situ using scanning electron microscopy (Inskeep, unpl). The turnover of autotroph biomass due to viral predation releases organic carbon enabling the production of heterotroph biomass, which leads to increased total community activity (Fig 4A). The predicted autotroph turnover is greater than 80% at the observed relative population abundances (Figure H in S13 File). To put this turnover value in perspective, it falls within the experimentally observed range of 26 to 1200% reported in other studied geothermal systems [43]. Cell lysis provides one mechanism of carbon and energy exchange; alternatives, such as metabolite exchange, could also contribute to these relative rates and abundances. Metabolite exchange is not expected to be the sole mechanism of exchange in these communities as it would necessitate greater than 80% of all autotrophically fixed carbon be secreted (Figure H in S13 File). The presented analysis does not exclude these mechanisms; indeed, a combination of lysis and metabolite exchange likely occur to varying degrees in all microbial communities. In addition to the lysis mechanism of carbon and energy transfer, simulations were run which considered only metabolite exchange between autotroph and heterotroph populations subject to the previously described community constraints. The metabolite exchange mechanism was simulated where the autotroph secreted monomer distributions consistent with the autotroph biomass composition to facilitate comparison with simulations of the lysis mechanism. The predicted maximum and minimum biomass concentrations were relatively unaffected (supplemental material). The simulated lysis mechanism required cellular energy expenditures by the autotroph to polymerize the monomer pools; this cellular energy was lost to the community when the organic material was transferred to the heterotroph. This energy expenditure was not required during simulated metabolite exchange mechanism conserving community energy and the oxygen used to produce it. However, the oxygen required for polymerizing monomer pools was very small relative to the oxygen required to fix DIC resulting in minimal differences in the predicted maximum and minimum biomass concentrations (Supplemental material).
The relative population abundances observed in situ suggest that the Fe(III)-oxide mat communities have metabolic activity that maximize energy acquisition from the environment thereby making them more competitive than those that exhibit lower rates of energy acquisition; this theory is known as the maximum power principle [44]. The ecological benefit to maximizing energy acquisition may be the dilution of autotroph biomass, which increases community diversity and has been shown to promote resilience and stability of the entire community from phage and/or predatory bacteria [45,46]. Applying the maximum power principle to metagenome-derived models may provide theoretical context for community structure-function relationships in other systems.
Analysis of the carrying capacity (i.e. maximum sustainable population numbers) of Fe(III)-oxide mats indicated that minimum and maximum total biomass concentrations exist where the specific growth rate of autotroph or oxygen flux into the mat become limiting, respectively. The minimum predicted biomass carrying capacity of 0.2 mg cm-3 (Fig 3) corresponds remarkably well with the calculated biomass concentration in Beowulf Spring (0.3 mg cm-3) (Table E in S13 File), which suggests that this microbial community is governed by the maximum specific growth rate of autotroph. Increases in heterotroph abundance would require an autotroph specific growth rate higher than the value observed in vitro [8]. This reduction in biomass concentration, and therefore autotroph limitation, is supported by high channel flow rate (20–30 cm s-1) and associated shear force in Beowulf Spring [12]. Conversely, the maximum predicted biomass carrying capacity of 2.0 mg cm-3 (Fig 4B and Figure I in S13 File) corresponds well with the calculated biomass concentration of OSP Spring (2.2 mg cm-3) (Table E in S13 File), which suggests that biomass production at this site is governed by the flux of oxygen. Indeed, in OSP Spring the channel flow rate is 10-fold lower than Beowulf (2–5 cm s-1), which results in less shear force and more limited oxygen transfer [12]. The presented modeling framework combined with ecological theory provides a powerful tool for understanding factors that control community structure and function, and can be applied to many natural and/or engineered microbial systems.
A variety of additional oxygen consuming processes in the Fe(III)-oxide mats could explain the higher amounts of measured oxygen flux relative to measured Fe(III)-oxide deposition rates [14] (Fig 5). For example, at least three major oxygen consuming processes (i.e., additional electron donors) could account for the higher measured oxygen flux than predicted for Fe(II) oxidation alone; these include the oxidation of reduced sulfur species, As(III), and reduced carbon from landscape carbon sources [14]. Small amounts of sulfide (< 5 μM) and occasional flocs of elemental sulfur can be present in Fe(III)-oxide depositing zones of acid-sulfate-chloride springs, such as Beowulf Spring, where high sulfide at discharge results in the deposition of elemental sulfur upstream of Fe(II) oxidation [5]. However, analysis of landscape carbon as an additional stimulant for heterotrophic activity could not account for the Fe(III)-oxide deposited for a given oxygen flux in the low biomass concentration simulations expected to be more representative of Beowulf Spring (Fig 5). Additionally, 25 to 30 μM As(III) is present in many acid-sulfate-chloride springs, and has been shown to be oxidized to As(V) by a bacterial autotroph population(i.e. Hydrogenobaculum spp.), which is commonly found in high abundance upstream of the examined mat community [47]. Elemental analysis of the poorly-crystalline Fe(III)-oxide phases indicates that significant amounts of As(V) are incorporated into the solid phase [5]. Any of these reduced species would result in consumption of additional oxygen if oxidized and warrant further study.
The integration of in situ measurements and metagenome-based in silico analyses provide unprecedented approaches to understand microbial interactions and community function. The analyses applied in the presented study provide context for the relative population abundance of autotrophs and heterotrophs, the minimum and maximum biomass concentration of microbial mats, and the possible effects of additional electron donors. The multiscale analyses presented here suggests that microbial interactions contribute to emergent properties of complex organization, such as increased productivity. These principles are likely shared across microbial and macro ecology.
In silico stoichiometric models were constructed for autotroph M. yellowstonensis str. MK1 (NCBI Taxon ID 671065, GOLD ID Gi04920) and heterotroph Geoarchaeum str. OSPB (NCBI Taxon ID 1448933, GOLD ID Gi0000638) in the following 5 step process. 1) Initial models were constructed using RAST [48–50]. 2) The elementally and electronically balanced reactions that represent carbon and energy metabolism were manually curated based on genome sequence (Joint Genome Institute—Integrated Microbial Genomes (JGI-IMG) [51]), prior genome analyses [6–8,10,16,17], literature surveys, and the MetaCyc [52,53] and KEGG databases [54]. 3) Reactions were assumed to close pathway gaps that would have resulted in auxotrophy or deviated from observed physiological behavior [17]. 4) The modeled macromolecular compositions of biomass for each population were 2% DNA, 11% lipid, 11% polysaccharide, 16% RNA, and 60% protein based on previous reports [55]. Genomes of each organism were mined for monomer distributions of DNA, RNA, and protein, based on the GC content, the ribosomal subunits, and the average amino acid distribution of all protein encoding genes, respectively (supplemental material, [51]). 5) Maintenance energy requirements were adjusted to calibrate the M. yellowstonensis and Geoarchaeum str. OSPB models to observed yields for Acidothiobacillus ferroxidans, a representative thermoacidophilic autotrophic Fe(II) oxidizing bacterium [56,57] and Alicyclobacillus acidocaldarius, a representative thermoacidophilic heterotrophic bacterium [58], respectively. Nongrowth associated maintenance energy (0.11 mmol ATP g biomass-1 h-1 for both M. yellowstonensis and Geoarchaeum str. OSPB) was calculated for 65°C assuming 50 KJ mol ATP-1 (supplemental material) [59], which determined the growth associated maintenance energy (171 and 149 mmol ATP g biomass-1 for M. yellowstonensis and Geoarchaeum str. OSPB, respectively).
EFMA-based cellular metabolisms (Fig 2A and 2B) were used as input reactions for FBA-based sensitivity analysis of electron donors available to the community. Briefly, the net exchange reactions from the EFMs were expressed on a specific biomass or specific cellular energy basis, such as iron per autotroph (mol Fe(II) per gram of autotroph), by normalizing to the biomass formed. The maximum specific growth rate (0.1 h-1), relative population abundances (0.3–0.5 autotroph to heterotroph), and DIC incorporation fractionation were then used to establish constraints on the maximum and minimum rates and population abundances (See supplemental material for explicit constraints and optimization criteria for the different environmental electron donor scenarios).
In silico stoichiometric models were constructed using CellNetAnalyzer version 2014.1 [60], and exported to RegEFMTool version 2.0 [61] to enumerate EFMs for each metabolic model. The macronutrients available to both populations were modeled as illustrated in Fig 1. Metabolic reconstructions for both populations, including reactions, genomic evidence, specific literature references, metabolites, stoichiometric balance, and gene regulatory rules can be found in the supplemental material. SBML files provided were exported and tested for import using CellNetAnalyzer version 2017.3 and the CNA SBML parser. Community FBA was performed using functions from the COBRA Toolbox [62] and overall reactions obtained from EFMs deemed ecologically competitive based on resource utilization. Functions used for the community analysis are available in the supplemental material. All computations were processed on a machine with at most two Intel Xeon X5690 and 120 GB RAM.
|
10.1371/journal.pntd.0004348 | Molecular Detection and Identification of Spotted Fever Group Rickettsiae in Ticks Collected from the West Bank, Palestinian Territories | Tick-borne rickettsioses are caused by obligate intracellular bacteria belonging to the spotted fever group (SFG) rickettsiae. Although Spotted Fever is prevalent in the Middle East, no reports for the presence of tick-borne pathogens are available or any studies on the epidemiology of this disease in the West Bank. We aimed to identify the circulating hard tick vectors and genetically characterize SFG Rickettsia species in ixodid ticks from the West Bank-Palestinian territories.
A total of 1,123 ixodid ticks belonging to eight species (Haemaphysalis parva, Haemaphysalis adleri, Rhipicephalus turanicus, Rhipicephalus sanguineus, Rhipicephalus bursa, Hyalomma dromedarii, Hyalomma aegyptium and Hyalomma impeltatum) were collected from goats, sheep, camels, dogs, a wolf, a horse and a tortoise in different localities throughout the West Bank during the period of January-April, 2014. A total of 867 ticks were screened for the presence of rickettsiae by PCR targeting a partial sequence of the ompA gene followed by sequence analysis. Two additional genes, 17 kDa and 16SrRNA were also targeted for further characterization of the detected Rickettsia species. Rickettsial DNA was detected in 148 out of the 867 (17%) tested ticks. The infection rates in Rh. turanicus, Rh. sanguineus, H. adleri, H. parva, H. dromedarii, and H. impeltatum ticks were 41.7, 11.6, 16.7, 16.2, 11.8 and 20%, respectively. None of the ticks, belonging to the species Rh. bursa and H. aegyptium, were infected. Four SFG rickettsiae were identified: Rickettsia massiliae, Rickettsia africae, Candidatus Rickettsia barbariae and Candidatus Rickettsia goldwasserii.
The results of this study demonstrate the geographic distribution of SFG rickettsiae and clearly indicate the presence of at least four of them in collected ticks. Palestinian clinicians should be aware of emerging tick-borne diseases in the West Bank, particularly infections due to R. massiliae and R. africae.
| Tick borne rickettsial diseases may have similar clinical characteristics, yet epidemiologically and etiologically different diseases. To date, no studies have been conducted to detect potential tick vectors of rickettsiae in the West Bank. Therefore, we aimed to identify tick species and to determine the presence of Rickettsia pathogens in naturally infected ixodid ticks. The overall prevalence of SFG rickettsiae detected in ixodid ticks in nine Palestinian districts was 17%. Our results document for the first time the finding of two important human pathogens carried by ixodid ticks in the West Bank: R. massiliae and R. africae, the agent of African tick bite fever. Genetically, the detected Rickettsia spp. clustered into 3 different groups: R. massiliae, C. R. barbariae and C. R. goldwasserii. Most of Rickettsia-infected ticks were collected from dogs, camels and sheep, increasing the risk of transmitting rickettsial infections to animal owners, shepherds and farmers. These findings highlight the importance of hard ticks and their potential hazard for human health in the West Bank.
| Tick-borne spotted fever group (SFG) rickettsioses are caused by obligate intracellular Gram-negative bacteria belonging to the genus Rickettsia [1]. Feeding ticks can transmit these microorganisms to humans and animals. Various vertebrates are suspected to serve as reservoirs for Rickettsia species; however, some are susceptible to rickettsial infections and may develop rickettsemia following tick bite [2]. The human disease may present as a fever with clinical symptoms including headache, rash, and occasional eschar formation at the site of the tick bites [3].
Mediterranean spotted fever (MSF), caused by Rickettsia conorii, is transmitted by the brown dog tick, Rhipicephalus sanguineus, which is well adapted to urban environments. Previous studies in Israel have documented the presence of two spotted -fever group (SFG) rickettsiae: the tick-borne rickettsia, Rickettsia conorii israelensis and the flea-borne rickettsia, Rickettsia felis [4], [5]. Rickettsia conorii israelensis has been described in Tunisia, Libya, Sardinia-Italy, and Portugal [6]. Furthermore, a number of other SFG pathogenic rickettsiae including Rickettsia africae, Rickettsia massiliae and Rickettsia sibirica mongolitimonae have been detected in ticks from Israel in addition to some rickettsial species such as Candidatus Rickettsia barbariae and Candidatus Rickettsia goldwasserii which were not associated with diseases, to date [7], [8], [9].
The number of newly described SFG rickettsiae has increased in recent decades [4]. Sequence analysis of PCR-amplified fragments targeting genes encoding the citrate synthase (gltA) [10], Rickettsia-specific outer membrane protein (ompA) [11], the 17kDa lipoprotein precursor antigen gene (17 kDa) [12], and the ribosomal 16S rRNA gene [13] has become a reliable method for the identification of Rickettsia species. Molecular typing of infectious agents is important for better understanding of ecological niches and identifying circulating strains and their virulence. Although various Rickettsia species are found in ticks from Israel; to date, no entomological survey has been carried out in the West Bank, and no clinical data or reports for the presence of tick-borne pathogens are available. Thus, this study aimed at identification of the circulating hard tick vectors and Rickettsia species in naturally infected ixodid ticks collected from the West Bank, using PCR and sequence analysis with special focus on their potential threat for humans and animals.
To identify the circulating hard ticks in the West Bank and to evaluate the presence of rickettsial infection in these ticks, one to ten hard ticks per animal host, for a total of 1,123, were collected during January to April, 2014 from dogs, camels, sheep, a horse, a wolf, and a tortoise residing in nine districts in the West Bank. The districts are located in three zones in the central, northern and southern regions of the country (Fig 1). All ticks were gently removed from their hosting animals by forceps or hand, and individually placed into small, labeled plastic tubes containing 70% ethanol for morphological identification. The ticks were identified using standard taxonomic keys [14], [15], [16], [17] and stored at −20°C until DNA extraction.
A maximum of five ticks of different tick species per hosting animal were randomly selected for DNA extraction. Genomic DNA was individually extracted from a total of 867 tick samples. Prior DNA extraction, individual ticks were washed with phosphate-buffered saline (PBS), air dried for 10 min on tissue paper and separately sliced into small pieces by a sterile scalpel blade then manually homogenized with a sterile micro pestle, resuspended in 200 μl of lysis buffer and 20 μl of proteinase K. After overnight incubation at 56°C with a continuous gentle shaking, the DNA was extracted using the QIAamp DNA tissue extraction kit (Qiagen, Hilden, Germany) following the manufacturer's protocol. Purified DNA was stored at 4°C until use. Three μl of template DNA (approximately 100–200 ng per tick) were used for each PCR.
Screening for the presence of rickettsial DNA was carried out by conventional PCR targeting a 250-bp fragment of the ompA gene using 107F and Rm299 primers as described previously [7] with the following modification: final volume of 25μl using PCR-Ready Supreme mix (Syntezza Bioscience, Jerusalem) including primers at 1μM final concentration. Positive samples were further characterized targeting a 426-bp portion of the 16SrRNA and a 265-bp portion of the 17kDa protein gene as previously described [18], [19]. For Rickettsia species identification, strong positive samples were sent for sequencing. DNA extract of the first ompA positive sample, identified as R. massiliae, by direct sequencing, was used as a positive control and ultra pure water were used as a negative control in each amplification reaction.
DNA sequences of the positive PCR products were assembled using Bioedit software, used in a BLAST search (ncbi.nlm.nih.gov/blast) and aligned with sequences of other rickettsial species registered in the GenBank. To infer relationships between the obtained amplicons and other reference sequences published in GenBank, a phylogenetic tree was constructed using MEGA6 program.
Statistical analysis was done using the SPSS program v20. Two –tailed t- test and Pearson’s correlation were performed. P-value <0.05 was considered statistically significant.
The animal population was residing in different farms throughout the West Bank. Prior to ticks sampling, the animal owners were verbally informed about the goals of the project and the sampling protocol. All owners gave their verbal informed consent to collect ticks from their animals. The study was approved by the ethics committee at the Faculty of Medicine in Al-Quds University-Palestine (EC number: ZA/196/013).
A total of 1,123 hard ticks were collected from 320 animals (234 dogs, 68 sheep, 10 camels, 5 goats, one horse, one wolf and one Mediterranean spur-thighed tortoise). From each infested animal one to ten ticks were collected. Of them, 547 were male ticks (48.7%), 511 females (45.5%) and 65 nymphs (5.8%). All tick samples were identified to the species level as follows: Rhipicephalus sanguineus (n = 694), Rhipicephalus turanicus (n = 191), Rhipicephalus bursa (n = 16), Rhipicephalus spp. (n = 21), Haemaphysalis parva (n = 100), Haemaphysalis adleri (n = 20), Hyalomma dromedarii (n = 68), Hyalomma impeltatum (n = 5), Hyalomma aegyptium (n = 4), and Hyalomma spp. (n = 4) (Table 1).
A set of 867 ticks comprising eight different species were screened for rickettsial DNA targeting Rickettsia–specific ompA (Table 2). A sample was considered positive when PCR yielded a fragment with the expected length (250 bp) of the ompA rickettsial gene. The overall prevalence of rickettsiae infection was 17% (148/867) in all ticks (Table 2). The detection of rickettsial DNA was significantly higher in female ticks (20.4%) than in male (15.7%) and nymph ticks (4.8%, p<0.01). The overall prevalence of rickettsial DNA was markedly higher in ticks collected from Nablus (34.3%; 66/192) compared to other districts (p<0.01) (Fig 1).
The infection rates in Rh. turanicus, Rh. sanguineus, Haemaphysalis adleri, H. parva, Hyalomma dromedarii and H. impeltatum ticks were 41.7, 11.6, 16.7, 16.2, 11.8 and 20% respectively. None of the ticks belonging to Rh. bursa (0/13) and H. aegyptium (0/4), taken from sheep and one tortoise, respectively, were infected (Table 2).
Among the positive samples (n = 148), identification of rickettsial DNA based on sequencing of the ompA amplicons were successfully obtained from 63 (42.6%) samples which showed strong bands on agarose gel. These samples were subsequently subjected to two additional amplification reactions targeting the 16SrRNA and 17kDa genes of Rickettsia species. Successful sequences were only obtained from (35/63) and (36/63) by 16SrRNA and 17kDa PCR, respectively.
BLAST analysis of the positive ompA sequences revealed 4 different rickettsial species. Twenty eight ticks were tested positive for R. massiliae-DNA including 15 Rh. turanicus, 11 Rh. sanguineus, one Haemaphysalis parva and one H. adleri, all obtained from dogs and sheep. C. R. barbariae-DNA was found in 12 ticks: 5 Rh. turanicus, 3 Rh. sanguineus and 4 H. dromedarii. The DNA of C. R. goldwasesrii- was detected in 17 ticks: 2 Rh. turanicus, 7 Rh. sanguineus, 6 H. parva and 2 H. adleri. Four ticks were positive for Rickettsia species found in four Rh. sanguineus ticks, Furthermore, R. africae-DNA was detected in two ticks, H. impeltatum and H. dromedarii obtained from two different camels in Jericho (Fig 2A). There were no cases in which multiple rickettsiae species were detected in the same infected tick.
Phylogenetic analysis based on ompA sequences revealed three main clusters: R. massiliae, C. R. barbariae and C. R. goldwasserii. Cluster I, representing the R. massiliae group (n = 28). In this cluster, the nucleotide ompA sequences of R. massiliae identified in this study were identical to each other and to the respective R. massiliae reference sequence (accession no. KJ663746.1) deposited in the NCBI GenBank (Fig 2A). The same samples also showed one cluster and had 100% similarity to the R. massiliae reference strains (NR074486.1 and JN871727.1) based on the 16sRNA and 17kDa analysis, respectively (Fig 2B and 2C). The DNA sequence of Rickettsia canadensis (CP003304.1) that do not belong to the SFG [20], was used as an out group in the analysis of ompA gene while the sequences of Anaplasma phagocytophilum (EU 436157.1) and Rickettsiaceae bacterium (CP009217.2) were used as out groups in the analyses of the 16SrRNA and 17kDa genes, respectively. On the basis of the ompA sequences, DNA sequences of four amplicons (1.4G, 1.4H, 14.8 A, 14.8 B) had several nucleotide differences and showed 96–98% sequence identity to the ompA sequences of R. massiliae identified in this study and to the reference strain of R. massiliae (KJ663746.1). These samples showed 92% and 89% sequence identity to the ompA sequences of the reference strain sequences of R. aeschlimannii (KF791253.1) and R. raoultii (KR608786.1), respectively. The four samples were further characterized by 17kDa and 16S rRNA genes, the partial 17kDa gene sequence of these samples revealed 94% and 93% similarity to the reference strain sequences of R. raoultii (KT261760.1) and R. conorii (JN182793.1), respectively. Two of them (14.84A and 14.8B) formed a separate branch with in this group complicating the further confirmation of this Rickettsia spp. (Fig 2C). However, none of these samples (n = 4) were successfully sequenced based on 16SrRNA gene.
Cluster II represents the C. R. goldwasserii group. The ompA sequences had 100% similarity to the C. R. goldwasserii reference sequence (HM136928.1) whilst the 17kDa sequences of the same samples had 100% similarity to the corresponding sequence of an incompletely described Rickettsia sp. Belarus (JQ711214.1) (Fig 2C). The 16SrRNA sequences showed 99% similarity to that of R. rickettsii (NR103923). Cluster III represents the C. R. barbariae group; the ompA and 16SrRNA sequences had 100% similarity to the Candidatus Rickettsia barbariae reference sequences (JF700253.1 and EU272189.1), respectively (Fig 2A and 2B). When 17kDa sequences were obtained from the same samples, they showed 99% sequence similarity to the homologous fragments of R. raoultii (KT261760.1) and to the other unidentified Rickettsia sp. (KM386654.1) as revealed by BLAST analysis.
A sub-cluster of two ompA sequences was identified as R. africae and showed 100% sequence similarity with a homologous fragment of R. africae (JF700254) detected in Hyalomma detritum from the Golan Heights [9]. The 17kDa sequences of these two samples had also 100% nucleotide identity to that of R. africae (KF646137.1) but they were not detected by the 16S rRNA PCR.
The overall prevalence of SFG rickettsiae detected in ixodid ticks in nine Palestinian districts was 17%. Ticks detected in this study belonged to three genera (Rhipicephalus, Haemaphysalis and Hyalomma) and collected from different host animals. The findings of this study highlight the importance of hard ticks for human health in the West Bank. Rhipicephalus sanguineus was the most prevalent tick species found in this study. It parasitized a wide range of mammals including dogs, sheep, goats and horses; however the main vector for SFG rickettsiae detected in this study was Rh. turanicus. Our results document the detection of two important human pathogens in ticks from the West Bank, R. massiliae and R. africae. R. massiliae was the most prevalent rickettsial species detected in this study. This pathogen was first isolated from Rh. sanguineus in Marseille in 1992 [21], and since then it has been detected in ticks within the genus Rhipicephalus in Greece, Spain, Portugal, Switzerland, Sardinia (Italy), Morocco and Israel [22], [23], [4], [24], [25], [7]. The first isolation of R. massiliae was reported from a human patient in Sicily in 1985 and identified in 2005 [26]. The clinical presentation of R. massiliae infection has been previously described [23],[27],[28]. Common clinical signs include fever, night sweats, headache, maculopapular rash and necrotic eschar at the tick bite site. In agreement with other studies, all ticks infected with this pathogen in this study belonged to the genus Rhipicephalus except for two which belonged to the genus Haemaphysalis obtained from the same dog in Jenin district. However, detection of R. massiliae in Haemaphysalis punctata ticks was previously reported in southeast England [29].
In this study, most of R. massiliae- infected ticks were removed from dogs (77%) and to a lesser extent from sheep (23%), increasing the risk of transmitting rickettsial infections to the animal owners, i. e. dog owners, shepherds and farmers. The phylogenetic tree based on partial DNA sequences of ompA, 16SrRNA and 17 kDa showed higher genetic variability among the R. massiliae strains using ompA gene than 16SrRNA and 17 kDa loci. The second SFG pathogenic Rickettsia found in this study was R. africae, the agent of African tick bite fever. The detection of R. africae, in Hyalomma spp. collected from camels in the West Bank confirms the results of a previous report associating R. africae with Hyalomma ticks in Egypt and Israel [30], [8], [9]. Livestock movements and migratory birds may play a role in the geographic spread of R. africae [31]. We observed a strong geographic correlation between the overall prevalence of rickettsial DNA in ticks collected from Nablus (Northern district) compared to the prevalence of infected ticks collected from other districts in the north, Ramallah (centre) and Hebron (south). Future studies with high representative number of ticks are required to address the comparative importance of geographic distribution on the infection rate of these ticks in the West Bank.
Candidatus. R. goldwasserii was also detected in Rhipicephalus ticks collected from dogs in the northern region of the West Bank. This Rickettsia was first detected in two Haemaphysalis ticks (H. adleri and H. parva) from golden jackals in Israel. Phylogenetic analysis based on concatenated four gene fragments (gltA-ompA-sca4-ompB) indicated that the nucleotide sequences of these SFG rickettsiae belonged to a novel phylogenetic lineage related to C. R. siciliensis detected in Rh. turanicus ticks. [32]. In the present study, the identification of C. R. goldwasserii in Rh. sanguineus and Rh. turanicus, in addition to the already known Haemaphysalis spp, expands the current knowledge concerning tick species that host C. R. goldwasserii in our region. Based on ompA phylogeny, high genetic homology was observed among the C. R. goldwasserii group identified in this study. However, these samples were found to be 100% similar to the corresponding sequence of a not well characterized Rickettsia sp. Belarus and 99% similar to that of R. rickettsii based on 17kDa and 16srRNA respectively. Thus, for a more accurate classification of this uncultivated SFG Rickettsia, further testing and phylogenetic analysis with additional genes is needed since no sequences of these two latter genes of C. R. goldwasserii were available in the GenBank.
This is the first study to report the presence of C. R. barbariae in 9.6% of Hyalomma ticks in the West Bank. The presence of C. R. barbariae has been previously reported in several Rhipicephalus spp. in Portugal, Italy, France, Cyprus and later in Rhipicephalus ticks flagged from the vegetation in Israel [24],[33],[2], [9]. However, no ompA sequence differences were observed in C. R. barbariae DNA detected in Hyalomma or Rhipicephalus ticks collected from camels, dogs and sheep residing in different localities throughout the West Bank. In conclusion, the findings presented in this study provide evidence for the presence of R. massiliae and R. africae in different ixodid ticks collected from various regions in the West Bank. In addition to Rhipicephalus species, members of the genus Hyalomma and Haemaphysalis may also play an important role in the epidemiology of SFG Rickettsia spp. Clinicians in the West Bank and neighboring countries should consider a range of SFG diseases in the differential diagnoses of patients present with fever of unknown origin and clinical signs compatible with rickettsioses.
|
10.1371/journal.ppat.1005835 | The Structural Architecture of an Infectious Mammalian Prion Using Electron Cryomicroscopy | The structure of the infectious prion protein (PrPSc), which is responsible for Creutzfeldt-Jakob disease in humans and bovine spongiform encephalopathy, has escaped all attempts at elucidation due to its insolubility and propensity to aggregate. PrPSc replicates by converting the non-infectious, cellular prion protein (PrPC) into the misfolded, infectious conformer through an unknown mechanism. PrPSc and its N-terminally truncated variant, PrP 27–30, aggregate into amorphous aggregates, 2D crystals, and amyloid fibrils. The structure of these infectious conformers is essential to understanding prion replication and the development of structure-based therapeutic interventions. Here we used the repetitive organization inherent to GPI-anchorless PrP 27–30 amyloid fibrils to analyze their structure via electron cryomicroscopy. Fourier-transform analyses of averaged fibril segments indicate a repeating unit of 19.1 Å. 3D reconstructions of these fibrils revealed two distinct protofilaments, and, together with a molecular volume of 18,990 Å3, predicted the height of each PrP 27–30 molecule as ~17.7 Å. Together, the data indicate a four-rung β-solenoid structure as a key feature for the architecture of infectious mammalian prions. Furthermore, they allow to formulate a molecular mechanism for the replication of prions. Knowledge of the prion structure will provide important insights into the self-propagation mechanisms of protein misfolding.
| The structure of the infectious prion (PrPSc), which is responsible for Creutzfeldt-Jakob disease in humans and bovine spongiform encephalopathy, has escaped all attempts at elucidation due to its propensity to aggregate. Here, we use the repetitive organization inherent in amyloid fibrils to analyze the structure of GPI-anchorless PrP 27–30 via electron cryomicroscopy. Fourier-transform analysis of averaged fibril segments indicates a repeating unit of 19.1 Å. In agreement with this observation, 3D reconstructions reveal that each fibril contains two distinct protofilaments and that the height of each PrP 27–30 molecule in these fibrils is ~17.7 Å. Together the data indicate a four-rung β-solenoid structure as a key feature for the architecture of infectious mammalian prions. The data conflict with all previous models for the structure of PrPSc and allow the formulation of a molecular mechanism for the replication of prions.
| Little is known about the structure of the infectious prion protein, the infectious agent causing prion diseases such as sheep and goat scrapie, bovine spongiform encephalopathy or “mad cow disease”, chronic wasting disease in cervids (deer, elk, moose, and reindeer), and Creutzfeldt-Jakob disease in humans. The structure of these infectious conformers is essential to understanding prion replication and the development of structure-based therapeutic interventions. The non-infectious, cellular prion protein (PrPC), which has its highest expression levels in neurons, is misfolded through a posttranslational process into an altered, infectious conformer termed PrPSc or prion [1]. The structure of recombinant PrP, which approximates the structure of PrPC, has been solved repeatedly by NMR spectroscopy [2] and X-ray crystallography [3]. PrPC consists of an unfolded N-terminal domain and a largely α-helical C-terminal domain, which contains three α-helices and a short, two-stranded ß-sheet [2,3]. In contrast, PrPSc has been found by a variety of methods to contain predominantly ß-sheet structure [4].
PrPSc and its N-terminally truncated variant, PrP 27–30, are generally insoluble and prone to aggregation into an assortment of quaternary structures, such as amorphous aggregates, 2D crystals, and amyloid fibrils. None of these aggregation products are amenable to conventional structural analysis techniques such as X-ray crystallography or solution NMR spectroscopy. To overcome the paucity of experimental data on the structure of the infectious prion, molecular modeling has been used by a variety of researchers to predict its structure. Little agreement exists among the published molecular models regarding the nature of the infectious conformer and a large range of different folds have been put forward [5], with the most recent entry proposing a parallel in-register ß-sheeted structure [6]. While none of the published models satisfy all experimental restraints [5], a ß-helical architecture has been suggested as a likely candidate for the structure of the infectious prion [7,8].
To investigate the structure of the infectious prion, we used electron cryomicroscopy (cryo-EM) to record and analyze the structure of brain-derived, murine prion protein amyloid. Brain-derived PrPSc has a high level of molecular heterogeneity due to its GPI-anchor and N-linked carbohydrates, which increases the difficulty to analyze its structure considerably. On the other hand, brain-derived PrPSc, as the disease-relevant conformer, can provide insights into the infectious state, which the more well-behaved, recombinantly-derived PrP amyloid, cannot [9]. By using transgenic mice expressing a GPI-anchorless form of the prion protein, which is also substantially underglycosylated [10], a more homogeneous version of the prion protein could be analyzed. GPI-anchorless PrPSc is deposited predominantly as large fibrillar amyloid plaques, which allows for a milder purification procedure compared to traditional purification approaches [11]. In prion-infected mice expressing GPI-anchorless PrP, PrPSc retains its full infectivity, while the neuropathology is similar to a cerebral amyloid angiopathy (CAA) as seen in Alzheimer’s disease [12] and hereditary prion protein amyloidosis [13,14]. This difference in apparent neuropathology is not surprising given that infectivity and toxicity of PrPSc are not unequivocally linked [15]. After all, GPI-anchorless PrPSc cannot attach to the cell membrane, and therefore exhibits a different cellular and tissue distribution. Nonetheless, the homogeneous preparation of GPI-anchorless, infectious prions facilitated novel insights into the structure of PrPSc by cryo-EM.
GPI-anchorless PrP 27–30, which, for consistency, is named according to the molecular weight of GPI-anchored PrP 27–30, was purified from the brains of transgenic mice expressing GPI-anchorless PrP that were infected with prions of the Rocky Mountain Laboratory (RML) strain [10,16]. The purification procedure took advantage of both the in vivo formed amyloid fibrils of GPI-anchorless PrPSc and their partial resistance against Proteinase K (PK) digestion. Limited proteolysis of GPI-anchorless PrPSc, similar to the digestion of GPI-anchored PrPSc, removes just the first ~66 N-terminal residues, leaving a ~17 kDa PK-resistant, GPI-anchorless, unglycosylated fragment, and a very modest amount of monoglycosylated protein [10,17,18] (Fig 1A and 1B). To verify that the purified GPI-anchorless PrP 27–30 preparations were still infectious, as reported by others, we inoculated a cohort of wild-type mice with the purified prions. All animals developed the typical neurological signs of RML-prion disease. Furthermore, biochemical and histopathological analyses demonstrated that the brains of infected mice showed all hallmarks of pathogenic prion disease, for example partial PK-resistance, presence of vacuolation, gliosis, and PrPSc deposition in the brain tissue (Figs 1B and 2). The observed, moderately-extended incubation period compared to animals inoculated with untreated brain homogenate (Fig 1C) is compatible with the use of PK treatment, which is known to degrade PrPC and PK-sensitive forms of PrPSc, thereby reducing the apparent prion titer [19,20]. In addition, the highly aggregated nature of the purified prions is also known to lower the effective prion titer [21,22].
The cryo-EM images from GPI-anchorless PrP 27–30 fibrils were examined in detail. Images showed fibrils ~10 nm wide, composed by two intertwined, twisting protofilaments, with a space between them (Figs 3, 4 and S1). A clearer view of the fibrils was obtained in 3D tomograms (S1 Video and S2 Fig), which allowed a more facile visualization of individual fibrils. Separate fibrils were found to display either a left- or right-handed twist, or to be essentially straight. This apparent variability with respect to the fibril helicity of GPI-anchorless fibrils had been observed before in negatively stained sample preparations [17]. Reconstructed tomograms allowed measurement of fibril widths, which were confirmed to be 9.55 ± 1.15 nm (standard deviation; n = 261) (S1 Fig). The limited dispersion of the values is similar to that of other amyloids [24].
The raw images contain high-resolution information, as surmised from Fourier-transform analyses routinely showing a 4.8 Å cross-β signal for individual fibrils along their axis (Fig 3) that can be readily interpreted as originating from β-strands running perpendicular to the fibril axis. It is notable that the orientation of the cross-β signal is strictly dependent on the orientation of the individual amyloid fibrils, which proves its origin (Fig 3, black and white arrows). Furthermore, Fourier-transform analyses of nearby empty areas of ice or carbon film never showed a 4.8 Å signal (Fig 3, dotted boxes), indicating that the 4.8 Å signal is indeed a cross-β signal similar to what was seen in X-ray fiber diffraction from PrP 27–30 [25]. In fact, to our knowledge, this is the first time that electron cryomicroscopy and the use of a second generation direct electron detector (Falcon II) has allowed to detect the 4.8 Å cross-β signal in Fourier-transform analyses of otherwise unprocessed cryo low-dose electron micrographs of individual amyloid fibrils or small bundles of fibrils.
Encouraged by this, we set out to extract as much additional structural information from the images as possible. In particular, we aimed at identifying the individual PrP subunits stacked in each protofilament. Unfortunately, fibrils were highly aggregated and did not display a constant helical periodicity. This impeded the use of the iterative helical real-space reconstruction (IHRSR) algorithm, which has frequently been used to process electron micrographs of amyloid fibrils to obtain high-resolution structural information [26]. Instead, we applied two single particle approaches to analyze a large number of short fibril segments from high magnification electron micrographs.
In the first single particle analysis approach, cryo-EM images were selected that presented clear Thon rings to enable correction of the contrast transfer function (CTF). In total, we extracted 1305 non-overlapping isolated fibril segments that were then aligned, classified, and averaged (S3 Fig). An averaged power spectrum of 1072 aligned segments showed a 4.8 Å intensity (Figs 5A and S4), characteristic of the cross-β structure of amyloid fibrils, also detected in many raw images (vide supra). The arc-shape of the 4.8 Å signal was indicative of an imperfect alignment of the fibrils. To overcome this problem, we analyzed each of the protofilaments that make up the fibril individually, which revealed 19.1 Å and ~40 Å signals upon Fourier-transform analysis (Figs 5A, S4 and S5). These spacings correspond to 4 and 8 multiples of 4.8 Å β-strands, and indicate the existence of a structural subunit with a height of 4 β-strands, that associates vertically with another subunit to form a higher order dimeric structure. Another feature, the absence of a strong ~10 Å signal on the equator of the Fourier-transforms (Figs 5A and S5), is generally interpreted to indicate the presence of β-helical and β-solenoidal structures [25,27]. The ~10 Å signal is commonly seen with stacked β-sheet structures such as Aβ(1–40) amyloid [28], and given the fact that our electron micrographs readily display the 4.8 Å cross-β signal in Fourier-transform analyses of individual fibrils (Fig 3), the ~10 Å signal could be expected if a stacked β-structure were present. Therefore, the most parsimonious explanation of this spacing hierarchy is the presence of a 4-rung β-solenoid as the basic subunit along the protofilament axis.
In an alternative reference-free single particle approach, we selected 2725 fibril segments from individual protofilaments, and performed eigenvector data compression and unsupervised automatic classification (Fig 5B). This analysis was performed without alignments and thereby avoiding reference bias (see materials and method for details and S6 Fig). The average of the amplitude spectra of all 20 classes (Fig 5B) revealed peaks at ~20 Å and ~40 Å (Fig 5D), in good agreement with the first approach. Furthermore, individual class averages (Fig 5B) showed distinct densities along the fibril axis with an average height of ~20 Å (Fig 5C), corroborating the other measurements.
Considering all this, we reasoned that the volumetric information provided by individual fibrils should provide an independent assessment of the conclusion that GPI-anchorless subunits are ~20 Å "tall". We picked isolated fibrils with images clearly showing a crossover point and covering at least one half turn (180°) without overlap to other fibrils. Given the already mentioned extensive clumping of the fibrils, which showed a high degree of lateral aggregation and, most of the times, overlapped with neighboring fibrils, only a few fibrils meeting these visual requirements were selected (Figs 4 and S1). Based on their morphology and dimensions, these fibrils are representative of the majority of specimens seen in the micrographs, and no bias, other than the mentioned selection requirements, took place to choose them.
We then generated 3D reconstructions of these fibrils, taking advantage of the fact that the 2D image of a helical object contains rotated projections of its 3D surface. Thus, we segmented the isolated fibrils into overlapping boxes along the helical axis and analyzed the fibril segments as single objects. Each box is a different view of identically the same fibril rotated around and translated along the helical axis [29]. By measuring the helical repeat distance of the fibril (images shown in Fig 4) we could estimate the angular orientation of each box in the set (S7 and S8 Figs).
The 3D reconstruction of one of these individual GPI-anchorless PrP 27–30 fibrils, with a maximum width of 9.1 nm and a crossover distance of 95 nm, showed two approximately 50 x 29 Å oval-shaped protofilaments (Fig 6). Fibril reconstruction statistics are listed in S1 Table. The density profile of the cross-section suggested several densities distributed within the core of the protofilament (Fig 6F), compatible with a ß-helical or ß-solenoidal fold and similar to what has been seen in the cross-section of the fungal HET-s prion [30]. The 3D reconstructions of the other isolated GPI-anchorless PrP 27–30 fibrils also contained two protofilaments twisting about the fibril axis (S9 Fig). While the general shape and features of protofilaments were remarkably consistent, the four fibrils were structurally heterogeneous with respect to the protofilament orientation (Fig 7), their widths (9.5 nm, 9.4 nm, and 8.7 nm) and crossover distances (76 nm, 67 nm, and 98 nm), which is a common hallmark of amyloid fibrils [31,32], but is also likely to be a reflection of analytical variability, indicating the resolution limitations of our approach. However, the level of resolution attained in these 3D reconstructions allowed us to calculate the average height of a GPI-anchorless PrP 27–30 monomer within each protofilament.
The average protein density has been estimated to be 0.8129 Da/Å3 [31,33]. However, considering that a highly compact amyloid might have a slightly different density value, we calculated the density value of a HET-s monomer stacked in a HET-s prion fibril ([30] PDB: EMD-2946). The value obtained was of 0.903 Da/A3. Based on a molecular mass of 17,148 Da [18], the calculated molecular volume a GPI-anchorless monomer is 18,990 Å3. Consequently, the average height per monomer came to 17.7 Å (Table 1). A similar calculation using the generic 0.8129 Da/Å3 protein density value would result in a monomer height of 19.7 Å. These height value calculations confirm that the ~20 Å spacings detected by our two independent single particle analyses originate from individual GPI-PrPSc subunits stacked along the protofilament axis, lending further support to the 4-rung β-solenoid interpretation for the structure of PrPSc.
Together, these data support a model in which the structure of PrPSc and PrP 27–30 consists of a four-rung ß-solenoid with a central, ß-strand-rich core, which is also supported by results obtained with X-ray fiber diffraction [25]. Fig 8 shows a cartoon representing the key elements of the prion architecture surmised from the data obtained in the present studies. It needs to be emphasized that this is not an atomistic model, but a cartoon only—meant to visualize the overall architecture of a four-rung ß-solenoid configuration. In order to account for a molecular height of 19.2 Å (4 x 4.8 Å) the approximately 144 residues of PrP 27–30 must adopt a coiled ß-sheet conformation (Fig 8B). This ~19 Å constraint is particularly relevant because it was obtained across many class averages, i.e. it is not characteristic of just one particular fibril architecture, but rather, emerges as a universal feature of the majority of fibrils present in the preparation. These data agree very well with data obtained from X-ray fiber diffraction of preparations of purified PrPSc samples of different kinds [25]. Although the actual number of residues in ß-sheet conformation per coil remains unclear, hydrogen-deuterium exchange and limited proteolysis studies have shown that brain-derived PrPSc consists of several ß-strands connected by short turns and loops [4,18]. Likely, the location of proline residues and the cysteine disulphide bridge pose constraints to the threading of the solenoid, with these elements likely located at corners [31,34].
Our cryo-EM images of GPI-anchorless PrPSc fibrils, and their subsequent analysis, show that they consist of two intertwined protofilaments, in agreement with a recent study of negatively stained PrPSc fibrils [35]. Each protofilament exhibits an approximately ellipsoidal cross-section with a linear volume of ~9,900 Å3/nm. Given the fact that cryo-EM preserves the native structure of specimens, this information sets a structural restraint for the conformation of GPI-anchorless PrPSc. One important implication is that PrPSc subunits can only fit into protofilaments with the observed dimensions (Table 1), if they are folded up onto themselves. Based on the routine observation of regular 4.8 Å cross-ß signals in individual GPI-anchorless PrP 27–30 fibrils (Fig 3), a ß-solenoid arrangement is the easiest way to accommodate the peptide into the available protofilament volume.
This arrangement would result in 4 x 4.8 Å (~19 Å) repeats along the protofilament axis, which is exactly what our two independent single particle analyses revealed (Figs 5 and S5), together with an additional ~40 Å signal that likely corresponds to a vertical pairing of two PrPSc subunits, as seen in HET-s prion fibrils [30]. Therefore, our cryo-EM data revealed a ß-solenoid architecture as the basic element for the structure of the mammalian prion GPI-anchorless PrPSc (Fig 8), which is in agreement with previous results obtained by X-ray fiber diffraction for other prions such as RML and Sc237 PrP 27–30 [25]. As an important corollary, our cryo-EM data are incompatible with models based on alternative architectures, such as the parallel in-register ß-sheet fold, which is based on a single, superpleated protofilament with a molecular height of only 4.8 Å [6]. It is noteworthy that a β-solenoidal architecture has been demonstrated, as already mentioned, for the HET-s prion [30,36], and has been proposed for insulin [31] and SH3 amyloid [37] fibrils based on cryo-EM data.
While important structural elements still need to be defined, such as which residues participate in the ß-strands that form each solenoid rung, and which ones are located in turns and connecting loops, what we have learned about the structure of GPI-anchorless PrP 27–30 and its four-rung ß-solenoid architecture, allows us to extrapolate about possible templating mechanisms that control the replication of infectious prions in vivo. In-phase stacking of identical residues along the peptide chain, as was proposed by the parallel in-register ß-sheet model [6], can be ruled out, as mentioned above, due to the experimentally determined height constraints. Therefore an alternative templating mechanism must explain the replication of prions in general, and the fidelity required for transmitting distinct prion strains in particular [19].
Templating based on a four-rung ß-solenoid architecture must involve the upper- and lowermost ß-solenoid rungs. These edge strands are inherently aggregation-prone, as they are predestined to propagate their hydrogen-bonding pattern into any amyloidogenic peptide they encounter [38]. In fact, the ß-strands of native proteins that contain a ß-solenoid are capped by loops and other structures to block unregulated propagation of ß-sheets. Furthermore, the elimination of the capping structures results in edge-to-edge-driven oligomerization of the "de-capped" ß-solenoids [39]. Thus, it is easy to conceive that these upper and lower ß-solenoid rungs can template an incoming unfolded PrP molecule to create additional ß-solenoid rungs.
It is noteworthy that the molecular forces responsible for the templating − hydrogen-bonding, charge and hydrophobic interactions, aromatic stacking, and steric constraints − are fundamentally similar to those operating during DNA replication. Obviously, the exquisite specificity of the A:T and G:C pairings is lacking and instead, a much more complex array of forces controls the pairing of the pre-existing and nascent ß-rungs. Once an additional ß-rung has formed, it creates a fresh "sticky" edge ready to continue templating until the incoming unfolded PrP molecule has been converted into another copy of the infectious conformer.
Furthermore, the stacking of GPI-anchorless PrP 27–30 molecules into amyloid fibrils, or, in other words, the way in which templating occurs, is either based on a) a head-to-tail orientation resulting in amyloid fibrils with intrinsic polarity (Fig 8), or b) a head-to-head and tail-to-tail orientation, which would result in generally apolar amyloid fibrils. In the former case, templating of ß-sheets would involve direct contact between different parts of the molecule, i.e. heterotypic templating. In the latter case, the same protein stretches will come into contact principally, but homotypic templating would result in two PrPSc molecules with opposite handedness. Alternatively, a head-to-head arrangement could also rely on heterotypic templating, if different parts of the molecule interact with each other. Experimentally, we observed a ~40 Å signal in Fourier-transform analyses of the fibril segments (Figs 5, S4 and S5) that may originate from a dimeric arrangement. The presence of a distinct dimer signal supports a head-to-head and tail-to-tail arrangement, but, ultimately, higher resolution data are needed to distinguish between these dimer options.
In summary, we present data based on cryo-EM analysis that strongly support the notion that GPI-anchorless PrPSc fibrils consist of stacks of four-rung ß-solenoids. Two of such protofilaments intertwine to form double fibrils, in agreement with a recent report based on tomography of negatively stained PrPSc samples [35]. The four-rung ß-solenoid architecture of GPI-anchorless PrPSc provides unique and novel insights into the molecular mechanism by which mammalian prions replicate.
Heterozygous GPI-anchorless PrP transgenic mice (tg44+/-) were developed by Dr. Bruce Chesebro (NIH Rocky Mountain Laboratories, Hamilton, MT, USA) [10]. The mice were crossed to generate homozygous GPI-anchorless PrP animals (tg44+/+) [16] and genotyped by tail DNA analysis using the PCR protocol described in Chesebro et al. [10].
Homozygous GPI-anchorless PrP transgenic mice (tg44+/+) [10,16] were inoculated intra-cerebrally (IC) in the right temporal lobe, at 6 weeks of age, with 20 μl of a 2% RML-infected mouse brain homogenate (BH), prepared in 2X (w/v) PBS. After 365 days post inoculation, the asymptomatic mice were euthanized, the brains were harvested, rinsed in PBS, and stored at -80°C until use.
For the bioassays two groups of six wild-type C57BL/6 mice were inoculated IC with either 20 μl of a 2% GPI-anchorless RML prion BH or with an equivalent amount of purified GPI-anchorless PrP 27–30 resuspended in 5% (w/v) glucose in 2X PBS. Also a new cohort of six wild-type mice was inoculated IC with 20 μl 2X PBS as negative control. Mice were monitored until the appearance of clinical signs, at which time they were euthanized and the brains were removed.
Animal experiments were carried out in accordance with the European Union Council Directive 86/609/EEC, and were approved by the University of Santiago de Compostela Ethics Committee (protocol 15005AE/12/FUN 01/PAT 05/JRR2).
The Kaplan-Meier plot survival analyses of the wild-type mouse groups inoculated with GPI-anchorless RML prion BH and purified GPI-anchorless PrP 27–30 were compared using the Gehan-Breslow-Wilcoxon test.
Immediately after extraction, the brain was fixed in formalin and then sliced into four transversal sections by cutting the brain caudally and rostrally to the midbrain and at the level of the basal nuclei. The sections were dehydrated by immersion in solutions of progressively higher ethanol concentration and, finally, with xylene before being embedded in paraffin. Haematoxylin-eosin was used to stain 4 μm thick sections. Additional sections were mounted on 3-triethoxysilyl-propylamine-coated glass slides for immunohistochemical (IHC) studies.
IHC for the detection of PrPSc was performed as follows. Deparaffinized sections were subjected to epitope unmasking treatments: Immersed in formic acid and boiled at low pH (6.15) in a pressure cooker and pre-treated with proteinase K. Endogenous peroxidases were blocked by immersion in a 3% H2O2 in methanol. Then, the sections were incubated overnight with anti-PrP mAb 2G11 primary antibody (1:100, kindly supplied by Dr. Eoin Monks, University College Dublin, Ireland) and subsequently visualized using the Dako EnVision system K400111/0 (Dako, Glostrup, Denmark) and 3,3’diaminobenzidine as the chromogenic substrate. Additional sections were incubated with a rabbit polyclonal antibody against glial fibrillary acidic protein Z0334, 1:600 (Dako, Glostrup, Denmark) to visualize astrocytic activation. For the glial fibrillary acidic protein detection epitope unmasking treatments were omitted, but the same visualization system was used. As a background control, the incubation with the primary antibodies was omitted.
GPI-anchorless PrP 27–30 was isolated using a slightly modified version of the method of Baron et al. [11]. During the purification, total PrPSc was treated with 10 μg/ml of proteinase K (PK) at 37°C for 1 h. PK treatment yielded a shortened, protease-resistant form termed PrP 27–30, due to its similarity to GPI-anchored PrP 27–30, which has an apparent molecular mass of 27–30 kDa [40]. The final GPI-anchorless PrP 27–30 pellet was resuspended in 100 μl of deionized water and treated with lipase (porcine pancreas lipase, Sigma No. 62300) at 1 μg/ml for 2 h at 37°C. To trap the fatty acids released by the lipase, bovine serum albumin (BSA) was added to a final concentration of 10 mg/ml. The sample was centrifuged at 22,000 g for 20 min, twice, and the pellet containing the PrP 27–30 fibrils was resuspended in 100 μl of deionized water. Finally, fibrils were sonicated with three pulses at a 50% amplitude with a probe ultrasonic homogenizer (Cole Parmer Instrument CO., Chicago, IL, USA), and the purified protein was stored at 4°C or -80°C. The sample purity was assessed by SYPRO Ruby Protein Gel Stain. The yield of GPI-anchorless PrP 27–30 was ~35 μg per mouse brain (BCA protein assay [41]).
SDS-PAGE was performed in 15% polyacrylamide gels. For Sypro Ruby staining, the gel was washed with ultrapure water, fixed for 1 hour with 10% methanol and 7% acetic acid, followed by overnight incubation with Sypro Ruby solution (Lonza, Rockland, ME, USA) at room temperature with gentle agitation and protection from light. For immunoblotting, the gel was transferred to an Immobilon-P PVDF membrane (Millipore, Billerica, MA, USA) and probed with the 3F10 antibody (recognizing residues 137–151) at a 1:5,000 dilution. Peroxidase-labeled anti-mouse antibody was used as the secondary antibody at a 1:5,000 dilution.
Samples were prepared by pipetting a 3 μl sample, mixed with a gold fiducial solution (15 nm gold particle size), onto a freshly glow-discharged Lacey carbon grids (Ted Pella Inc., Redding, CA, USA). The grids were plunge-frozen in liquid ethane in a Vitrobot Mark IV (FEI, Eindhoven, The Netherlands). Cryo-EM data were collected using a Titan Krios microscope equipped with Falcon II direct electron detector (FEI, Eindhoven, The Netherlands), operated at 300 kV. Low-dose imaging conditions with 20 electrons per Å2 were applied. Images were collected at 1–3 μm underfocus. All micrographs were recorded at a pixel size of 1.34 Å per pixel.
Tilt series data were obtained on a Titan Krios (FEI, Eindhoven, The Netherlands), operated at 300 kV on a Falcon II direct electron detector (FEI, Eindhoven, The Netherlands), under low-dose conditions with 50 electrons per Å2. Tomograms of selected areas were obtained from -70 to +70 degrees with a 1.5° tilt increment. The defocus range of the data set was 5–8 μm, with a pixel size of 6 Å. The tilted images were aligned and reconstructed with both the FEI Inspect3D software (FEI, Eindhoven, The Netherlands) and with Tomo3D [42] and the SIRT reconstruction method (30 iterations).
261 fibril widths were obtained from 9 different reconstructed tomograms. Measurements were determined using ImageJ software, following the “Measuring distances between points” manual.
Single particle analysis: Only cryo-EM images presenting clear Thon rings were used for processing. The CTF was determined by CTFFIND3 [43] and phases were corrected with Bsoft [44]. Straight fibril segments were extracted manually from 400 micrographs using EMAN’s boxer program [45]. A total of 1305 non-overlapping segments were picked with a box size of 200 x 200 pixels (26.8 x 26.8 nm). In order to align the segments along the y-axis, a symmetrized average from an iterative alignment using the straightest particle in the set as a reference, was used as a starting template. Subsequently, particles were classified by multivariate statistical analysis implemented in IMAGIC [46]. Also, the aligned particles were cut into single protofilaments, realigned and reclassified, resulting in a new data set of 2610 particles. Finally, the summed amplitude spectra of all class averages was calculated and further analyzed.
Reference-free single particle analysis: Image processing was performed with the IMAGIC-4D software [47]. The same 400 selected micrographs (4096 x 4096 pixels) were normalized by a posteriori camera correction to remove camera artifacts [48] (S6 Fig). Amplitude spectra were used to determine the CTF parameters and perform CTF-correction by phase flipping (using CTF2D-FIND and CTF2D-FLIP programs). The particles were picked from the best patches of normalized and band-pass filtered micrographs by the PICK-M-ALL program, using a single featureless rectangular reference created by smearing a filament along its length. From the selected patches containing clear fibrils, the best 2725 particles were selected based on standard statistical analyses (averages, standard deviations, cross correlation coefficients). Multivariate eigenvector data compression was then applied, followed by automatic unsupervised classification, to create 20 classes. No alignments were applied to avoid reference bias.
3D fibril reconstruction: Suitable fibril images were selected from a set of 1284 cryo-EM images. An individual fibril, presenting at least half a helical turn (180°), was segmented along the helical axis using EMAN’s boxer program [45] into overlapping boxes of 300 x 300 pixels (40.2 x 40.2 nm). The boxes were centered and spaced 1–5 pixels apart along the helical axis. By measuring the repeat distance of the helical fibril, we could estimate the angular orientation of each box in the set. The underlying assumption was that each box represented a different view of identically the same fibril, where each view is rotated and translated according to the helical twist. By assigning the angles to each box in the set, a preliminary 3D reconstruction was generated by back projection. Two-fold symmetry was then imposed. The preliminary 3D reconstruction was refined against other 3D reconstructions generated from the same fibril using different step sizes [29]. SPIDER software was used for the reconstruction [49]. In order to remove the rippling artifact created by the overlapping segments, the procedure continued with a second stage, using IMAGIC software [46]. The refined 3D reconstruction was rotated 90° around its x-axis and sliced along the yz plane. The new set of cross-sectional projections was aligned, centered, and averaged. The averaged density was symmetrized, replicated along the fibril length, and in-plane rotation angles were assigned based on the helical twist. Finally, the collection of 2D projections was used to assemble the 3D volume. Reconstructions were visualized in UCSF Chimera [50]. The thresholds for the cryo-EM maps were fixed based on the fibril widths measured in the raw images.
The monomer height calculation was accomplished by determining the cross-section area (Å2) of each protofilament in Chimera [50] and then calculating the corresponding volume using an experimentally determined molecular mass for the GPI-anchorless PrP 27–30 of 17,148 Da [18], and a mean protein density of 0.903 Da/Å3; calculated based on a HET-s monomer stacked in a HET-s prion fibril ([30] PDB: EMD-2946).
|
10.1371/journal.pgen.0030135 | Unbiased Gene Expression Analysis Implicates the huntingtin Polyglutamine Tract in Extra-mitochondrial Energy Metabolism | The Huntington's disease (HD) CAG repeat, encoding a polymorphic glutamine tract in huntingtin, is inversely correlated with cellular energy level, with alleles over ∼37 repeats leading to the loss of striatal neurons. This early HD neuronal specificity can be modeled by respiratory chain inhibitor 3-nitropropionic acid (3-NP) and, like 3-NP, mutant huntingtin has been proposed to directly influence the mitochondrion, via interaction or decreased PGC-1α expression. We have tested this hypothesis by comparing the gene expression changes due to mutant huntingtin accurately expressed in STHdhQ111/Q111 cells with the changes produced by 3-NP treatment of wild-type striatal cells. In general, the HD mutation did not mimic 3-NP, although both produced a state of energy collapse that was mildly alleviated by the PGC-1α-coregulated nuclear respiratory factor 1 (Nrf-1). Moreover, unlike 3-NP, the HD CAG repeat did not significantly alter mitochondrial pathways in STHdhQ111/Q111 cells, despite decreased Ppargc1a expression. Instead, the HD mutation enriched for processes linked to huntingtin normal function and Nf-κB signaling. Thus, rather than a direct impact on the mitochondrion, the polyglutamine tract may modulate some aspect of huntingtin's activity in extra-mitochondrial energy metabolism. Elucidation of this HD CAG-dependent pathway would spur efforts to achieve energy-based therapeutics in HD.
| Huntington's disease (HD) is a tragic neurodegenerative disorder caused by a CAG repeat that specifies the size of a glutamine tract in the huntingtin protein, such that the longer the tract, the earlier the loss of striatal brain cells. A correlation of polyglutamine tract size has also implicated huntingtin in the proper functioning of mitochondria, the cell's energy factories. Here we have tested the prevailing hypothesis, that huntingtin may directly affect the mitochondrion, by using comprehensive gene expression analysis to judge whether the HD mutation may replicate the effects of 3-nitropropionic acid (3-NP), a compound known to inhibit mitochondria, with loss of striatal neurons. We found that, while mutant huntingtin and 3-NP both elicited energy starvation, the gene responses to the HD mutation, unlike the responses to 3-NP, did not highlight damage to mitochondria, but instead revealed effects on huntingtin-dependent processes. Thus, rather than direct inhibition, the polyglutamine tract size appears to modulate some normal activity of huntingtin that indirectly influences the management of the mitochondrion. Understanding the precise nature of this extra-mitochondrial process would critically guide efforts to achieve effective energy-based therapeutics in HD.
| The CAG trinucleotide repeat in the Huntington's disease gene (HD) is highly polymorphic in humans, with alleles ranging from ∼6 to >100 units encoding a variable polyglutamine tract in huntingtin, a large (>350 kDa) HEAT domain protein [1]. The gene was discovered because alleles over ∼35–37 units, whether inherited in one copy or, in rare cases, two copies, are associated with the onset of Huntington's disease (HD) symptoms, including dance-like movements, cognitive decline, and psychiatric disturbance [1]. This intriguing disease-initiating mechanism, while relatively insensitive to dosage, is exquisitely progressive with allele size, such that the age at onset of HD symptoms is progressively decreased as CAG length is increased [1].
The early HD pathology comprises the selective loss of medium-size spiny neurons in the striatum that forms the HD pathological grading system [2]. The ability of mitochondrial respiratory chain poisons such as succinate dehydrogenase inhibitor 3-nitropropionic acid (3-NP) to produce HD-like striatal-specific cell loss has implied a role for mitochondrial dysfunction in HD pathogenesis [3]. Indeed, numerous studies have demonstrated that deficits in measures of energy metabolism become manifest in presymptomatic and symptomatic HD brain and peripheral tissues [4–11].
Investigations of the early consequences of the HD CAG repeat in human lymphoblastoid cell lines have recently implicated the polyglutamine tract size in huntingtin in modulating cellular ATP/ADP ratio across the entire non-HD and HD range [12]. The longest alleles were associated with the lowest ATP/ADP ratios, while alleles in the non-HD range were associated with progressively higher energy levels [12]. Moreover, consistent with a role for the polyglutamine tract in influencing an intrinsic huntingtin function in energy metabolism, the targeted deletion of the short seven-glutamine tract from murine huntingtin yielded elevated cellular ATP, with early senescence, and improved motor performance in HdhΔQ/ΔQ mice [13].
The shared downstream consequences of the HD CAG tract and 3-NP have implied that the polyglutamine tract in huntingtin, like 3-NP, may directly affect the mitochondrion [14]. Huntingtin is detected throughout the cell, in the nucleus and in the cytoplasm, where it can associate with mitochondria [15], implicating a direct “toxic” interaction [15–17]. However, recent findings, including studies in STHdhQ111/Q111 striatal cells, with a knock-in juvenile onset CAG repeat accurately expressed as endogenous huntingtin with 111 glutamines [18], suggested that mutant huntingtin may influence mitochondrial biogenesis/function by decreasing Ppargc1a transcription [19,20]. This gene encodes peroxisome proliferative activated receptor gamma, coactivator 1 alpha (PGC-1α), a key cofactor for Nrf-1 and other mitochondrial transcription regulators.
To probe the earliest consequences of the HD CAG mechanism, we tested the mitochondrial hypothesis by using unbiased gene expression analysis to monitor the extent to which the accurate expression of the HD CAG repeat, in STHdhQ111/Q111 striatal cells, may reproduce the consequences of 3-NP challenge. The results confirmed a shared downstream energy collapse but did not indict direct 3-NP-like effects of the HD CAG repeat on the mitochondrion. Instead, the data have elevated the candidacy of extra-mitochondrial pathways in huntingtin modulation of energy metabolism.
We and others have demonstrated previously that STHdhQ111/Q111 cells, compared to wild-type STHdhQ7/Q7 cells (expressing endogenous seven-glutamine huntingtin), exhibited mitochondrial energy phenotypes similar to the effects of 3-NP treatment, including decreased mitochondrial respiration [21] and ATP synthesis [12,21,22]. However, STHdhQ111/Q111 cells also have been shown to display phenotypes opposite to the effects reported for 3-NP challenge, such as increased, instead of decreased, levels of the reactive oxygen species (ROS) scavenger glutathione [23], suggesting that 3-NP might not precisely mimic the effects of the HD mutation.
To explore this notion, we examined additional energy phenotypes and observed, as reported for 3-NP challenge, that the HD mutation decreased mitochondrial membrane potential (Figure 1A) and elevated the lactate/pyruvate ratio (Figure 1B) indicative of altered energy homeostasis. Indeed, in human lymphoblastoid cells, this ratio was increased in severity with HD CAG repeat size (Figure 1C), demonstrating that mitochondrial dysfunction is likely to be a consequence of the HD CAG size-dependent mechanism. However, while 3-NP, as reported [24], was associated with increased ROS, as measured by hydrogen peroxide levels, the STHdhQ111/Q111 cells, compared to wild-type cells, exhibited decreased ROS (Figure 1D). Moreover, while 3-NP treated wild-type cells, as expected, displayed decreased succinate dehydrogenase activity, the longer HD CAG repeat expressed in the STHdhQ111/Q111 cells did not significantly change the activity of this respiratory chain component (Figure 1E). Thus, while superficially similar, the energy phenotypes displayed by mutant striatal cells did not in detail recapitulate the effects of 3-NP challenge, implying distinct underlying pathways.
To delineate the effects of the HD CAG repeat and 3-NP treatment, without making a priori assumptions about the underlying biology, we performed global analysis to monitor the expression of the mitochondrial and the nuclear genomes. As sequences for genes located on the mitochondrial genome were not represented on the murine Affymetrix MG 430 2.0 arrays microarrays, which we used to analyze the nuclear genome (below), the expression of nine of the 13 mitochondrial genes encoding respiratory components was assessed using specific RT-PCR assays. The mRNA levels for each of these genes was dramatically reduced in 3-NP treated cells, compared to untreated wild-type striatal cells, whereas mRNA levels did not differ significantly between STHdhQ111/Q111 and wild-type cells (Figure S1). Thus, consistent with unchanged succinate dehydrogenase activity, mutant huntingtin did not reproduce the effects of 3-NP on the mitochondrial genome, implying that the energy deficits in STHdhQ111/Q111 cells might instead stem from altered expression of the nuclear genes that regulate the mitochondrion.
We then performed unbiased analysis of the nuclear genome expression datasets to determine the extent to which the consequences of the HD CAG repeat may mirror the effects of 3-NP challenge. The results of principle component (Figure 2A) and cluster analysis (Figure 2B) of all probes demonstrated that, whereas the replicate datasets were highly related, the wild-type, 3-NP treated, and mutant cell datasets were quite distinct. Thus, rather than giving rise to similar effects, which may differ in magnitude, the consequences of the HD CAG repeat and 3-NP appeared to be fundamentally different. Indeed, at a stringent false discovery rate (FDR) (q < 0.005), approximately the same proportion of all probes (3%) was significantly altered either by the HD CAG repeat mutation (comparing mutant versus wild-type cells) (Table S1) or by 3-NP challenge (comparing 3-NP treated wild-type versus wild-type cells) (Table S2), with the HD CAG yielding some changes of larger magnitude (fold-change) than 3-NP. However, little overlap was detected between the two probe lists (summarized in Figure 2C and 2E), yielding low correlation coefficients between HD CAG and 3-NP changed probes (Figure 2D). Indeed, for the most part, these represented different genes, which are plotted by chromosome in Figure S2. Thus, while the impact, in terms of number and magnitude of changes, was similar, HD mutation did not reproduce the molecular effects of 3-NP.
The mitochondrial hypothesis predicted that both the HD mutation and 3-NP treatment would alter mitochondrial energy genes, prompting an examination of the small fraction (0.18%) of all probes significantly altered by both insults. These represented 83 known genes (plotted by chromosome in Figure S3 and listed in Table S3), which did not highlight the mitochondrion but instead spotlighted decreased cytosolic energy production (Figure 3). More sensitive tests of groups of genes in functional pathways, using the Gene Ontology (GO) biological process (Figure 4A) and sigPathway gene set analysis [25] (Figure 4B), confirmed that the HD mutation and 3-NP were both associated with highly significant decreases in carbohydrate metabolism and, as reported previously [26], lipid (sterol/cholesterol) biosynthesis (Figure 4C). Thus, striatal cells, unlike glia or other cell types, may possess a limited capacity to adjust glycolytic flux in response to changes in mitochondrial ATP synthesis [27], thereby providing a potential explanation for the ability of 3-NP to mimic the early loss of striatal neurons in HD striatum.
Although the notion that the HD mutation and 3-NP might commonly alter the expression of nuclear encoded mitochondrial genes did not appear to be borne out, inspection of the data did reveal significant changes in a few mitochondria-related energy genes in mutant striatal cells, which, notably, were unchanged by 3-NP (Figure 3). Ppargc1a (encoding PGC-1α) was decreased, as reported previously [19]. Mitochondrial components in the transfer electrons from NADH to the respiratory chain (Ndufa12 and Ndufa3) or in the transport of hydrogen ions (Atp6v0e2) were decreased. By contrast, mRNAs encoding the iron sulfur-binding factor of complex III (Uqcr) and a nonenzymatic component of the ATP synthase complex (Atp5j2) were elevated. Moreover, consistent with elevated glutathione [23], the expression of genes that detoxify free radical derivatives (Ldh2, Gss, and Glo1) was increased, suggesting that altered redox state may contribute distinctly to low energy (and lipid) metabolism in mutant striatal cells.
The decrease in Ppargc1a mRNA implied that mitochondrial regulatory transcription factors such as Nrf-1, which are coregulated by PGC-1α, might be expected to improve energy metabolism in mutant cells. Indeed, over-expression of Nrf-1 in STHdhQ111/Q111 cells altered the mRNA levels of known Nrf-1 target genes (Table S4) but, as reported for PGC-1α [19], only mildly improved both cellular lactate/pyruvate ratio (Figure 5A) and cell survival following respiratory chain inhibition (Figure 5B). Thus, coupled with the apparent dearth of changes in mitochondrial factors, the modest effects of boosting Nrf-1 strongly suggested that mutant huntingtin may not directly perturb the mitochondrion.
Consequently, to more rigorously examine this possibility, we used gene set enrichment analysis (GSEA) to specifically determine whether mitochondrial pathways or processes may be perturbed by 3-NP or the HD mutation. GSEA examines, as a group, genes that form a functional pathway, thereby capturing small effect sizes that when considered in single gene analyses may not have reached our stringent statistical threshold. We tested two sets of ∼1,500 nuclear genes that were either GO annotated or predicted by the integrative genomics program MAESTRO [28] to encode mitochondrial products, as well as a third set comprising the group of 902 genes that were common to both lists (Table S5; Figure 6A and 6B). The GSEA results, plotted as enrichment scores in Figure 6C, demonstrated that, compared to wild-type cells, each of the gene sets was highly enriched by 3-NP treatment. In striking contrast, no significant enrichment of any of the mitochondrial gene sets was detected in STHdhQ111/Q111 cells. Thus, these findings, which were consistent with the results of the mitochondrial genome analysis (Figure S1), clearly revealed that the HD mutation did not reproduce the direct effects of 3-NP on the mitochondrion. Therefore, contrary to predictions from decreased Ppargc1a mRNA or interactions of mutant huntingtin with the mitochondrion [14–17,19,20], our results consistently implied that the HD CAG mechanism may primarily influence energy metabolism via extra-mitochondrial cellular pathways.
Since our data supported the view that the processes by which the HD mutation and 3-NP may lead to energy starvation were largely distinct, we reasoned that the pathways that did mediate the effects of the huntingtin polyglutamine tract would be those that were not affected by 3-NP challenge. To support this approach, we tested whether the early presymptomatic consequences of the HD CAG repeat in medium-size spiny striatal neurons in vivo might be recapitulated by accurate expression of the expanded HD CAG repeat in cultured STHdhQ111/Q111 cells or by 3-NP challenge of wild-type cells. Microarray analysis of mRNA from medium-size spiny striatal neurons, obtained by laser capture microscopy (LCM) from post-mortem brain, has been reported [29]. The major class of LCM genes judged to be the most significantly altered by the HD CAG in that study, all with decreased expression, yielded a set of 38 mouse genes (Table S14) that were tested by GSEA in our striatal cell HD CAG and 3-NP datasets. The results demonstrated that this human LCM gene set was significantly decreased in the STHdhQ111/Q111 cells (enrichment score [ES] 0.51, p-value 0.000, FDR q-value 0.180) but was not altered in the 3-NP treated cells (ES 0.38, p-value 0.271, FDR q-value 0.713). Thus, the early molecular consequences of the HD CAG repeat in striatal neurons in human brain also become manifest as a consequence of accurate expression of the expanded repeat in the cultured STHdhQ111/Q111 cells but these were not reproduced by 3-NP challenge. This supported the approach of attempting to identify the processes by which huntingtin may modulate energy metabolism by examination of HD CAG repeat–specific changes.
Therefore, stringent statistical filtering criteria were used to identify those gene sets that were enriched in mutant striatal cells but not in 3-NP treated wild-type cells (Tables S6–S12). Remarkably, a major class of pathways uniquely enriched in STHdhQ111/Q111 cells, summarized in Figure 4A and 4B, pointed to processes implicated in huntingtin's essential normal activities [30–33]: development/morphogenesis (central nervous system, mesoderm, and embryo); growth, cell motility, migration, and locomotion; cell adhesion; neuronal processes; and TGF-β and BMP signaling. A minor class, which denoted increased gap junction channel (connexin), phosphate, and anion transport, may also prove to be related to huntingtin function. These results, therefore, strongly implicated huntingtin normal function in the capacity of the polyglutamine tract to influence mitochondrial function and cellular energy metabolism in STHdhQ111/Q111 cells.
Notably, this finding is consistent with a genetic gain-of-function hypothesis for the mechanism that initiates the HD pathogenic process. The polyglutamine tract size may progressively influence energy homeostasis by increasing some intrinsic huntingtin activity. Alternatively, it may capitalize on an opportunity afforded by huntingtin function to modulate an unrelated cellular component.
What might this normal huntingtin activity be? One possibility, inferred from previous work with huntingtin-deficient cells [34,35], as well as with STHdhQ111/Q111 cells [18], may reflect a role for huntingtin in intracellular iron trafficking. However, huntingtin's normal activities impact a broad range of cellular processes, from vesicle trafficking to the proper regulation of gene transcription [33,36,37]. As a starting point, therefore, we reasoned that worthy candidates for huntingtin modulation of energy metabolism might be found among the transcriptional regulators that most frequently orchestrate the gene changes in STHdhQ111/Q111 cells, but not 3-NP treated cells.
Unbiased transcription factor analysis coupled with GSEA yielded five HD CAG–specific candidates (Figure 7). Interestingly, NRSF/REST, associated with altered expression of genes in STHdhQ111/Q111 cells and in HD brain [37], did not emerge as a candidate, and NRSF/REST target genes or genes with upstream RE1 binding sites were not enriched as a consequence of the HD CAG repeat (unpublished data). The top candidates for HD CAG specific regulators were CAAT-Box (e.g., NF-Y, cEBP, and CTF) and nuclear factor kappa B (Nf-κB), with weaker support for Nrf-1 and Sp-1. Most of these candidates, including Nrf-1, have been shown to exert extra-mitochondrial effects on energy metabolism, affecting, e.g., fatty oxidation, glycolysis, glutathione synthesis, and cell stress-sensing [38–40]. Moreover, by associating exclusively with mutant huntingtin (or fragment), PGC-1α, a coregulator with Nrf-1, Sp-1, and its coregulator TAF4, as well as core components of Pol II transcriptional machinery (TFIID, TFIIF) have been implicated in HD pathogenesis [19,41–43].
However, the mechanism by which the HD CAG modulates cellular energy levels was manifest even at non-HD-causing lengths, in the cadre of normal huntingtin function [12]. Nf-κB/Dorsal/RelB has been linked previously to huntingtin's normal function [44]. Nf-κB was a modifier of huntingtin carboxyl terminal fragment–induced phenotypes in Drosophila [44]. Furthermore, endogenous normal human huntingtin can associate with the p50 subunit of Nf-κB [44], which binds to target gene promoters in a redox-dependent manner [45]. This raises a new hypothesis that merits investigation in a variety of suitable polymorphic HD CAG systems, namely, that huntingtin polyglutamine tract length may influence Nf-κB-mediated signaling, perhaps via redox sensing, either as a cause or a consequence of modulating energy metabolism.
In summary, the results of this study demonstrated that the early (presymptomatic) consequences of the juvenile onset HD CAG allele, accurately expressed in STHdhQ111/Q111 striatal cells, did not in most details mimic mitochondrial respiratory chain inhibitor 3-NP, although both metabolic challenges produced a common response—the collapse of mitochondrial and cytosolic energy processes. This reveals the limited utility of 3-NP lesion models for uncovering the pathways by which the polyglutamine tract in huntingtin may influence energy metabolism or for prioritizing agents that may modify these processes. Indeed, the data uniformly refute the widely held view of a direct mutant huntingtin-specific effect on the mitochondrion. Instead, our data implicate an effect of the polyglutamine tract on some normal activity of the huntingtin protein in extra-mitochondrial energy metabolism, perhaps redox sensing.
It is interesting to note that redox sensing cell signaling, via ROS-dependent pathways, is emerging as a regulator of glucose and lipid metabolism in health, aging, and disease [46]. If huntingtin proves to be involved, the molecules and gene products that modify oxidation-sensitive signaling in metabolic disorders may become candidates for modifying huntingtin-regulated mitochondrial metabolism in HD. By the same token, agents that modify redox sensing signaling in HD may also be of interest in cancer and a variety of complex metabolic disorders.
STHdhQ7/Q7 and STHdhQ111/Q111 cells, generated from striatal primordia of wild-type HdhQ7/Q7 and homozygous mutant HdhQ111/Q111 knock-in mouse embryos, respectively [18], were cultured in DMEM (33 °C, 5% CO2, 10% FBS, and 400 μg/ml G418). Challenge of STHdhQ7/Q7 cells with 3-NP (1 mM) was for 48 h. Previously genotyped human lymphoblastoid cell lines (HD CAG 17/18, 15/42, 21/43, 21/66, 19/70, 41/48, and 45/50) were cultured in RPMI-1640 (37 °C, 5% CO2, and 5% FBS).
Mitochondrial membrane potential was measured by incubation of cells with JC-1 (2.5 μg/ml) for 20 min at 33 °C, followed by flow cytometry (FACScalibur, http://www.bdbiosciences.com/), counting 104 cells in the FL1 (monomer) and FL2 (aggregate) channels and calculating FL2/FL1 ratio.
Succinate dehydrogenase activity, measured using succinate as a substrate and 3-(4,5-dimethylthiazol-2-yl)-5-(3-carboxymethoxyphenyl)-2-(4-sulfophenyl)-2H-tetrazolium as an electron acceptor, was normalized by protein concentration [47]. ROS levels were monitored by determinations of hydrogen peroxide concentration, using a chemiluminescent hydrogen peroxide detection kit (Assay Designs, http://www.assaydesigns.com/). Data were quantified compared to a standard curve, and normalized to cell number. Lactate and pyruvate concentrations in cleared lysates (3% perchloric acid, sonication) were determined by HPLC analysis (Aminex column, Bio-Rad Laboratories, http://www.bio-rad.com/; 0.6 ml/min 0.05 mM H2SO4; UV detection at 210 nm) using appropriate standards. All measurements derive from triplicate experiments, using duplicate samples. Data are given as the mean, +/− one standard deviation.
Reverse transcription PCR (RT-PCR) analysis for the genes listed in Table S5 was performed (Bio-Rad iCycler) with gene-specific primer sets listed in Table S13. The ΔΔCT method was used to calculate gene expression levels, compared to β-actin control [48].
Total RNA (5 μg), isolated from triplicate cell cultures, was converted using SuperScript II reverse transcriptase (Invitrogen. http://www.invitrogen.com) to labeled cRNA, and 25 μg of labeled probe was hybridized to Affymetrix MG 430 2.0 arrays (http://www.affymetrix.com). Expression data was generated and normalized using RMA [49] and significant probes were identified by FDR (q < 0.005) [50]. A selection of genes judged to be significantly altered according to these criteria was tested by semi-quantitative RT-PCR analysis (primer sets in Table S13), with ∼80% (13 of 16 randomly chosen genes) concordant with the microarray results. For pathway analysis, significant (p < 0.05) GO terms in significantly altered probe sets were identified by GO analysis (DAVID 2006) [51] and, using all probes, significantly enriched (q < 0.01) pathways were identified by permutation-based GSEA (sigPathway) [25]. For the GSEA comparison of the mouse and human presymptomatic striatal cell changes, the LCM gene set comprised the 38 most robust LCM genes, all decreased by the HD CAG repeat, drawn from the table in Hodges et al. [29], that could be unambiguously mapped to a corresponding mouse locus. Cluster [52], dChip [53], and Bioconductor [54] were used for clustering and data visualization.
STHdhQ111/Q111 cells were transiently transfected with control vector (pSG5) or Nrf-1 mammalian cell expression plasmid (0.2 μg of plasmid/well) using lipofectamine as described by the manufacturer (Invitrogen). At 48 h post-transfection, lactate/pyruvate ratios were determined, as described above, and viability following 3-NP challenge was measured by MTS cytotoxicity assay (Promega, http://www.promega.com/). Nrf-1 activity was judged by measuring the expression of a selection of Nrf-1 target genes (Table S4), using semi-quantitative RT-PCR analysis with gene-specific primer sets (Table S13).
For significant genes, putative transcription factor binding sites were identified from the MAPPER database [55], by searching 500 bp upstream from the transcription initiation site and then selecting the top ten transcription factors, by score, for each gene. A total of 4,400 hits (358 transcription factors) and 6,684 hits (371 transcription factors) were recorded for HD CAG and 3-NP treatment, respectively, and the frequency for each was then calculated relative to the total number of hits. As a statistical test, the 100 top target genes for each transcription factor that exhibited 2% or higher frequency in each comparison were identified and used in the transcription factor analysis, to build a custom set for gene set enrichment analysis [56].
MIAME compliant microarray data were deposited at the Gene Expression Omnibus (GEO, http://www.ncbi.nlm.nih.gov/geo/) with Accession number GSE3583. |
10.1371/journal.ppat.1001115 | Involvement of the Cytokine MIF in the Snail Host Immune Response to the Parasite Schistosoma mansoni | We have identified and characterized a Macrophage Migration Inhibitory Factor (MIF) family member in the Lophotrochozoan invertebrate, Biomphalaria glabrata, the snail intermediate host of the human blood fluke Schistosoma mansoni. In mammals, MIF is a widely expressed pleiotropic cytokine with potent pro-inflammatory properties that controls cell functions such as gene expression, proliferation or apoptosis. Here we show that the MIF protein from B. glabrata (BgMIF) is expressed in circulating immune defense cells (hemocytes) of the snail as well as in the B. glabrata embryonic (Bge) cell line that has hemocyte-like features. Recombinant BgMIF (rBgMIF) induced cell proliferation and inhibited NO-dependent p53-mediated apoptosis in Bge cells. Moreover, knock-down of BgMIF expression in Bge cells interfered with the in vitro encapsulation of S. mansoni sporocysts. Furthermore, the in vivo knock-down of BgMIF prevented the changes in circulating hemocyte populations that occur in response to an infection by S. mansoni miracidia and led to a significant increase in the parasite burden of the snails. These results provide the first functional evidence that a MIF ortholog is involved in an invertebrate immune response towards a parasitic infection and highlight the importance of cytokines in invertebrate-parasite interactions.
| Schistosoma mansoni, a parasitic blood fluke that causes intestinal schistosomiasis in humans, requires an intermediate host, a freshwater snail of the genus Biomphalaria for its transmission. Infection of the snail triggers marked cellular and humoral immune responses, but the molecular mechanisms of these responses are not well known. We have identified and characterized the involvement of a snail homologue of the cytokine MIF (Macrophage Migration Inhibitory Factor) in the snail immune responses to infection by the parasite. By using biochemical and molecular approaches in combination with in vitro culture and in vivo gene knock down, we have demonstrated the role of snail MIF in the regulation of the snail innate immune system. In particular, MIF regulates the proliferation and activation of the hemocytes, the macrophage-like snail defense cells, and the encapsulation response. This shows for the first time that MIF has a conserved cytokine function in an invertebrate and underlines the interest of the schistosome-snail model in the study of innate immunity.
| Schistosomiasis, the second most widespread human parasitic disease after malaria [1], is caused by helminth parasites of the genus Schistosoma and more than 200 million people in 74 countries suffer from the pathological consequences of this disease [2]. Human infection requires contact with freshwater in which infected snails (the intermediate hosts of schistosomes) have released cercariae larvae that penetrate human skin. The complex interaction between the intermediate snail host and the parasite and in particular between Schistosoma mansoni and the snail generally used for its laboratory maintenance, Biomphalaria glabrata, is of interest both in terms of transmission dynamics, but increasingly as a model for the study of the innate immune response and its evolution.
In order to protect themselves against pathogens, invertebrates use innate immune responses such as wound repair, coagulation, phagocytosis and encapsulation reactions, also used by vertebrates [3]. Major signalling pathways or effector molecules underlying innate immune responses of vertebrates and invertebrates are also shared, as for instance the Toll receptors described for the first time in Drosophila [4] or members of immunoglobulin superfamily such as the FREPs (Fibrinogen-RElated Proteins) in B. glabrata [5].
The need for regulation of cellular immunity and the parallels made between vertebrate and invertebrate innate immunity led to an intense search for invertebrate cytokines [6]. Cytokines specific to invertebrates, such as spätzle in Drosophila [4], astakine in Pacifastacus leniusculus [7], or CCF in Eisenia foetida [8] have been described, but to date, only very few orthologs of vertebrate cytokines have been incontrovertibly identified in invertebrates [9], one of which is Macrophage Migration Inhibitory Factor (MIF). MIF was one of the first mammalian cytokines to be discovered and has been described as a pivotal regulator of innate immune and inflammatory responses in mammals [10]. It was first characterized as a factor derived from activated T cells that inhibited random migration of macrophages [11], [12]. Since the first cloning of a MIF gene [13] many biological activities have been described, including stimulation of cell proliferation through ERK1/ERK2 pathway activation, activation of the response against endotoxin or Gram negative bacteria by upregulation of TLR4 (the signal-transducing molecule of the LPS receptor complex) expression and the suppression of p53-mediated growth arrest in macrophages challenged by LPS [10]. In addition MIF possesses intrinsic tautomerase activity (keto-enol isomerisation of small aromatic substrates such as L-dopachrome methyl ester) that is dependent on the post-translational cleavage of the initiating methionine to expose an N-terminal proline residue [14].
Interestingly, MIFs have been characterized in a wide variety of parasites, including nematodes and protozoans [15], [16] but the role of the cytokine has been mainly studied in the context of the host-parasite relationship with the emphasis on the effect of parasite MIF on the host immune response. For instance Ancylostoma MIF has been shown to bind to the human MIF receptor [17] and recombinant Brugia MIF induces the release of cytokines (IL-8, TNFα) from human macrophages [18]. Similarly, Plasmodium MIF is thought to influence the host immune response and the course of anemia during infection [16]. MIFs have also recently been identified in two species of mollusks, disk abalones [19], but currently, nothing is known about the role of MIF from the invertebrate host during an immune response to a pathogen. Strikingly, an exhaustive search of the S. mansoni genomic sequences (AB-G, unpublished) failed to find any MIF signature sequences These in silico findings are consistent with the in vitro work of others describing the absence of MIF homologs in parasitic trematodes [20].
The discovery in B. glabrata of a potential cytokine-like molecule displaying significant sequence similarity to MIF [21], raised the question of its potential involvement in the regulation of the snail immune response to parasite infection. In this report, we demonstrate that the MIF protein from B. glabrata (BgMIF) is expressed in circulating immune defense cells (hemocytes) of the snail as well as in the B. glabrata embryonic (Bge) cell line that has hemocyte-like features. We show that recombinant BgMIF (rBgMIF) possesses the conserved tautomerase enzymatic activity of the MIF family, induces cell proliferation (correlating with ERK phosphorylation) and inhibits NO-dependent, p53-mediated apoptosis in Bge cells. Moreover, knock-down of BgMIF in Bge cells inhibits the in vitro encapsulation of S. mansoni sporocysts and this correlates with an inhibition of p38 phosphorylation in these cells. Finally, in whole snails, we demonstrate the involvement of BgMIF in the snail anti-parasitic response towards S. mansoni. Furthermore, the tools developed here pave the way toward a better understanding of the complex interactions between S. mansoni and its molluscan snail host.
Alignment of MIF peptide sequences (Figure 1A) shows that BgMIF contains the N-terminal catalytic proline (Pro2) that is exposed by cleavage of the initiating methionine and is essential for tautomerase activity (see below and [14]). With 31% sequence identity to human MIF, BgMIF is less conserved than MIFs from two other mollusks, the bivalve abalones, Haliotis diversicolor sextus (39%) and Haliotis discus discus (35%). Several invariant active site residues [15] are conserved, including Lys32 and Ile64. The conserved Val106 residue is substituted by a Cys in BgMIF or by Leu in MIF from Ixodes scapularis, thus maintaining the presence of a hydrophobic residue at this position (Figure 1A). To further investigate the relationship between BgMIF and other MIFs, we performed a phylogenetic analysis (using two different analyses with similar results: see Methods) on selected vertebrate and invertebrate proteins (Figure 1B). The phylogeny of selected MIFs proved to be complex with numerous small clades and no strong relationship with taxonomy. Although BgMIF is clearly grouped in the phylogenetic tree with nematode MIF2 sequences [15], it is not closely related to other mollusk MIFs (Figure 1B).
A hallmark of all MIF family members is the enzymatic tautomerase activity; we expressed it as a recombinant protein (rBgMIF) in E. coli together with a site-directed mutant (rBgMIFP2G), in which the N-terminal Proline (Pro2) was substituted by Gly. We used rBgMIF and rBgMIFP2G to perform a tautomerase assay with mouse MIF (rMmMIF) as a positive control and L-dopachrome methyl ester as a substrate. The results (Figure 2) showed that rBgMIF displayed tautomerase activity comparable to that of the mouse MIF protein and that, as expected; the mutant rBgMIFP2G did not have any detectable activity. Therefore, as in all MIF family members, Pro2 is required for enzymatic activity of BgMIF. In addition we tested the inhibition of the tautomerase activity using the MIF antagonist ISO-1 a specific inhibitor of mammalian MIF [22]. rBgMIF treated with 100 or 200 µM of ISO-1 (Supplementary data Fig S1) was inhibited by more than 95% at both doses.
In contrast to most cytokines, MIF is constitutively expressed and stored in intracellular pools. MIF secretion is induced by inflammatory stimuli such as endotoxin (LPS) or tumor necrosis factor (TNF-α), as well as by hormones [10]. MIF is expressed by defense cells such as macrophages [23], monocytes, neutrophils, dendritic cells and other cell types in tissues in contact with the host's natural environment [10]. We examined tissue specific expression of MIF in B. glabrata snails by western blotting of protein extracts from various snail organs using an antiserum raised against two peptides derived from the BgMIF sequence. This antiserum was shown to recognize native BgMIF (Figure 3A). A single band corresponding to BgMIF was found in all tissues tested, including the albumen gland, digestive tract, heart (hematopoietic organ) hepatopancreas, and foot (Figure 3A). In order to confirm the presence of BgMIF in B. glabrata hemocytes, we performed both western blotting and immunolocalisation analyses. BgMIF was detected in hemocyte lysates (Figure 3A) and immunolocalized in the cytoplasm of hemocytes. BgMIF was found to be more abundant in well spread hemocytes, termed granulocytes, than in unspread hemocytes or hyalinocytes [24], [25] (Figure 3B–D).
ELISA tests performed with anti-BgMIF serum allowed us to detect BgMIF in plasma (cell-free hemolymph) and to demonstrate that the amount of BgMIF in plasma progressively decreased during infection by S. mansoni (34% of decrease at 48 h post-infection) (Supplementary data Figure S2).
Bge cells represent the only existing molluscan cell line and display hemocyte-like immune functions [26]. They have previously been described to share with hemocytes a fibroblastic origin and the ability to recognize and phagocyte or encapsulate foreign material including larval trematodes [26], [27]. In order to assess the pertinence of this cell line as an in vitro system for the analysis of BgMIF activity and function, we first searched for the presence of BgMIF in Bge cells. BgMIF was readily detectable in these cells by western blotting (Figure 3A) and immunolocalization showed that, as in hemocytes, BgMIF could be detected in the Bge cell cytoplasm (Figure 3E–G). In order to determine whether Bge cells could also release BgMIF protein upon immune stimulation, as observed in vitro for mammalian macrophages [23], we cultured Bge cells in the presence of S. mansoni excretory-secretory products (ESP) that have been shown to modulate gene expression in these cells [28]. BgMIF secretion was induced by ESP at 30 µg/mL (protein) with an apparent maximum at a dose of 120 µg/mL (Figure 3H).
We have also carried out a western blot of ESP from sporocysts with the anti-BgMIF antiserum and as expected in view of the absence of MIF signature sequences from the S. mansoni genome, no cross-reactive bands were detected (data not shown).
In mammals, MIF stimulates the proliferation of quiescent fibroblasts in an “ERK sustained activation” dependent manner [29]. We determined whether BgMIF promotes cell proliferation by stimulating quiescent Bge cells with rBgMIF for 24 h and measuring BrdU incorporation. Purified rBgMIF stimulated cell proliferation in a dose dependent manner from a concentration of 50 ng/ml, with a maximum incorporation rate at 100 ng/ml (Figure 4A).
ERK-MAPK pathway activation is associated with mammalian MIF induced cell proliferation. To investigate whether BgMIF activated ERK, quiescent Bge cells were treated with rBgMIF and the cell lysates examined for ERK phosphorylation by Western blot analysis using phospho-specific anti-ERK antibodies. MIF induced phosphorylation of a B. glabrata ERK homolog in a dose and time-dependent fashion (Figure 4B). ERK phosphorylation was detected as early as 2 h and was sustained for at least 24 h as previously described for ERK in the NIH/3T3 fibroblast cell line (Figure 4B) [29]. In addition, U0126 the specific inhibitor of MEK (mitogen-activated protein kinase/ERK kinase), the upstream kinase of ERK [30], prevented the stimulatory effect of BgMIF on Bge cell proliferation (Supplementary data Figure S3) and ERK phosphorylation (data not shown), further indicating that MIF can induce proliferation via the ERK1/ERK2 pathway.
MIF was found to inhibit NO-induced intracellular accumulation of p53 and, therefore, p53-mediated apoptosis in macrophages [31]. To investigate whether BgMIF inhibits apoptosis induced by NO accumulation, Bge cells were treated with the NO donor SNGO and different concentrations of rBgMIF. The proportion of apoptotic cells, labeled by the TUNEL method, was quantified using FACS analysis. As in mammalian macrophages, SNGO induced a significant level of apoptosis in Bge cells that was decreased in a dose-dependent manner by the addition of BgMIF (Figure 5A) and except for the lowest concentration of rBgMIF (25 ng/ml), the decrease in the percentage of apoptotic cells was statistically significant (Figure 5B). In mammalian macrophages, it has been shown that NO treatment is associated with a coordinate increase in the phosphorylation of p53 on Ser15, and that immunoblotting for phosphorylated p53 is a sensitive way of detecting the influence of MIF on intracellular p53 [31]. Examination of B. glabrata ESTs and genome sequences have allowed us to characterize a p53 ortholog (GenBank accession number: GU929337). We therefore examined whether inhibition of apoptosis in Bge cells treated with rBgMIF could be related to a decrease in NO-induced p53 accumulation in these cells. Western blot analysis of cell lysates using a phospho-specific (Ser15) anti p53 antibody showed that rBgMIF inhibited p53 phosphorylation in Bge cells and suggested that this mechanism participated in the suppression by rBgMIF of NO-induced apoptosis (Figure 5C).
We have demonstrated that Bge cells secrete BgMIF when they are incubated with S. mansoni ESP (Figure 3H). In these experiments, we additionally observed that Bge cells aggregated and changed their form upon ESP induction (Figure 6A Ctrl+ESP) like mammalian macrophages induced by LPS [32], nevertheless this phenotype had not previously been described for B. glabrata cells and was not due to contaminating endotoxin in the ESP preparation (see Methods). In order to determine whether this aggregative behavior was regulated by BgMIF, we used RNAi to knock-down (KD) its expression in Bge cells, using dsRNA against BgMIF (dsMIF) or dsRNA against luciferase (dsLuc) as an unrelated control. The efficiency of BgMIF KD was confirmed by the marked decrease (70%) of BgMIF transcripts observed after a 3 day incubation with dsMIF, as compared to incubation with dsLuc (Figure 7A). When S. mansoni ESP (120 µg/mL) was added to cells treated with dsRNA, aggregation was observed in dsLuc (Figure 6B dsLuc+ESP) but not in dsMIF treated cells, which remained well-individualized, with numerous round and unspread refringent cells (Figure 6B dsMIF+ESP), suggesting that BgMIF is involved in the regulation of Bge cell activation induced by parasites. S. mansoni ESP have been shown to stimulate the p38 MAPK signaling pathway in Bge cells [33] manifested by the phosphorylation of Bgp38. We therefore tested the increase in phosphorylation of Bgp38 in response to the incubation with ESP. We detected a rapid activation of Bgp38 after 15 min in dsLuc treated cells, while in dsMIF treated cells p38 was not activated (Figure 6C).
We next determined the effect of BgMIF-induced cell activation using the in vitro model of S. mansoni sporocyst encapsulation by Bge cells [27], [34]. To Bge cells treated with dsMIF or dsLuc for 72 h, we added 48 h in vitro-transformed sporocysts and followed interaction of Bge cells with sporocysts for a further 72 h. Control (as well as dsLuc treated) Bge cells (Figure 8A) readily migrated towards and encapsulated the sporocysts as previously observed [34] but dsMIF-treated cells showed a markedly reduced ability to encapsulate the sporocysts (Figure 8A). The proportion of encapsulated sporocysts was indeed significantly reduced (p<0.05) in dsMIF treated Bge cells compared to dsLuc treated cells or untreated control cells (Figure 8B).
Since BgMIF promotes cellular responses to immune stimulation in vitro, we examined its role in the activation of hemocytes in B. glabrata snails confronting a parasitic infection. We first analyzed the circulating hemocyte population in non-infected versus infected snails, using flow cytometry based mainly on size (forward scatter-FSC) and granularity (side scatter-SSC) dot plot distribution. In non-infected snails the content of circulating hemocytes was shown to be very heterogeneous and we could discriminate two subpopulations, R1 (small and medium hemocytes) and R2 (large hemocytes) (Figure 9A). 24 h following infection, the population of circulating hemocytes showed a marked reduction in the R2 subpopulation of large cells or granulocytes (Figure 9A), which together with hyalinocytes, make up the heterogenous cell population present in healthy snails [25]. This decrease in circulating granulocytes is linked to their migration toward the tissues invaded by miracidia [35], [36].
In order to determine whether these cellular changes were regulated by BgMIF, we performed RNAi KD in whole snails by microinjecting 15 µg dsMIF or dsLuc into the cardiac sinus. BgMIF expression was monitored three days after dsRNA injection. We observed a decrease in BgMIF transcripts and protein in both whole snails and circulating hemocytes (Figure 7B–7C) treated with dsMIF, as compared to dsLuc treated animals. Next we infected control and dsMIF or dsLuc-treated snails by S. mansoni miracidia and analyzed the circulating hemocyte content 24 h after infection. Compared to dsLuc controls, dsMIF-treated infected snails exhibited a hemocyte profile similar to that found in non-infected snails (Figure 9B). These results corroborated the data obtained in vitro with Bge cells and further supported a role for BgMIF in the hemocyte response during a parasite infection of B. glabrata.
Finally, we tested the effect of BgMIF silencing of the snails on the level of infection observed with S. mansoni miracidia. We observed that dsMIF treated snails have significantly more parasites than dsLuc treated and control, untreated snails (Figure 10). These results further show that BgMIF is essential for the control of the immune response of snail against parasite infection.
Invertebrate immune systems have now become a major research focus for investigating broader questions such as the diversity of immune responses including those against parasitic and viral infections [37], [38], processes involved in the immunity of lophotrochozoan invertebrates, or the question of cytokine-dependent regulatory processes [9], [39]. Here we investigated the role of MIF, a putative ortholog of the vertebrate cytokine in the immune response of a lophotrochozoan invertebrate, the gastropod Biomphalaria glabrata towards its natural parasite Schistosoma mansoni.
The BgMIF protein sequence shares sequence and structural homologies with other members of the MIF family, including residues that are invariant across the whole MIF family (Asp9, Pro 56 and Leu88), or involved in the tautomerase enzymatic activity (Pro2, Lys32, Ile64). In addition, analysis of the secondary structure shows that the BgMIF protein is composed of four α-helices and four β-sheets as for other MIF family members. Results obtained with the recombinant BgMIF proteins indicated that BgMIF has a conserved dopachrome tautomerase enzyme activity dependent on the Pro2 residue and the results obtained with ISO-1, which interacts with the catalytic active site residues of mammalian MIF [22], show that the catalytic active site of BgMIF is conserved (even though residues Tyr 95 and Asp 97 in human MIF are respectively Val (as in the tick I. scapularis, Figure 1A) and Lys (as in the abalone MIFs) in BgMIF). The requirement for the enzymatic activity of MIF proteins for their biological activity has not been established and the use of a tautomerase-null MIF gene knock-in mouse model indicated that it was not involved in the growth-regulatory activity of the cytokine [40]. However, it does appear that the catalytic site could be important for pro-inflammatory activity, but the reason for this is still unknown. One possibility is that the catalytic site of MIF affects its binding to the MIF receptor, CD74, and its activation [41]. It has also been suggested that the tautomerase activity may be a vestige of a role in the invertebrate melanotic encapsulation response against microbial invasion, reflecting an ancestral role of the protein in the innate immune response [42]. In the context of BgMIF the significance of this tautomerase activity remains unknown as is the hypothetical interaction of the catalytic site with an as yet uncharacterized BgMIF receptor. On the other hand, the immunological relevance of this activity does not seem to be related to melanotic encapsulation, since melanization has not been described for B. glabrata.
In contrast to most cytokines, MIF is constitutively expressed and stored in intracellular pools and does therefore not require de novo protein synthesis before release into the extracellular milieu. These features provide MIF with the capacity to be released immediately and to act as an effector molecule regulating innate immunity [10]. Macrophages, which represent the first cellular barrier of defense towards pathogen invasion, are an important source of MIF protein, but MIF is also expressed in the tissues that are in contact with the host's natural environment. In this study we showed that BgMIF protein was constitutively present in all the snail tissues tested. In addition we found BgMIF protein in the cytoplasm of hemocytes. It was more abundant in the granulocytic forms that are involved in defence responses including phagocytosis or encapsulation of pathogens, than in the cytoplasm of hyalinocytes. In addition we have demonstrated the presence of BgMIF protein in the plasma of snails and its decrease after infection by S. mansoni. This decrease correlates with the migration of granulocytes towards infected tissues. We further found BgMIF protein in the cytoplasm of Bge cells that share hemocyte characteristics. Bge cells also secreted BgMIF protein in the presence of Schistosoma released products, thus validating the use of these cells in the bio-assays we developed to assess the biological activities of recombinant BgMIF.
When they enter B. glabrata snails, S. mansoni miracidia are readily located by the hemocytes and trigger marked cellular and humoral responses involving both migration of hemocytes toward the site of infection and an increase in hemocyte production [43]. In addition, reactive nitrogen and oxygen species (RNS and ROS) produced by the hemocytes can damage miracidia/newly-transformed sporocysts [38], but at the same time they can also induce hemocyte apoptosis. Important biological activities of MIF are likely to play a key role in this immune response, including the activation of MAPK such as ERK1/2 and the inhibition of p53-mediated apoptosis [10]. Although the MAPK pathways have only been partially characterized in mollusks, previous studies on B. glabrata have documented the involvement of ERK1/2 and other MAPK family members in signaling events leading to cellular immune responses in this snail [33], [44], [45], [46]. Using Bge cells as an in vitro model, we demonstrate in this work the induction of cell proliferation by rBgMIF in a dose-dependent manner. This proliferation was correlated with an increase in phosphorylated ERK in Bge cells as early as 2 h that was sustained for at least 24 h. This activity seems directly linked to the proliferation of the defense cells in response to immune stimulation.
The inhibition of NO-induced intracellular p53, which in turn inhibits apoptosis, is a well-documented effect of mammalian MIF that leads to a sustained proinflammatory function in macrophages exposed to LPS [31]. We showed that incubation with rBgMIF significantly reduced the apoptotic response of Bge cells, induced by SNGO as an NO-donor. The inhibition of apoptosis was accompanied by a reduction in the amount of phosphorylated Bgp53, the p53 ortholog recently identified in B. glabrata snails. These data further support the conservation of essential functions of BgMIF that enable it to regulate the immune response in this invertebrate.
BgMIF thus presents conserved activities of the MIF family and in order to address the question of its involvement in the innate immune response we first determined its role in the interaction of Bge cells with the sporocyst larvae of S. mansoni. We have optimized the KD of BgMIF transcripts and showed for the first time that the reduction of protein level was related to a resulting phenotype in vitro. When Bge cells were stimulated with S. mansoni ESP, they acquired an activated phenotype (aggregation) that was no longer observed in cells treated with dsMIF. In addition we showed that this aggregative phenotype was correlated with phosphorylation of Bgp38, an activation known to promote cell adhesion [33]. Using the in vitro co-cultivation system of Bge cells and in vitro-transformed sporocysts [27], we observed that the KD in BgMIF expression led to fewer Bge cells migrating towards and encapsulating the sporocysts, resulting in a lower percentage of encapsulated sporocysts than observed in control conditions. These results suggest that BgMIF is intimately involved in the response of the snail to parasites.
We next performed BgMIF KD in vivo using the microinjection of dsRNA previously described for B. glabrata snails [47] and analyzed its consequences on the hemocyte population of snails exposed to infection by S. mansoni miracidia. After 24 h following infection, the population of circulating hemocytes in non-interfered snails showed a marked reduction in the number of large cells (granulocytes), which, together with hyalinocytes, form the heterogeneous hemocyte population present in healthy snails [25]. This decrease in circulating granulocytes is due to their mobilization by the parasites and their migration towards the invaded tissues [35], [36]. When we examined the hemocyte populations of infected dsMIF-treated snails, we did not observe such a reduction in the number of granulocytes, suggesting that BgMIF is necessary for hemocyte activation during the in vivo response to the parasite. This change in the behavior of hemocytes was accompanied by a significant increase in the number of sporocysts establishing in dsMIF-treated snails, underlining the importance of MIF in regulating the innate immune response toward the parasite. The absence of hemocyte activation may be due to the lack of a signal generated by BgMIF released under normal conditions. It has been demonstrated that MIF deficient macrophages are hyporesponsive to stimulation by LPS or Gram-negative bacteria stimulation and that this is due to TLR4 downregulation [48]. In dsRNA-treated snails we also showed that the absence of hemocyte migration correlated in these cells with a down regulation of a Toll receptor ortholog, BgToll1, which we have recently identified (unpublished data). These data suggest that the BgToll1 expression may be regulated by BgMIF and that BgMIF facilitates the activation of hemocytes and their migration towards invaded tissues.
Taken together, the results presented here demonstrate the involvement of BgMIF in the innate immunity of B. glabrata. This is the first functional study of a molecule involved in the regulation of the anti-parasite response in B. glabrata, and the tools developed here pave the way towards a better understanding of the complex interactions between medically important helminths and their molluscan snail hosts.
All animal experimentation was conducted following the Nord-Pas de Calais Region and the Pasteur Institute of Lille guidelines for housing and care of laboratory animals and performed in accordance with institutional regulations after review and approval by the Nord-Pas de Calais Region (Authorization No. A59-35009) and Pasteur Institute (Authorization No. AF/2009) Animal Care and Use Committees.
Adult (6–10 mm in diameter) B. glabrata snails (albino strain), were raised in pond water and fed ad libitum. A Puerto-Rican strain of S. mansoni was maintained by passage through B. glabrata snails and Mesocricetus auratus. Miracidia were isolated from infected hamster livers and maintained in complete Chernin's balanced salt solution [49] (CBSS supplemented with 1 mg/ml glucose and 1 mg/ml trehalose) for 48 h to achieve in vitro transformation into mother sporocysts as described in [50]. Mother sporocysts and/or excretory-secretory products (ESP)-containing CBSS were then collected and used. The B. glabrata embryonic (Bge) cell line (ATCC CRL 1494; Rockville, MD), was maintained at 26°C under normal atmospheric conditions in complete Bge medium [51], supplemented with 10% fetal bovine serum (FBS; Sigma), and antibiotics (100 U/ml penicillin G; 0.05 g/ml streptomycin sulphate, 25 µg/ml amphotericin B, Sigma).
Total RNA and protein from individual snails was extracted using the TRIZOL reagent (Invitrogen) according to the manufacturer's instructions. Total RNA from Bge cells was extracted using the Rneasy Mini kit (Qiagen) according to the manufacturer's instructions. For hemocytes the collected hemolymph [52] was divided in two tubes for extraction of protein and for extraction of RNA. For cDNA synthesis, RNA from whole snails (1 µg) and Bge cells (0.1 µg) were used for reverse transcription using SuperScript III reverse transcriptase (Invitrogen) and the oligo(dT)20 primer.
A partial cDNA sequence (EST GenBank accession number: CK989824[21]) was used to design specific primers and perform 5′ and 3′ RACE amplification (SMART RACE cDNA Amplification kit, Clontech) according to the manufacturer's instructions. The complete BgMIF coding sequence was then amplified using primers containing XhoI and XbaI restriction sites respectively (see Supplementary data Table S1 for primer sequences). The PCR products were digested and cloned into the bacterial expression vector pET303 Ct-His (Invitrogen) (BgMIF construct). A BgMIF mutant construct (P2G pET303 Ct-His) was generated by site directed mutagenesis using primers (Supplementary data Table S1) encoding glycine instead of proline after the initiating methionine (BgMIFP2G construct). Sequence alignments and analysis were carried out using the DNAStar Lasergene programme package and the BioEdit v7.0.1 package (http://www.mbio.ncsu.edu/BioEdit/bioedit.html).
For phylogenetic analysis, multiple amino acid sequence alignments were performed using MUSCLE [53] or 3DCoffee [54]. The maximum likelihood tree was obtained with PhyML 3.0 [55] or MrBayes [56] at LIRMM (http://www.phylogeny.fr/) using the WAG model of substitution with four substitution rate categories and estimated gamma shape parameter and proportion of invariant sites. Branch support values were based on 500 bootstrap replicates with PhyML or 100000 replicates for MrBayes.
Recombinant C-terminally His-tagged full-length rBgMIF and rBgMIFP2G fusion proteins were expressed using the pET303 Ct-His vector in E. coli BL21 (DE3) pLys strain. One liter of bacterial culture was grown to an OD600 nm of 0.4 and induced by addition of isopropyl β-1-D-thiogalactopyranoside to a final concentration of 0.4 mM. After 3 h at 30°C, cells were harvested, lysed and purified with Ni-NTA agarose resin (Qiagen) according to the manufacturer's instructions. Briefly, cells were resuspended in lysis buffer (50 mM NaH2PO4, 300 mM NaCl, 10 mM imidazole) and disrupted by successive freeze-thaw cycles in liquid nitrogen. The soluble protein fraction was mixed with Ni-NTA agarose and incubated under agitation for 1 hour at 4°C. The resin was then washed (50 mM NaH2PO4, 300 mM NaCl, 20 mM imidazole) and finally tagged proteins were eluted with 50 mM NaH2PO4, 300 mM NaCl, 200 mM imidazole. The purified protein was dialyzed against endotoxin-free PBS overnight and the content of remaining endotoxin was measured with Limulus Amoebocyte Lysate (Cambrex). Recombinant proteins used in the bioassays contained less than 200 pg endotoxin/mg of protein.
Tautomerase activity was measured using a D-dopachrome tautomerase assay as described previously [14], [22]. Briefly, a fresh solution of D-dopachrome methyl ester was prepared by mixing 4 mM L-3,4-dihydroxyphenylalanine methyl ester with 8 mM sodium periodate for 5 min at room temperature that was then placed on ice 20 min before use. Activity was determined at room temperature by adding D-dopachrome methyl ester to a cuvette containing 50 mM rBgMIF, BgMIFP2G or a commercial mammalian MIF (mouse MIF), rMmMIF (R&D systems), in 25 mM potassium phosphate buffer pH 6.0 and, 0.5 mM EDTA. For the inhibition assays the MIF inhibitor, (S,R)-3-(4-hydroxyphenyl)-4,5-dihydro-5-isoxazole acetic acid methyl ester (ISO-1, Merck) was dissolved in Me2SO at various concentrations and added to the cuvette with rBgMIF prior to the addition of the dopachrome. The decrease in absorbance at 475 nm was monitored for 5 min using a UV/visible Spectrophotmeter Ultraspec 2100 (Amersham).
Antibodies used in this study were as follows: anti-Actin (Abcam), anti-phospo-p53Ser15, anti-phospho-p44/42 MAPK (Erk1/2) (Thr202/Tyr204), anti- p44/42 MAPK (Erk1/2) and anti-Phospho-p38 MAPK (Thr180/Tyr182) (Cell Signaling). An anti BgMIF antiserum was produced in a rabbit using the Ac-HKDITEIASEFLQKSQK-amide, Ac-VRVPALFCQFDGHLHGH-amide peptides and the polyclonal sera were purified using a peptide linked resin column (Proteogenix). For western blot analysis cells or snails were lysed in Tris-buffered saline (50 mM Tris-Base, 150 mM NaCl, pH 7.5) containing 1% NP-40, 0.5% deoxycholic acid, 0.1% SDS, 2 mM EDTA, and 1 mM PMSF), cellular debris was pelleted and the supernatants were adjusted for protein concentration and diluted with reducing SDS-PAGE sample buffer. For BgMIF analyses the total protein extracts were separated by SDS-PAGE in pre-cast 16% tricine gels (Invitrogen) and transferred to a PVDF membrane with 0.2 µm pore size (Millipore). For actin, ERK, p38 and p53 analyses the total protein extracts were separated by SDS-PAGE Tris-glycine gels and transferred to a PVDF membrane with 0.45 µm pore size (Millipore). Western blot analyses were then performed using the SNAP id system (Millipore) according to the manufacturer's instructions.
For real-time PCR analyses total RNA (1 µg of RNA from a 5 snail pool and one tenth of the RNA obtained from 2×105 Bge cells or collected haemocytes) was reverse transcribed using SuperScript III reverse transcriptase (Invitrogen). For Q-PCR analyses, cDNAs used as templates were amplified using the SYBR Green Master Mix (Invitrogen) and the ABI PRISM 7000 sequence detection system (Applied Biosystems). Primers (Sup. Table 1) specific for B. glabrata ribosomal protein S19 (Genbank accession number CK988928;[21]), and BgMIF, were designed by the Primer Express Program (Applied Biosystems) and used for amplification in triplicate assays.
The organ distribution of BgMIF protein was determined in adult snails. Organs (albumen gland, hepatopancreas, foot, heart and digestive tract) were excised, sectioned and homogenized at 4°C with Tris-buffered saline. The cellular debris was pelleted and the supernatants were adjusted for protein concentration, diluted with reducing SDS-PAGE sample buffer and 15 µg of total protein was analyzed by SDS-PAGE and western blotting as described above.
For the measurement of BgMIF protein contained in the plasma (cell free hemolymph) after infection we used an indirect ELISA protocol. Briefly, the cell hemolymph for five snails was pooled and centrifuged to pellet the cells. Wells of a PVC microtiter plate were coated with 100 µl of hemolymph diluted by half in PBS and plates were incubated for 2 h at room temperature. Then, after the coating solution was removed, plates were washed three times by filling the wells with 200 µL of PBS. The solutions or washes were removed by flicking the plate over a sink. The remaining protein binding sites in the coated wells were blocked by adding 200 µL of blocking buffer (5% bovine serum albumin (BSA) in PBS) per well. Plates were covered and incubated for 2 h at room temperature. Plates were then washed twice with PBS, and 100 µL of anti-MIF antibody (from rabbit, diluted 1/500, Proteogenix) diluted in blocking buffer was added to each well. Plates were covered and incubated overnight at 4°C. Subsequently, plates were washed four times with PBS, and 100 µL of conjugated secondary antibody (antirabbit, diluted 1/5000, Jackson Immuno Research) diluted in blocking buffer was added to each well. Then, plates were covered and incubated 2 h at room temperature. After four washes, the horseradish peroxidase (HRP) activity was measured using the colorimetric substrat TMB (3,3–5,5 –Tetramethylbenzidine) blue substrate (Roche Applied Science). A standard with rBgMIF ranging between 0–230 ng/ml was used. The measurements were made twice in triplicate with two different infection experiments using a microplate reader MRXII (Dynex Technologies) at 450 nm; the haemolymph was pooled before infection and 6, 24 and 48 h post infection.
For immunolocalisation assays, circulating haemocytes or Bge cells were extracted or cultured as described above and allowed to adhere to glass slides, washed with PBS, fixed in 4% paraformaldehyde for 10 min and permeabilised by a 4 min treatment with Triton X-100 at 0.1%. Slides were saturated for 90 min with PBS containing 1% bovine serum albumin (BSA) and normal goat serum (1/50) at room temperature (RT). This blocking step was followed by an overnight incubation with rabbit anti-BgMIF polyclonal serum (diluted at 1/100 in PBS-BSA 1%). After three washes the slides were incubated with goat anti-rabbit Alexa Fluor 488 IgG (1/500 in PBS-BSA 1%, Molecular Probes) for 2 h at room temperature (RT). Slides were then stained with Hoechst 33342 and rhodamine-labeled phalloidin (1/1000 in PBS, Sigma) for 10 min at RT, washed and mounted with Fluoromont G (Interchim). For control slides, anti-BgMIF polyclonal serum was incubated with the peptides used as immunogens for 1 h at RT and the slides were then treated as described above. Samples were analysed by confocal microscopy using a LSM 710 inverted microscope (Zeiss).
All the confocal imaging was performed with a LSM710 microscope (Zeiss) and a Plan Apochromat objective (63×1.4 NA oil immersion). The associated software (Zen 2008) enabled the adjustment of acquisition parameters. The rhodamine (red) signal was excited at 561 nm and emission was collected from 570 to 700 nm. The Alexa488 (green) signal, in contrast, was excited at 488 nm and emission was collected from 490 to 530 nm. The nuclear Hoechst dye signal was excited at 405 nm and emission was collected from 410 to 470 nm. Fluorescent signals were collected sequentially, with a 4 lines average, and resulting images are 2048×2048 (or 1024×1024) pixels in size. By setting the photomultiplier tubes and the pinhole size (1 AU) correctly, there was no signal bleed-through. The images were treated with ImageJ (NIH) and Photoshop CS3 (Adobe).
The protocol used was adapted from [29]. Briefly Bge cells (2×104 cells/well) were cultured until semi-confluent in 96-well plates in complete Bge medium. The cells then were synchronized by culture in 0.5% FCS-containing Bge medium overnight. The medium was then replaced by fresh medium (control condition) or medium containing different concentrations of rBgMIF. After incubation with rBgMIF the cells were pulsed with 10 µM of BrdU (Sigma) for 2 h and the proliferation was measured by ELISA method as described in [57]. In order to test the effect of inhibition of the ERK pathway, cells were treated with the MEK inhibitor U0126 (Cell Signalling) at 10 µM or Me2SO (solvent) for 30 min prior to the addition of rBgMIF and were then treated as above.
For analysis of sustained activation of ERK, cells (2×105cells/well) were cultured and synchronized as described above. Cells were exposed to various concentrations of rBgMIF, for 2 h, 8 h and 24 h, and then lysed as described above. Cell lysates were used for western blot analysis of phosphorylated and total ERK content.
The apoptosis assay used was adapted from [31]. Bge cells (semi-confluent) were cultured in 6-well plates in complete Bge medium. Cells were pretreated for 12 h with rBgMIF at different concentrations. The NO donor, S-nitrosoglutathione (SNGO, Sigma) was then added at 250 µM for 8 h. Apoptosis was measured by the Terminal deoxynucleotidyl transferase mediated dUTP Nick End Labeling (TUNEL) assay (Roche Applied Biosystems), following the manufacturer's instructions. Briefly, the cells were fixed in paraformaldehyde 4% for 1 h, washed and permeabilized with sodium citrate 0,1%/Triton -X 100 0.1% for 2 min on ice. Cells were incubated with “labeling solution” for 1 h at 37° C, washed with PBS and the number of positive cells was visualized on a FACSCalibur flow cytometer (Becton Dickinson) and the data were treated with CellQuestPro software (Becton Dickinson). The data are displayed using a logarithmic scale and the results are represented as the percentage of cells undergoing apoptosis. Activation of p53 was investigated by performing a western blot on total protein extracts from Bge cells using the anti-p53ser15P antibody and the anti-actin antibody.
PCR products were amplified from the pCR2.1 TOPO vector containing the complete BgMIF sequence, purified (Wizard SV Gel and PCR Clean up system, Promega) and used as a template for T7 transcription and synthesis of BgMIF dsRNA (MEGAScript T7 kit, Ambion). The firefly (Photinus pyralis) luciferase gene dsRNA (pGL3 vector, Promega) was used as a control (see Sup. Table 1).
Bge cells (2×105cells/well) were cultured in 12-well plates in complete Bge medium. For the analyses of BgMIF excretion the medium was changed for complete CBSS and the cells were stimulated with different quantities (30, 60, 120, 240 and 480 µg/ml of protein content) of S. mansoni ESP (prepared as described in [33]) for 12 h. Cell supernatants were collected, centrifuged for 10 min at 800 g to eliminate non adherent cells, then concentrated 10-fold by membrane filtration with a 10 kDa cut-off (Centricon, Amicon). For the study of Bgp38 activation, the protocol was adapted from [33]. Briefly, cells were exposed to 120 µg/ml of ESP for 5, 15 and 30 min, and the protein extracts were analysed for the phosphorylated p38 and actin content by western blotting. The content of endotoxin was measured with Limulus Amoebocyte Lysate (Cambrex). ESP used in the bioassays contained less than 17 pg endotoxin/mg of protein.
Each dsRNA (2 µg) was transfected into confluent cultures of Bge cells using the FUGENE HD transfection reagent (Roche Applied Biosystems), following the manufacturer's instructions. For the experiments with EPS, the Bge medium was changed to complete CBSS 2 days after the addition of dsRNA. On the third day 120 µg/mL of ESP products were added to the medium and the presence or absence of an aggregation phenotype was determined. For encapsulation experiments, the medium was replaced by fresh medium and S. mansoni mother sporocysts cultured in complete CBSS for 48 hours, were added (500 sporocysts/well) 3 days after the addition of dsRNA. The co-culture was maintained 4 days to allow the observation of an in vitro encapsulation phenotype as described previously [50]. Aggregation and encapsulation phenotypes were observed using an Eclipse TS100 optical microscope (Nikon) and the images were acquired with a DS-Fi1digital camera (Nikon) and treated with Photoshop CS3 (Adobe). For encapsulation experiments, 250–300 sporocysts were counted per assay and the results were represented as the percentage of sporocysts completely covered in adhering cells.
Each dsRNA (15 µg in 10 µl of sterile CBSS) was injected into the cardiac sinus of B. glabrata snails, using a 50 µl Hamilton syringe with a 26s needle (Hamilton). Three days after injection, hemocytes were isolated from three snails per group and the snails were individually frozen in liquid nitrogen for extraction of RNA and soluble protein. Knock down efficiency was checked by real-time PCR and western blot analyses. Snails were infected three days after injection, with 20 S. mansoni miracidia. 24 h post infection the hemolymph of three snails was pooled and diluted by half in complete CBSS containing citrate/EDTA (50 mM sodium citrate, 10 mM EDTA, and 25 mM sucrose) [58] and the composition of the hemocyte population in each condition was assessed by FACS analyses using SSC and FSC parameters in a FACSCalibur flow cytometer (Becton Dickinson). The hemocyte population was analyzed in pools of three snails.
In order to determine the number of mother sporocysts (Sp1) present in the head–foot region of infected snails, mollusks were fixed 15 days post-exposure as described [59], [60]. Briefly, the snails were relaxed in water containing an excess of crystalline menthol for 6 h. The snail shell was removed and the body was fixed in modified Raillet–Henry's solution (930 ml distilled water, 6 g NaCl, 50 ml Formol 40%, 20 ml acetic acid). The head–foot zone was dissected and visually inspected. In each snail Sp1 were readily observable as translucent white bodies within an opaque yellow tissue background.
All data were expressed as mean plus or minus SE. The statistical significance of differences was assessed using the Mann–Whitney U test for nonparametric data or Student's t-test using the program StatView (Abacus Concepts). P values of less than .05, 0.01 or .001 were used to indicate statistical significance.
Nucleotide sequence data reported in this paper are available in the GenBank database under the accession numbers ACR81564 (BgMIF), GU929337 (Bgp53).
|
10.1371/journal.pgen.1003682 | In Vivo Bypass of 8-oxodG | 8-oxoG is one of the most common and mutagenic DNA base lesions caused by oxidative damage. However, it has not been possible to study the replication of a known 8-oxoG base in vivo in order to determine the accuracy of its replication, the influence of various components on that accuracy, and the extent to which an 8-oxoG might present a barrier to replication. We have been able to place a single 8-oxoG into the Saccharomyces cerevisiae chromosome in a defined location using single-strand oligonucleotide transformation and to study its replication in a fully normal chromosome context. During replication, 8-oxoG is recognized as a lesion and triggers a switch to translesion synthesis by Pol η, which replicates 8-oxoG with an accuracy (insertion of a C opposite the 8-oxoG) of approximately 94%. In the absence of Pol η, template switching to the newly synthesized sister chromatid is observed at least one third of the time; replication of the 8-oxoG in the absence of Pol η is less than 40% accurate. The mismatch repair (MMR) system plays an important role in 8-oxoG replication. Template switching is blocked by MMR and replication accuracy even in the absence of Pol η is approximately 95% when MMR is active. These findings indicate that in light of the overlapping mechanisms by which errors in 8-oxoG replication can be avoided in the cell, the mutagenic threat of 8-oxoG is due more to its abundance than the effect of a single lesion. In addition, the methods used here should be applicable to the study of any lesion that can be stably incorporated into synthetic oligonucleotides.
| In the course of normal cellular functions, many types of reactive oxygen species are produced that can lead to oxidative damage in the cell. DNA bases are subject to the formation of various oxidative lesions; one of the most common is the production of 8-oxoG which can pair relatively well with A instead of C, leading to GC→TA transversion mutations. In this work, we have been able to place a single 8-oxoG in the yeast chromosome and observe its replication. We find that in a wild-type cell its replication is surprisingly accurate due primarily to two components: DNA mismatch repair, which recognizes an A inserted opposite the 8-oxoG and initiates its removal and subsequent re-replication; and translesion synthesis in which the existence of the 8-oxoG induces a switch from a normal replicating DNA polymerase to a specialized DNA polymerase, Pol η, that can accurately replicate an 8-oxoG. We also find that the 8-oxoG can cause the replicating strand to shift to the newly replicated strand in the sister chromatid, thus avoiding the 8-oxoG lesion. These findings indicate not only how cells deal with a known DNA lesion, but also demonstrate how other such lesions can be studied in the future.
| All DNA bases are subject to a variety of different types of damage due to reactive oxygen species (ROS) [1]. Among the most common and most mutagenic is 7,8-dihydro-8-oxoguanine, or 8-oxoG, which is mutagenic because of its tendency to pair with adenine and thus create GC to TA transversion mutations [2], [3]. In the yeast Saccharomyces cerevisiae there are several mechanisms either to repair 8-oxoG lesions or to prevent 8-oxoG-induced mutations. 8-oxoG lesions opposite C, which would be formed by oxidative damage of double-stranded DNA, are removed by the glycosylase Ogg1 [4], [5], which has little if any activity on 8-oxoG paired with other bases [6]. Mismatch repair (MMR) plays an important role in preventing mutations due to oxidative damage [7], and it has been shown that yeast MutSα, consisting of the Msh2 and Msh6 subunits, recognizes A replicated opposite an 8-oxoG lesion and thereby prevents mutations [8]. Thus in S. cerevisiae, MMR appears to replace the function of MutY, which is absent [8], [9]. MutSβ, consisting of the Msh2 and Msh3 subunits appears to play no role in 8-oxoG repair [8].
For 8-oxoG lesions that are not removed prior to replication, translesion DNA synthesis (TLS) is importantly involved in bypass, with Pol η playing the major role in yeast. A variety of biochemical experiments using oligonucleotide templates with an 8-oxoG lesion have demonstrated that Pol η replicates through an 8-oxoG lesion, usually inserting a C [10]–[12]. This accuracy is explained by structural studies that show Pol η with a template containing an 8-oxoG lesion can hold the lesion in an anti conformation, permitting a C to be inserted [13]. In contrast, Pol δ is ten-fold less accurate and efficient in bypassing 8-oxoG [14] and Pol ε does not bypass 8-oxoG at normal dNTP concentrations, but does, inaccurately, at damage-induced levels of dNTPs [15]. Genetic studies are more complicated because the existence of an 8-oxoG lesion can be inferred only by its mutation signature, generally in an ogg1 background that greatly increases the amount of 8-oxoG in DNA. Such studies were used to show the involvement of MMR in preventing mutations due to 8-oxoG [8], the role of Pol η in accurate replication of 8-oxoG [11], [16], and the lack of a significant role for Pol ζ [16], [17]. The interplay of TLS and MMR is not completely clear. It was proposed that MMR was responsible for recruiting Pol η for bypass [18] but a detailed study of 8-oxoG bypass and repair concluded that Pol η acted independently of MMR [19]. It appears that monoubiquitination of PCNA is necessary for most TLS and in yeast this step is carried out by the Rad6-Rad18 heterodimer [20], [21]. Genetic studies implicate RAD6 and RAD18 as well as RAD30 (the gene encoding Pol η) but not REV3 (the gene encoding the catalytic subunit of Pol ζ) in 8-oxoG tolerance [16].
Most TLS is assumed to occur at the replication fork [20], [21], although it can occur after S phase [22]. Furthermore, as there appear to be different replicative polymerases on the leading and lagging strands of replication [23], one might expect 8-oxoG tolerance could exhibit strand differences. There are only a limited number of such studies. Using a reversion analysis of a URA3 mutation, it was found that 8-oxoG was preferentially repaired on the lagging strand of replication [24]; most of the differential repair was ascribed to the preferential activity of MMR on the lagging strand [25]. Using a mutation analysis of ogg1-dependent mutations in a SUP4-o reporter assay, the lagging strand bias of MutSα was observed, as well as a lagging strand bias for accurate Pol η bypass [19].
It is not clear what effect an 8-oxoG lesion has on replication. Some work has suggested that an 8-oxoG lesion has no effect on replication [10], [18], whereas a stall site was observed in vitro at a nucleotide prior to an 8-oxoG lesion with Pol δ but not Pol η [11]. An in vivo study inferred replication stalling or blockage from a mutational analysis [19]. Lesions that block or stall replication forks appear to be tolerated, especially in yeast, by homologous recombination [26]. Recent interest has focused on tolerance mechanisms by template switching in which a blocked 3′ end invades the replicating sister strand, either by a fork regression or strand invasion [26]. Such mechanisms of template switching appear to be dependent on polyubiquitination of PCNA by a complex of Ubc13-Mms2-Rad5 [20]. Because the substrate of Ubc13-Mms2-Rad5 is PCNA monoubiquitinated by Rad6-Rad18, template switching would also be expected to be dependent on Rad6 and Rad18 [20]. In addition to its role in polyubiquitination, the helicase function of Rad5 may also be important in template switching [27], [28].
Rather than using an ogg1 mutant background, a more direct method of analyzing 8-oxoG bypass in vivo would be to introduce DNA containing a defined lesion directly into cells. Plasmids containing a single-strand gap with an 8-oxoG or 8-oxoG in duplex DNA have been introduced into E. coli and mammalian cells [29]–[32] and a plasmid treated with methylene blue to induce oxidative damage was introduced into yeast for analysis [33]. The problem with the use of plasmids for analysis, in addition to the difficulty of substrate construction, is that the mechanism of replication may differ from that within the chromosome and various forms of recombinational bypass may also differ. Another approach would be to transform cells with single-stranded oligonucleotides (oligos) containing an 8-oxoG lesion. Transformation of yeast with oligos was first performed in Fred Sherman's laboratory [34], [35] and the method has subsequently been used to study various lesions carried into yeast by oligos [36]–[41]. However, in most cases the lesion itself was responsible for generating a phenotype and with one exception [36] the mechanism of transformation with oligos was not fully understood.
In order to study 8-oxoG bypass, we wanted to introduce the 8-oxoG on an oligo that would create a selectable phenotype that would be independent of the presence of the 8-oxoG lesion. Such an experimental design allows us to study both replication across the 8-oxoG and bypass of the 8-oxoG by template switching outside of a context of overall increased oxidative damage in the cell. As detailed below, we find evidence that the 8-oxoG lesion does stall replication; that only Pol η is able to replicate 8-oxoG accurately; that template switching is invoked frequently in the absence of Pol η; and that MMR strongly influences the outcome of 8-oxoG replication.
We sought a system in which a damaged base could be placed into the chromosome independent of an oligo-induced reversion event and so needed a low spontaneous reversion rate coupled with a tight selection. We turned to the set of trp5 point mutations we previously constructed [42]. These strains contain a mutation at either nucleotide position 148 or 149 and can only be reverted to the wild type phenotype by restoring the original TRP5 sequence [42]. The plan was therefore to revert a mutant TRP5 gene with an oligonucleotide containing the wild-type base along with a damaged base at a different location in the oligo. A potential problem was that the region surrounding the mutant base is highly conserved, constraining the location of any damaged base. We therefore created the mutant trp5-G148Cm gene (Figure 1) [43]. Because this mutant trp5-G148Cm gene is placed close to a dependable origin of replication, and is present in both orientations relative to the origin, we know which strand is replicated as leading and which as lagging and can reverse the replication strands by using a strain of opposite TRP5 orientation [42]. In order to use oligos to incorporate a segment of DNA, it was necessary to know the frequency of co-incorporation of nucleotides in a given oligo. Using oligos with markers spread throughout the length of the oligo (Oligo N, Figure 1), we determined that for an oligo of 40 nt in length, a central core of 10–15 nt was incorporated with a greater than 90% frequency [43]. Those results suggested that it was feasible to use oligos of that length for our experiments.
We had initially hoped to investigate damaged base bypass by transforming with an oligo which contained one normal base to revert the Trp- phenotype and another damaged base placed in a silent position where any base incorporation would be tolerated. However, our prior experiments [43] as well as a number of preliminary experiments indicated that we needed a method to mark incorporation of bases on both sides of the damaged base in order to be sure that we were observing bypass, and not partial incorporation of the relevant region of the oligo. These goals were accomplished by transforming with Oligos G and GO (Figure 1). The C at position 20, highlighted in yellow, creates a Trp+ phenotype upon incorporation; the G at position 12, highlighted in blue, if incorporated, creates a new SphI site. The G at position 15 is an 8-oxodG in Oligo GO and is highlighted in red, forming an 8-oxoG-G mismatch with the trp5-G148Cm sequence. Oligo G is identical, with a G instead of an 8-oxodG at position 15. This mismatch was deliberately chosen, as one of the main glycosylases processing 8-oxodG, Ogg1, should have little or no activity on an 8-oxodG-G mismatch [44], [45], and the efficiency of its removal by another glycosylase, Ntg1, is low, if it exists [46]. In addition, 8-oxodG, when bypassed, is very unlikely to template a G, so if the original sequence at that position is maintained, that would be strong evidence either of removal of the 8-oxodG before replication, or a failure to bypass the 8-oxodG. The expectation for 8-oxodG bypass is that either a C or A is incorporated. If an A is incorporated opposite the 8-oxodG, a BfaI site is created, thus allowing a simple restriction digestion to indicate a mutagenic bypass of the 8-oxodG. In summary, at the site in question, a G on the coding strand indicates that 8-oxoG was either not used as a template for replication or was removed before replication, a C indicates that 8-oxoG was bypassed accurately, and an A indicates inaccurate replication of 8-oxoG.
The overall design of the assay system and its expected results are illustrated in Figure 2A. Incorporation of the oligo can be selected by the Trp+ phenotype, and given that only 7 nt separate the base creating the Trp+ phenotype and the base creating an SphI site, we initially expected that all Trp+ cells should contain a new SphI site. What we found as analysis proceeded is that a fraction of oligos, even those containing all normal bases, exhibited “partial removal” as indicated in Figure 2A: in the presence of MMR, a substantial fraction of cells (as much as 30% or more) transformed by Oligo G (containing only normal bases) were Trp+ but did not contain an SphI site [43]. Those results were explained by a failure of MMR to recognize the C-C mismatch created by the oligo during MMR-directed excision from the 5′ end of the oligo [43]. Such results were seen only in the presence of MMR and with Oligo G and Oligo GO, but not with Oligo UG or UGO, as will be detailed below. It is in the second round in which the oligo sequence, now fully incorporated into the genome, is replicated for the first time. In Trp+ cells that were transformed by Oligo GO and contain the SphI site, DNA synthesis must have used the 8-oxoG for a template, and the base inserted can be subsequently analyzed. In the absence of MMR, all Trp+ cells would be expected to contain an SphI site, and that is true for cells transformed by Oligo G but not for all strains transformed by Oligo GO. The failure of cells transformed by Oligo GO to contain an SphI site could be explained by the process of template switching, in which the replication fork switches to use the newly replicated strand of the sister chromatid [26].
Strains with a variety of different genotypes were transformed by Oligo G and Oligo GO and assayed for the presence of an SphI site. The results for strains of the R orientation are shown in Figure 3A. Results for strains of the F orientation are shown in Figure S1 and the numbers of colonies analyzed for each strain are given in Table S1. Because of the problem of partial oligo removal discussed above, strains with an active MMR cannot be analyzed for template switching (i.e. Trp+ transformants lacking an SphI site) with Oligo GO. In MMR-defective strains, with the exception of rad30 msh6 strains lacking both MMR and Pol η, the SphI site is created in almost all Oligo GO transformants. If the lack of the SphI site in that background is due to template switching, it should be blocked by loss of Rad5, Mms2, or Rad18 [20], [27], [47], [48]. That is seen to be true, as rad5 rad30 msh6, mms2 rad30 msh6, and rad18 rad30 msh6 strains show minimal loss of the SphI site.
Our previous results had suggested that placing a base creating an additional mismatch 3′ of the C-C mismatch in Oligo G would prevent the partial removal of the oligo illustrated in Figure 2A [43]. Therefore, we used Oligo UG and Oligo UGO (Figure 1) to repeat a subset of the experiments shown in Figures 3A and S1. The results assaying presence of the SphI site in both R and F strains are presented in Figure 3B. As observed with Oligo GO, Oligo UGO displays template switching in the absence of both MMR and Pol η (Figure 3B). Loss of the SphI site is suppressed in rad5 rad30 msh6 strains. Oligo UG transformants in the presence of MMR showed little loss of the SphI site (Figure 3B). Oligo UGO transformants in rad30 strains also demonstrated little SphI site loss, and were significantly reduced in template switching compared to rad30 msh6 strains (Figure 3B). Therefore we can conclude that most template switching is suppressed by MMR. It appears in Figure 3B that the level of SphI site loss in rad30 Oligo UGO transformants is somewhat elevated compared to wild-type strains. The difference is not statistically significant in strains with the F orientation, and is only marginally significant (P = 0.03) in the R orientation.
8-oxodG is considered to be extremely mutagenic due to the frequency of misreplication, with an A inserted opposite the 8-oxodG. In order to measure the bypass accuracy of the introduced 8-oxodG, we selected only those revertants that were both Trp+ and contained an SphI site, as all of those revertants should have incorporated the intervening 8-oxodG into the genome. As illustrated in Figure 2, the 8-oxodG in oligos GO and UGO was placed opposite a G in the genome; thus removal of the 8-oxoG lesion would have resulted in retention of the original sequence at that point. Replication of the 8-oxoG lesion would be expected to yield only a C for accurate bypass or an A for inaccurate bypass, both leading to a change of sequence at that position. The replication accuracy could have been directly determined by sequencing each one of the revertants. However, as indicated in Figure 1, inaccurate replication with an A creates a novel BfaI site, allowing a direct measurement of accuracy without sequencing. In order to assess the validity of this approach, we sequenced 129 Trp+ revertants that contained the introduced SphI site but lacked the BfaI restriction site and found 126 C, 1 T and 2 G at that position, thus confirming the utility of the restriction site assay and the assumption that a C would be found in such cases; 22 out of 22 sequences that contained the BfaI site had an A as expected. The result of this assay in a variety of genetic backgrounds using Oligo GO is shown for strains of R orientation in Figure 4A and for the F orientation in Figure S2. All numbers are given in Table S1. To our surprise, not only was replication extremely accurate in wild-type cells, it was also highly accurate in the absence of MMR, averaging 94% in MMR-defective strains of both orientations compared to 97% in wild-type strains; the difference is not statistically significant. The source of the accurate replication was clearly Pol η, as in MMR-deficient strains in the absence of Pol η, the accuracy dropped to 36% in R orientation and 44% in F orientation (Figures 4A and S2; Table S1). The resulting 8-oxoG-A mismatch was efficiently recognized and corrected by MMR, as the replication accuracy in rad30 strains was 92% in R and 93% in F (Figures 4A and S2; Table S1), neither of which was significantly different from that measured in wild type or MMR-defective strains.
In order to confirm our results with Oligo GO, we conducted a reduced set of experiments with Oligo UGO (Figure 4B; Table S1). The accuracies measured in either rad30 or msh6 strains were not significantly different from each other. The double mutant combinations of msh6 rad30 were significantly lower, at 44% in F and 36% in R orientation.
The graphs of accuracy in Figures 4 and S2 demonstrate that there are basically two categories of strains: those strains with deficient Pol η and MMR, and those that have at least one of the two pathways intact (as discussed above, Rad18 is thought to be necessary for Pol η function, which is consistent with our results). In general within each group there is no statistically significant difference among strains, and there is a significant difference between strains in the two groups. It appears that rad5 strains could be an exception. In both the R (Figure 4A) and F (Figure S2) orientations, accuracy in rad5 strains is lower than in wild-type, and accuracy in rad5 msh6 strains is lower than in msh6 strains. The P values are marginal, ranging from 0.01 to 0.04, but the pattern is consistent in the four comparisons. Another question is whether there are differences between the accuracies observed in the two orientations of the TRP5 gene. The measured accuracy in R strains is lower than in F strains for Oligo GO in msh2, msh6, msh3 msh6, and for Oligo UGO in msh6 strains. However, only by combining the results in msh2, msh3, and msh3 msh6 strains for Oligo GO does the difference approach statistical significance (P = 0.05). Because the 8-oxoG is replicated in the second round, replication of the 8-oxoG would be on the leading strand in strains with the R orientation.
In this work, we have developed a method of analyzing bypass of 8-oxoG in vivo by using oligos to place a single 8-oxoG in a defined location in the chromosome. Once incorporated into the chromosome, the 8-oxoG lesion is not replicated until a second cell cycle, such that the measured replication is of a lesion fully integrated into the chromosome. The replication of 8-oxoG was surprisingly accurate and that accuracy was due to the synergistic action of MMR and Pol η. Although the general expectation was that 8-oxoG would not affect replication, 8-oxoG-induced template switching was observed, but for the most part only in the absence of both MMR and Pol η.
We know that Trp+ revertants containing a new SphI site must have resulted from replication past the 8-oxoG and can therefore measure the accuracy of that bypass. 8-oxoG is considered to be a very mutagenic lesion; therefore it was somewhat of a surprise that in wild-type cells it was replicated quite accurately. The accuracy of replication was 98% in F strains (replication on the lagging strand) and 97% in R strains. Even more surprising was the accuracy in MMR-deficient cells; pooling data from all genotypes lacking MutSα (msh2, msh6, msh3 msh6) gave 96% accuracy in F strains and 91% in R. As noted above, this difference is of marginal statistical significance (P = 0.05), but it is in agreement with experiments that showed a lagging strand bias for Pol η [19]. One important distinction between our measurement and other in vivo measurements is that our strains contain one 8-oxoG lesion above the background level of such lesions, whereas most other measurements have been made in Ogg1-deficient strains in which one would expect large numbers of additional 8-oxoG lesions. Given that the amount of MutSα in cells is low (one estimate is 1230 molecules of Msh2p and 5330 molecules of Msh6p per cell [49]), elevated levels of 8-oxoG in the cell could potentially titrate out MutSα. The source of accurate replication of 8-oxoG in MutSα-deficient strains is clearly Pol η, as in cells deleted for Pol η, 8-oxoG is replicated accurately only 40% of the time (Figure 4 and Table S1). Pol η is also in relatively low abundance in yeast (an estimated 1860 molecules per cell [49]) again suggesting a potentially misleading picture of 8-oxoG replication in cells with elevated levels of 8-oxoG. Thus our measurements indicate the levels of accuracy to be expected for repair of spontaneous levels of oxidative damage in normal conditions, but might not apply for cells under oxidative stress or with reduced levels of MMR or Pol η. Given that the accuracy of bypass in the absence of MMR was due to Pol η, our results support the independence of MMR and Pol η in maintenance of accuracy, as previously observed [19], and are not consistent with a model in which MMR is required to recruit Pol η [18].
Previous in vitro measurements had determined that yeast Pol η could replicate an 8-oxoG much more accurately than could Pol δ [11], [14], but it is impossible to extrapolate from such in vitro experiments using single DNA polymerases to an in vivo situation in a chromosomal context with multiple DNA polymerases available. The low accuracy observed in MMR- and Pol η-defective strains (40%) suggests that no other DNA polymerase in the cell is able to replicate 8-oxoG accurately. Therefore the high accuracy of 8-oxoG replication observed in MMR-defective strains indicates that most of the 8-oxoG replication must be due to Pol η. (Suppose Pol η is used for 75% of 8-oxoG bypass, and other polymerases with only 40% accuracy are used the rest of the time; the overall accuracy of bypass in the absence of MMR would be approximately 85%, considerably lower than what we observe.) In vitro experiments found a strong stall site with Pol δ just before replication of an 8-oxoG [11]; such a stall is likely the signal responsible on either replication strand for switching to synthesis by Pol η or switching templates.
Particularly because of the different abilities of Pol δ and Pol ε to bypass an 8-oxoG in vitro [14], [15], one might have expected a difference in replication fidelity due to replication of leading and lagging strands by different DNA polymerases [23], although we see little evidence for that here. The similarity in accuracy of leading and lagging strands in the absence of Pol η is surprising, particularly given the expected difference in the ability of Pol δ and Pol ε to bypass 8-oxoG. Our data suggest that the 8-oxoG lesion causes a stall; it has been hypothesized that a lesion on the leading strand could induce a switch to continued synthesis on the leading strand by Pol δ [50] and so it is possible that the synthesis observed on either strand across 8-oxoG in the absence of Pol η could be due primarily to Pol δ.
The design of these experiments made it possible to observe template switching, in which the replicating DNA polymerase uses DNA from the replicating sister chromatid as a source of template [26]. In strains deficient in MMR, template switching was observed only in strains lacking Pol η (Rad30), and in those cases, occurred in about half of the replication events on both the leading and lagging strands of replication (Figures 3 and S1). As expected for template switching, these events were not observed in strains lacking Rad5, Mms2, or Rad18.
Template switching was measured as Trp+ revertants that did not contain the SphI site introduced by the oligo. As explained above, experiments with Oligo GO could not examine possible template switching in the presence of MMR because of the loss of part of the oligo during transformation in some Trp+ colonies (Figure 2A). Such partial loss was not observed with Oligo UGO, and those experiments showed that MMR suppresses template switching, as such events were significantly lower in rad30 strains compared to rad30 msh6 strains in both orientations (Figure 3B). It is possible that some template switching events were missed in these assays, as only 4 nucleotides separate the 8-oxoG from the C needed to produce Trp+ cells. If in template switching, there is loss of more than 4 bases from the 3′ invading end, the resulting strain would be Trp- and therefore not observed.
Template switching could occur via fork regression on the leading strand [26], but on the lagging strand must occur via a mechanism involving homologous recombination [47]. What triggers template switching and how does MMR suppress template switching? A stall by the replicative DNA polymerase in advance of the 8-oxoG is a strong candidate for a template switching signal. As postulated above, as a replicative DNA polymerase encountered an 8-oxoG, it would stall and either induce a switch to Pol η replication or the replicative polymerase would switch templates and bypass the lesion in an error-free manner. In the presence of MMR, presumably the same template switching would occur but when the DNA copied from the sister chromatid was brought back to pair with the template strand, MMR would recognize the 8-oxoG-G mispair and initiate removal of the newly synthesized DNA, thus abolishing the effect of the template switch.
It is evident from Figures 3A and S1 that deletion of either RAD5 or MMS2 blocks template switching. However, deletion of RAD5 in either wild-type or msh6 strains appears to somewhat decrease accuracy in Oligo GO transformants, whereas deletion of MMS2 in the same strains does not (Figures 4A and S2). That result is consistent with a role for Rad5 in addition to template switching [51]. A rad5 strain is more sensitive to UV damage than an mms2 strain [52] and a role for Rad5 was observed in TLS of UV damage independent of Mms2-Ubc13 [53]. Although on many substrates the Rad5-dependent TLS might be mutagenic [54], it would appear from our work that the Rad5-mediated events are accurate in replicating 8-oxoG.
The sequence context of a lesion can affect its fate. The trp5-G148Cm strains we have made could accommodate a lesion at several different positions, and thus somewhat different sequence contexts. Another option would be to place a lesion in a sequence context of choice and integrate it into a strain that would select for the loop integration [36]. The potential difficulty with such methods is that the spontaneous background of such reversion events, particularly in the absence of MMR, is considerably higher than in the trp5-G148Cm strains. The use of oligos to place defined DNA damage at unique places in the chromosome is potentially very informative. With proper markers in the oligos, we have shown that the fate of a defined lesion can be measured in a completely normal chromosome context in a variety of genetic backgrounds.
The trp5-G148Cm mutation was created as described for the other trp5 mutations [42] using delitto perfetto [55] and created the sequence CGATGTTATCCAACTGGGA starting at position 138 of TRP5 with mutated bases underlined. The lys2CT1265GA mutation was similarly created by delitto perfetto. The genotypes of strains used in these experiments are given in Table S2. All gene deletions were created by one-step disruption with PCR generated fragments. In general gene deletions were made from a PCR fragment generated from the collection of yeast gene deletions [56]. The kanMX4 resistance marker was changed to hphMX4 or natMX4 by transformation with a fragment from pAG32 or pAG25, respectively [57]. For the msh6Δ::loxP deletion, the PCR fragment was from a strain in which MSH6 had been disrupted by a loxP-kanMX-loxP fragment that was subsequently excised by Cre expression [58].
Transformation was a modification of the method used previously [43], [59]. An overnight culture of a strain was diluted 1∶50 in YPAD [60], incubated with shaking at 30° to an OD600 of 1.3–1.5, washed twice with cold H2O, and once with cold 1 M sorbitol. After the final centrifugation, all solution was removed from the cells and a volume of cold 1 M sorbitol equal to that of the cell pellet added to resuspend the cells. For a typical transformation, 200 pmol of a Trp oligo and 200 pmol of LYS2TCARev40 (used to revert the lys2CT1265GA mutation) was added to 200 µl of this cell suspension in a 2-mm gap electroporation cuvette, and the mixture electroporated at 1.55 kV, 200 Ω, and 25 µF (BTX Harvard Apparatus ECM 630). Immediately after electroporation, the cell suspension was added to a volume of YPAD equal to that of the initial culture, and the cells incubated at 30° with shaking for 2 h. Cells were then centrifuged, washed with H2O, and plated on synthetic dextrose (SD) medium lacking either tryptophan or lysine [60] to select transformants. The number of Lys+ transformants served as a useful guide that a particular transformation experiment had worked, but was not correlated well enough with the number of Trp+ transformants to be used as an internal control (results not shown).
Individual Trp+ revertants were picked into 200 µl SD-Trp medium in 96-well deep well plates, grown overnight at 30° with shaking, a small aliquot of each transferred to fresh SD-Trp medium with a Boekel Microplate Replicator and grown overnight, and finally transferred with the replicator to another deep well plate for overnight growth in 300 µl YPAD. Cells were then transferred with the replicator to a PCR microplate containing 15 µl per well of 2 mg/mL Zymolyase 20T (USBiological) in 0.1 M Phosphate Buffer pH 7.4 and incubated at 37° for 30 min and 95° for 10 min. After incubation, 85 µl H2O was added to each well. PCR was performed using 5 µl of the lysate in a total volume of 50 µl of the recommended buffer with 0.3 µM trpseq2 and trpseq8 primers [43] and 0.5 µl Takara e2TAK DNA polymerase for 30 cycles. For restriction digestion, 5 µl of the PCR reaction was incubated with 2 units of either BfaI or SphI (New England Biolabs) in the recommended buffer in a total volume of 15 µl at 37° overnight and analyzed by gel electrophoresis.
The number of revertant colonies analyzed for each strain and oligo are given in Table S1. For each combination, usually 48 colonies were analyzed; in some instances experiments were repeated multiple times. For experiments repeated three or more times, means and standard deviations are shown in the figures. For comparison of results between strains, all data from a given strain and oligo were combined. Statistical calculations were performed using the VassarStats website (http://www.vassarstats.net/). In most cases, the number of samples was large enough for use of a chi-square test; in the remainder of cases, a Fisher's exact test was used. P values are given for each comparison in the relevant figure.
|
10.1371/journal.pcbi.1003278 | Machines vs. Ensembles: Effective MAPK Signaling through Heterogeneous Sets of Protein Complexes | Despite the importance of intracellular signaling networks, there is currently no consensus regarding the fundamental nature of the protein complexes such networks employ. One prominent view involves stable signaling machines with well-defined quaternary structures. The combinatorial complexity of signaling networks has led to an opposing perspective, namely that signaling proceeds via heterogeneous pleiomorphic ensembles of transient complexes. Since many hypotheses regarding network function rely on how we conceptualize signaling complexes, resolving this issue is a central problem in systems biology. Unfortunately, direct experimental characterization of these complexes has proven technologically difficult, while combinatorial complexity has prevented traditional modeling methods from approaching this question. Here we employ rule-based modeling, a technique that overcomes these limitations, to construct a model of the yeast pheromone signaling network. We found that this model exhibits significant ensemble character while generating reliable responses that match experimental observations. To contrast the ensemble behavior, we constructed a model that employs hierarchical assembly pathways to produce scaffold-based signaling machines. We found that this machine model could not replicate the experimentally observed combinatorial inhibition that arises when the scaffold is overexpressed. This finding provides evidence against the hierarchical assembly of machines in the pheromone signaling network and suggests that machines and ensembles may serve distinct purposes in vivo. In some cases, e.g. core enzymatic activities like protein synthesis and degradation, machines assembled via hierarchical energy landscapes may provide functional stability for the cell. In other cases, such as signaling, ensembles may represent a form of weak linkage, facilitating variation and plasticity in network evolution. The capacity of ensembles to signal effectively will ultimately shape how we conceptualize the function, evolution and engineering of signaling networks.
| Intracellular signaling networks are central to a cell's ability to adapt to its environment. Developing the capacity to effectively manipulate such networks would have a wide range of applications, from cancer therapy to synthetic biology. This requires a thorough understanding of the mechanisms of signal transduction, particularly the kinds of protein complexes that are formed during transmission of extracellular information to the nucleus. Traditionally, signaling complexes have been largely perceived (albeit often implicitly) as machine-like structures. However, the number of molecular complexes that could theoretically be formed by complex signaling networks is astronomically large. This has led to the pleiomorphic ensemble hypothesis, which posits that diverse and rapidly changing sets of transient protein complexes can transmit and process information. Our goal was to use computational approaches, specifically rule-based modeling, to test these hypotheses. We constructed a model of the prototypical yeast mating pathway and found significant ensemble-like behavior. Our results thus demonstrated that ensembles can in fact transmit extracellular signals with minimal noise. Additionally, a comparison of this model with one tailored to generate machine-like complexes displayed notable phenotypic differences, revealing potential advantages for ensemble-like signaling. Our demonstration that ensembles can function effectively will have a significant impact on how we conceptualize signaling and other processes inside cells.
| Much of our reasoning about the function of biological systems relies on the formation of multi-subunit protein complexes [1]. In some cases, such as the ribosome and the proteasome, these complexes take the form of intricate molecular machines with well-defined quaternary structures [2]–[4]. The overall structure of complexes formed during signal transduction, however, is considerably less clear. There are a few well-characterized signaling machines, like the apoptosome, and some have argued that the majority of structures produced by signaling networks would have a machine-like character [5], [6]. Most of the complexes formed during signal transmission and processing have not had their global three-dimensional structures experimentally determined, however, and as such we currently do not know the extent to which signaling occurs via machines [7]. Despite this uncertainty, the machine-like perspective on signaling complexes is pervasive in the literature, if often implicit; for instance, one commonly represents signaling networks graphically by drawing large complexes in which all of the relevant proteins interact simultaneously [8]–[14] (Fig. 1A). Although such diagrams are often presented as compact summaries of a set of interactions, they are certainly evocative of a machine-like structure, and lead naturally to analogies between signaling complexes and highly ordered objects such as circuit boards [7], [9].
One issue that complicates this machine-based picture is the fact that the protein interaction networks that underlie cellular signaling exhibit considerable combinatorial complexity; that is, they can (theoretically) generate anywhere from millions to 1020 or more unique molecular species [7], [15]–[17]. For example, even a single PDGF receptor dimer has ∼105 possible phosphorylation states, many of which could be (stably) occupied by any given molecule [7], [18]. A similar problem arises in protein folding: a polypeptide chain could theoretically adopt so many conformations that it is a priori difficult to understand how a protein folds quickly and stably into a single native structure [16], [19], [20]. Proteins have evolved energy landscapes with specific features in order to overcome this problem (which is known as the “Levinthal paradox”). In order to assemble well-defined signaling machines, signaling networks would similarly need to evolve specific “chemical potential landscapes” in order to drive the system to a specific set of quaternary structures [16], [19].
Mayer et al. have speculated, however, that signaling networks might not need to assemble machine-like structures at all in order to function [7]. This “pleiomorphic ensemble” hypothesis posits that heterogeneous mixtures of complexes drive cellular responses to external signals. Early work, based on systems of Ordinary Differential Equations (ODEs) that considered a few hundred molecular species, indicated that more diffuse “network” models of signaling could generate reasonable signaling behavior [21], [22]. The dearth of computational methods that can handle combinatorially complex networks has made it difficult to fully test the ensemble hypothesis in realistic networks, however [16]. As such, it is currently unclear if ensembles could even produce reliable responses to signals, or if there is any functional or evolutionary difference between networks that employ ensembles vs. machines.
Over the past 10 years, a set of rule-based methods have been developed that allow one to model the behavior of biological systems without an a priori reduction in the set of possible species that can be formed [11], [16], [21], [23], [24]. Given a model consisting of a specific set of protein interaction rules, we can exactly sample sets of protein complexes (or “conformations”) from the astronomically large set of all possible complexes the model can generate. In this work we employed these methods to investigate the possibility of signaling via ensembles in silico. We focused on the pheromone response network (Fig. 1A), one of multiple mitogen-activated protein kinase (MAPK) cascades in Saccharomyces cerevisiae. This thoroughly characterized signaling cascade involves the scaffold protein Ste5, which is thought to be a nucleation point for the formation of signaling complexes (Fig. 1B) and prevent crosstalk [8]–[10]. Since similar MAPK cascades are found in eukaryotic cells from yeast to humans [25], this network represents an excellent model system for exploring the influence of combinatorial complexity on signaling dynamics.
In our initial model, we included only those interactions (and their requisite molecular contexts) that have been explicitly characterized experimentally. We found that this model is able to fit available data on the response of the network to pheromone, despite exhibiting significant ensemble character. We also constructed an alternative set of rules that could assemble a scaffold-based signaling machine, similar to those typically drawn to graphically summarize the cascade [8]–[14] (Fig. 1A). Although this model does fit some of the available data, we found that it could not replicate the “combinatorial inhibition” of the pathway observed at high levels of Ste5 overexpression [12], [26]; instead, it displayed considerable robustness to such changes. We also demonstrated that TAP/MS, a common technique for experimentally determining the components of “molecular machines” via binary interactions [27], [28], could not distinguish between the complexes formed in these two models, despite their radically different character. Direct experimental tests of the ensemble hypothesis thus require the application of assays that can measure three-way or higher-order interactions, such as fragment complementation, fluorescence triple correlation spectroscopy or single-molecule approaches [29]–[35]. Our findings indicate that ensembles can indeed reliably transmit and process extracellular information, and their inherent plasticity in response to perturbations like scaffold overexpression implies that they may play a role in facilitating the evolutionary variation of signaling systems within cells [36].
A summary of the molecular interactions underlying the yeast pheromone response network may be found in Fig. 1A. Briefly, the signaling cascade is initiated by the interaction between extracellular pheromone molecules and a G-protein coupled receptor (GPCR), which induces dissociation between the α subunit (Gpa1) and βγ subunits (the Ste4-Ste18 complex, hereafter referred to as Ste4) of the G-protein [37]. Ste4 then recruits the scaffold protein, Ste5, which dimerizes, binds numerous kinases (Ste20, Ste11, Ste7) and promotes a phosphorylation cascade resulting in dual-phosphorylation and activation of the MAPK, Fus3 [38], [39]. As mentioned above, the vast majority of graphical depictions of this cascade involve simultaneous binding of all requisite proteins to Ste5 (Fig. 1A) [8]–[14], however to our knowledge there is no explicit experimental evidence that such a large scaffold-based complex is actually formed during signaling. Active Fus3 then translocates to the nucleus, regulating the expression of numerous mating-related genes via the transcription factor Ste12 [10].
To create a dynamical model of this cascade, we constructed a set of rules for these interactions and other events (e.g. post-translational modification, protein synthesis and degradation, nucleotide transfer). The rules themselves, which follow mass-action kinetics, were primarily derived from two sources: an online model (http://yeastpheromonemodel.org) [40] and an ODE model [11], both of which are based on comprehensive literature searches (Section 1 in Supporting Information Text S1). In our initial model, if a reaction (e.g. efficient phosphorylation of Fus3 by Ste7) requires conditions that have been experimentally characterized (e.g. Ste7 also bound to Ste5), they are explicitly represented in the rule. We added no additional constraints to this model, in order to: (a) see if existing knowledge of these interactions is sufficient to produce realistic network dynamics (Fig. 2) and (b) characterize whether they result in machine- or ensemble-like character. The rule set, written in the Kappa rule-based modeling language [41], contains 232 rules, 18 protein and 8 gene agent types and is available as a separate supporting file (“ensemble.ka” in Protocol S1). This model displays considerable combinatorial complexity: even if we only focus on complexes containing the Ste5 scaffold, the system can generate over 3 billion unique molecular structures (Section 3.5 in Text S1). We thus employed KaSim, an open source simulator for Kappa models, to consider the dynamics of the system without a reduction in its combinatorial complexity. Our general simulation strategy is described in detail in the Materials and Methods section and Section 2 in Text S1; a graphical schematic can be seen in Fig. 3A
The model described above has two types of parameters: initial copy numbers (i.e. concentrations) for each of the 18 protein agents and stochastic rate constants for each of the 232 rules. We obtained the initial conditions directly from experimental measurements of copy number in yeast cells [40], [42]. The stochastic rate constants were obtained from a combination of experimental data and parameter fitting. Briefly, 7% of the rate constant parameters in the model have been directly measured for yeast proteins, 68% were estimated from measurements on related proteins in other networks and 25% were completely unknown and thus given approximate values. In order to reproduce experimental observations with our model, we identified 111 rules that were likely to influence experimentally characterized trends and varied their rate constants.
We found that only 25 of these parameters had a strong impact on the dynamics of important observables in the model, and so we only modified those values during our fitting procedure. Of these 25, 22 had original estimates obtained from related proteins. In those cases, we restricted variation of the parameters to an increase or decrease of about one order of magnitude, to maintain similarity between the fitted value and the original estimate. Two of the remaining parameters had no available estimate, and so we restricted variations in those parameters to a biologically realistic range (a table with ranges for each type of parameter is available in Section 1.2 in Text S1). Finally, one parameter, the Gpa1 degradation rate, had been measured experimentally; we restricted variation in this parameter to a less than five fold change, a reasonable range given the inherent error in the experimental measurement [43]. Further details on how we identified and varied these parameters may be found in Section 1.2 in Text S1.
Since each simulation of this model requires over three hours of CPU time, we could not perform fits using standard techniques, nor could we employ statistical methods to understand the probabilistic structure of the parameter space [44], [45]. Therefore, we manually altered these 25 parameters (subject to the above constraints) and simulated the model with the updated rate parameters. We iteratively applied this procedure until the model successfully replicated the dose-response behavior of Fus3 with respect to pheromone (Fig. 2A) [13], [14], the temporal dynamics of G-protein activation (Fig. 2B) [37], and other experimental observations (Figs. S1, S2, S4, and S5 in Text S1). To test the robustness of our results to the particular simulation method, we translated our rules into the related BioNetGen Language (BNGL) and used the same parameters to simulate the model using the BNGL simulator NFsim [23]. The two software packages produced exactly the same dynamics for these rules (Figs. S4, S5 and Section 2.2 in Text S1). The BNGL version of the model is also available as a supporting file (“ensemble.bngl” in Protocol S1).
Given the large number of parameters in the model compared to the amount of data available for fitting, one should not construe the above results as implying this model represents a uniquely valid description of the system. Indeed, as we demonstrate below, even fairly different rule sets can provide (roughly) equivalent fits to this data; we thus cannot make any claims regarding the identifiability of the parameters or even the rule set itself [45], [46]. The point in this case is that it is possible to find some set of parameters that replicate the data, indicating that this model is at least consistent with available observations.
To determine if the model described above signals through ensembles, we implemented a pairwise comparison between the sets of complexes produced in two independent simulations i and j, using the Jaccard distance, which we refer to as “compositional drift” [16]:where Ci represents the set of unique complexes in simulated cell i, Δ and are the symmetric difference and union set operators, respectively, and |X| is the cardinality of set X. Given the complexes present in two simulated cells, drift is the number of complexes unique to either one cell or the other, divided by the total number of complexes in the union of the two cells. Drift can thus be interpreted as the probability that a complex found in one cell is not found in the other at a particular point in time. For example, d = 0 indicates identical sets of complexes, whereas d = 1 means the sets are pairwise disjoint. We only performed this comparison between multiple simulation replicates that started from exactly the same steady-state initial condition; thus d = 0 at t = 0 for all of our simulations (Fig. 3A; Sections 2.3 and 3.3 in Text S1). Note that this calculation takes into account any difference between complexes, whether the difference is in binding partners, phosphorylation states, or otherwise. Analysis of other potential criteria for differentiating complexes yielded similar results to those discussed below (Fig. S9 and Section 3.3 in Text S1).
We observed a marked increase of drift between simulations with pheromone (and thus signaling activity) as opposed to those without pheromone (Fig. 3B). At peak Fus3 signaling activity (t = 360 seconds), around 80% of all unique complexes were exclusive to one simulation or the other (Fig. 3B). Such small overlap indicates that individual cells utilize different sets of signaling complexes, consistent with the ensemble hypothesis [7], [16]. To confirm that this high level of drift is not an artifact of our chosen parameters, we generated over 1000 rule sets with randomized rate parameters (Section 2.4 in Text S1). In Fig. 3C we see the distributions of drift values among scaffold-based signaling species for both the validated model and models with randomized parameters at peak Fus3 signaling. Although the average random parameter set has somewhat lower drift than observed in our original parameter set, approximately 97% of the drift values from the models with randomized parameters were nonetheless greater than 0.8. The high level of drift among signaling species thus likely arises from the rules and interactions themselves rather than specific rate constants.
While the results in Fig. 3C indicate relatively high levels of heterogeneity at a particular time point, it could be that two different simulated cells utilize the same set of complexes, just at different times during signal transduction. We thus considered the differences between cells based on the union of all the unique complexes they sampled across the time points in our simulations (i.e. the points in Fig. 3B). We found that using the union of complexes across times only reduced absolute drift levels by about 10%, indicating a high degree of diversity between simulated cells across the entirety of the signaling dynamics (Fig. S10 in Text S1).
Our analysis of drift across time points raised the question of whether an individual simulation i maintains a specific set of complexes, or if the set changes over time. To answer this question, we used an alternative drift calculation, termed autodrift: di(t,t+Δt) instead of d(i,j). We found that simulated cells employ rapidly changing sets of complexes during peak signaling times in this model (Fig. 3D). Autodrift increased as a double exponential, with a longest time scale of approximately 0.5 s (Fig. 3D, inset, and Section 3.3 in Text S1). Indeed, within 5 seconds the difference between a cell and its past self achieves levels of drift similar to that observed between two completely independent cells in the population. This is consistent with observations from both modeling and experimental studies of epidermal growth factor signaling in mammals, where a diverse set of phosphorylated species forms rapidly during signaling [47], [48]. The rapid increase in drift also highlights the transient nature of the ensembles of complexes that are generated.
It is possible that the putative ensembles in this case merely represent a set of highly similar (though technically distinct) signaling species that form around a large “core” signaling complex. We thus examined in detail the structures of the scaffold-based species at various time points in our simulations. If a core complex were present, we would expect to see substantial conservation of protein binding patterns (ignoring phosphorylation state) in the set of unique complexes. Though Ste5 dimers are present in ∼70% of species during peak signal throughput, conservation significantly declines as the binding pattern is expanded to include more proteins (Fig. 4A). In fact, not once did we find a Ste5 dimer bound to all its potential interaction partners, indicating that the complex used in the standard graphical depiction of this phosphorylation cascade is one that would very rarely, if ever, occur in simulations of this model (Fig. 1A) [8]–[14].
It is possible that complexes in the ensemble model still assemble around a consistent core structure, just not the traditional representation of a scaffold-based core signaling complex that we intuitively expect (Fig. 1A). Since there are over 3 billion possible scaffold signaling structures in this model, however, we could not search for this core by enumerating all possibilities and looking at conservation patterns as in Fig. 4A. We thus used a straightforward clustering analysis to search for an alternative core structure. The signaling species generated in our model were clustered on the basis of the structural similarity between complexes, represented in this case by the graph edit distance metric, which is simply the number of changes (or edits) that would be required to form one complex starting from another. This distance accounts for differences in the members of a complex (i.e. the removal of a protein from a complex increases the distance) as well as differences in phosphorylation state, etc. (Fig. S12 and Section 3.4 in Text S1).
We implemented a hierarchical clustering algorithm based on this distance. Briefly, the algorithm chooses a representative complex from each cluster, called the “clustroid,” which is the complex with the lowest average graph edit distance to all other complexes in its cluster (Section 3.4 in Text S1). At each level of the hierarchy, the algorithm combines the two clusters whose clustroids are most similar, that is those with the minimum graph edit distance (i.e. the minimum between-cluster distance, or MBCD). This algorithm is initialized with each complex in its own cluster (meaning the complex is its own clustroid) and continues until the original set of complexes is partitioned into a given number of clusters. This number, which we call the “cutoff,” is a free parameter and is relatively arbitrary in our case (Fig. S15 and Section 3.4 in Text S1), so we repeated the clustering algorithm with numerous different cutoff values. We calculated the size of the largest conserved structural pattern as a function of the cutoff value for each cluster that contained ten or more complexes. We found that, on average, this conserved pattern contained less than 2 proteins (Fig. 4B), indicating substantial dissimilarity among clustered proteins; cutoff values producing clusters with 4 or more proteins in the conserved subgraph were very rare (Fig. S15 in Text S1). These results, combined with the dissimilarity between clusters generated from independent simulations (Fig. S13 in Text S1) and the high levels of drift we observe (Fig. 3B–D), underscore the strong ensemble character of this model.
The findings described above indicate that heterogeneous ensembles of complexes can indeed transmit and process extracellular information with levels of noise comparable to those observed experimentally (Figs. 2–4). To understand if machine-like complexes could also produce reliable signaling behavior, we constructed an alternative model with the goal of assembling signaling machines, which we defined to be stable, multi-subunit kinases based around the scaffold Ste5 [1], [5], [9]. Specifically, the machine we focused on consists of a Ste5 dimer, with each scaffold protein bound to a Ste4–Ste20 dimer and two kinases, Ste11 and Ste7 (Fig. 1A). Upon assembly and activation, this decameric structure binds and phosphorylates Fus3 according to standard mass-action kinetics [9].
In contrast to the previous model, we were forced to introduce a priori assumptions (neither experimentally supported nor specifically refuted) in order to generate stable signaling machines. The simplest possible approach would be to create rules and rates that render the desired machine complex incredibly stable. The decamer, however, is essentially never generated in our original model's simulations (Fig. 4A), so a machine model based purely on increasing the stability of the desired complex is unlikely to actually produce such machines in high quantities reliably. As mentioned above, this fact resembles the Levinthal paradox in protein folding: no matter how stable the native state of a polypeptide chain may be, proteins would essentially never fold if they randomly searched for this state on an otherwise “flat” energy landscape [19], [20]. Alternatively, evidence suggests that molecular machines assemble hierarchically in vivo [49], and so we added specific rules that determine the order in which binding and phosphorylation could occur between the scaffold and its associated proteins (Fig. 1B, red arrows). This represents a hierarchical energy landscape (extending the analogy to protein folding), where each consecutive step builds toward the formation of a “native” signaling machine [19]. For example, in the machine model, binding of Ste11 to the scaffold can only take place if Ste5 has dimerized and each scaffold is bound to a Ste4–Ste20 dimer. Beyond these scaffold assembly rules, no other alterations were made to the model.
The resulting rule set is sufficiently complex that it is impossible to directly estimate the number of unique species that the machine model could form. We thus translated this model from Kappa into BNGL and used available BioNetGen tools to calculate the total number of species for this rule set [50]; as with our ensemble model, the Kappa and BNGL versions of the machine model are available as supporting files (“machine.ka” and “machine.bngl”, respectively, in Protocol S1). This analysis indicated that the machine model could only generate a total of 1106 possible scaffold-based structures, a decrease of over 6 orders of magnitude compared to the ensemble model (Section 3.5 in Text S1). The hierarchical assembly rules in this case thus drastically constrain the set of possible species that the model can sample.
As with our original model, we subjected this alternative machine model to parameter variation and confirmed that it can reproduce experimental data (Figs. S6, S7, S8 and Sections 1.8 and 3.2 in Text S1). Although the dose-response and time-course trends of the machine and ensemble models are similar, they exhibit significantly different sets of signaling complexes. As expected, nearly half of all unique scaffold species in the machine model contained the decamer defined above (Fig. 4C), indicating wide conservation of the desired core signaling complex, in contrast to the complete lack of conservation observed in the ensemble model (Figs. 4A and 4B).
The set of species sampled in the machine model also differed dramatically from those produced by the ensemble model. As a gross estimate of this difference, we considered the cumulative number of unique scaffold-based species obtained by a set of simulations; that is, the total number of unique complexes that are found in a group of N simulated cells. In the machine model, this number rapidly approaches a maximum value as N increases, saturating at around 800 after considering only 100 simulations (Fig. 5A). The machine model thus samples about 70% of the 1106 possible scaffold complexes in a population of ∼100 cells. The behavior of the ensemble model is strikingly different, sampling a set of unique structures that is nearly two orders of magnitude greater than the machine model (approximately 70,000, Fig. 5A), and failing to saturate even after considering a population of 600 simulated cells. Although the total number of sampled species across these 600 cells is large, it is only 0.0022% of the 3 billion species the ensemble model could theoretically generate.
As one might expect given the results of Fig. 5A, we observed large differences in drift during peak signal output between the two models. On average, only 55% of unique scaffold complexes were exclusive to one of two simulations in the machine model, as opposed to 90% in the ensemble model (Fig. 5B). As with the ensemble model, we generated 1000 alternative machine models with randomized parameter sets to determine if the level of drift in this case was an artifact of the parameterization of the model. Though the distribution of drift values was fairly wide across these randomized models, in every case we observed considerably less drift than for the validated or randomized ensemble model (Fig. 5B). The rules underlying the machine model thus robustly produce dynamics that one might expect for well-established molecular machines like the ribosome or proteasome: a stable, heavily populated core structure with residual diversity arising from assembly intermediates and the association of substrates and/or regulatory factors.
Since these two models can both reproduce general pheromone-dependent trends, one might ask if it is possible to differentiate machine- and ensemble-like signaling processes directly using available experimental techniques. The most natural approach would be tandem affinity purification in conjunction with mass spectrometry (TAP/MS), which is widely employed as a high-throughput assay for the discovery and analysis of protein complexes [27]. For example, Gavin et al. employed a “socio-affinity” (SA) index designed to extrapolate binary TAP/MS interaction data in order to discover novel “eukaryotic cellular machines” via clustering analysis [27]. To determine whether this technique could discern the nature of in vivo signaling complexes, we characterized the signaling species generated in both the ensemble and the machine models using the SA index [27]. There is a high correlation between the SA scores produced from our two models' sets of species (Fig. 6A); clustering these scores using the commonly employed MCL algorithm [27], [28], [51], [52] results in essentially the same set of complexes (Fig. 6A, inset).
This leads to the question of whether one could ever detect any functional differences between ensembles and machines in a signaling context. Previous work has established the presence of “combinatorial inhibition” [26] (akin to the “prozone” effect [53]) in this particular cascade; increased expression of the Ste5 scaffold leads to a maximal response, past which further overexpression leads to a decline in signal output [12], [40]. We found that the ensemble model reproduces this behavior, while the machine model does not (Fig. 6B). In the ensemble model, the eventual decrease in signal response arises because the high quantity of scaffold proteins lowers the probability of cascade components (say, Ste7 and Ste11) binding the same scaffold dimer [26], [53], and so the rate of signal propagation is drastically reduced. The hierarchical assembly rules in the machine model, however, reduce drift by ensuring scaffold dimers can only bind Ste7 after Ste11 is already bound. Beyond a certain minimal point, increasing Ste5 concentration has no effect, since the only potential scaffold binding partners for Ste7 are already bound to Ste11, and thus can propagate signal.
To test if the difference in Fig. 6B was robust to variations in the rate parameters, we simulated 100 randomized ensemble models and 100 randomized machine models with three values of Ste5 concentration: Wild Type (WT), 12 times WT (12×) and 60 times WT (60×). We used these simulations to calculate the relative change in peak Fus3 activation (ΔFus3pp) between two pairs of scaffold concentrations: WT to 12×, and 12× to 60×. The validated ensemble and machine models both exhibit a positive ΔFus3pp (12× – WT), corresponding to an increase in Fus3 activation (the peak in Fig. 6B); all the randomized ensemble models, and most of the randomized machine models, displayed this same behavior (Figs. S16, S17 and Section 3.7 in Text S1). In the ensemble model, increasing Ste5 to 60× WT concentration decreases response, yielding a negative ΔFus3pp (60×–12×), while the machine model exhibits an approximately constant response across these concentrations (Figs. 6B and C). The randomized ensemble models also universally showed a decrease in Fus3 activation from 12× to 60× Ste5 concentration, indicating that combinatorial inhibition is a robust feature of the ensemble model. The randomized machine models, however, had mostly increases in Fus3 activation between these two concentrations, and in no case did we observe a decrease as large as that observed for the ensemble models (Fig. 6C). The relative lack of combinatorial inhibition in the machine model is thus likely a feature of the rules themselves, rather than the specific parameters chosen.
It should be noted that the machine considered here is an acyclic complex; that is, there are no ring-like motifs in the protein interaction map for Ste5 (Fig. 1A) [53]–[57]. Previous modeling studies indicate that ring-like structures can assemble efficiently into well-defined quaternary structures, at least in certain parameter regimes [57]. Nonetheless, overexpression of a single subunit in a heteromeric ring causes a marked decrease in the concentration of the assembled machine, indicating that ring-like structures can simultaneously exhibit a machine-like character and combinatorial inhibition [53], [56], [57]. We leave full consideration of the interplay between robustness and topology in the evolution hierarchical assembly pathways to future work [56], [57].
The nature of the signaling complexes formed during signal transduction is foundational to how we conceptualize and understand information processing in cells. This is particularly true of scaffolds, whose primary function is to serve as a platform for the formation of multicomponent complexes that transmit signals [9]. The question of whether these complexes align more with the machine or ensemble paradigm is thus crucial for developing a principled picture of the roles scaffolds play. For instance, it has been posited that Ste5 acts to insulate pheromone signals from activating other, related MAP kinase cascades by sequestering active Ste11 in a pheromone-specific complex. This view is inconsistent with the ensembles we observe, however, since those involve appreciable concentrations of free, active Ste11; in contrast, the machine model produces essentially no active Ste11 molecules that are not bound to the scaffold. The capacity of Ste5 to fulfill the role of insulator in this pathway, or the need to posit other mechanisms such as cross-inhibition [8], [9], is thus directly related to the degree of ensemble character the network displays, a fact that highlights the central role that reasoning about quaternary structure plays in developing and evaluating hypothetical signaling mechanisms.
Our findings indicate that certain experimental methods, such as TAP/MS, are ill-equipped to directly resolve the structural details of signaling complexes in living cells. The difficulty in this case lies with the inherently binary nature of co-purification assays: they can tell us that two proteins interact in some way, but they tell us very little about the global structural context of the complexes in which those proteins are found. For example, in our computational TAP/MS experiment, we see that the overall pattern obtained by “tagging” each protein and recording its interaction partners is essentially the same for both the ensemble and machine models (Fig. 6A). This is due to the fact that, while the types of quaternary structures formed varies considerably between the two models (Fig. 4), the probability of observing any given pairwise association between two proteins is essentially the same. Our results thus indicate that it is problematic to construe clusters obtained from TAP/MS data as representing “cellular machines” in the classic sense [1], [27].
In contrast, experimental methods that can capture ternary or higher interactions (i.e. the simultaneous association of three or more distinct proteins) could be used to provide direct evidence for (or against) the hierarchical assembly of a signaling machine. For instance, in the machine model, Ste7 only binds Ste5 after Ste11 is already bound. Observation of Ste7-Ste5 association in the absence of Ste11 binding to Ste5 would thus provide evidence against the type of signaling machine considered here (Fig. 1B). Methods such as fragment complementation assays and fluorescence triple correlation spectroscopy could likely be used to probe these types of ternary association dynamics [29]–[31]. Alternatively, recent advances in single-molecule (super-resolution) microscopy (e.g. methods like PALM and STORM) could potentially track the assembly of machine- or ensemble-like signaling complexes [32]–[35].
While direct experimental tests of the ensemble hypothesis are currently lacking, inherent functional differences between machine and ensemble models can be used to provide indirect evidence for or against a particular paradigm. For instance, the hierarchical assembly rules that are required to reliably construct a functional scaffold-based signaling machine prevent our machine model from replicating the experimental observation of combinatorial inhibition (Fig. 6B) [9], [12], [26]. Our analysis of machine models with randomized parameters indicate that this is likely a general observation: in order to exhibit combinatorial inhibition, signaling networks must have the capacity to sample large sets of complexes, ultimately leading to ensemble behavior (Fig. 6C). Although more work is clearly needed to unambiguously resolve the question of machines vs. ensembles, our findings on combinatorial inhibition indicate that at least some degree of ensemble character is likely present in yeast pheromone signaling. It is also clear that the assembly pathways employed to form machines can have measurable, phenotypic consequences. As a result, even if one could determine experimentally the small set of machine-like complexes employed by some network, making a model that employs these machines, but ignores the mechanisms necessary to generate them [11], [21], may not accurately capture the response of the system to perturbations.
The presence of ensemble character in signaling also highlights a potential evolutionary trade-off between machines and ensembles in terms of their phenotypic plasticity. Considering again the analogy to protein folding, adopting a well defined, thermodynamically stable tertiary structure clearly enables the function of a vast array of protein domains (i.e. the general protein structure-function paradigm) [58]. In some cases, however, it has been posited that “intrinsically unstructured” (or unfolded) protein domains may have a distinct functional or evolutionary advantage: for instance, they may display greater interaction plasticity, binding specifically yet transiently with a large number of protein targets [58], [59]. Similarly, a protein with a robust, stable quaternary structure (i.e. a machine) [1], [7], [16] may be beneficial for the conservation of universal cellular tasks, like protein synthesis and degradation. In the case of signal transduction, however, ensembles may offer greater functional and evolutionary plasticity. For example, modifying Ste5 expression levels produces altered, but nonetheless functional, responses without the need to introduce complex, coordinated mutations to the reaction network's rule set (Fig. 6B) [36]. In this sense, both intrinsically disordered proteins and pleiomorphic ensembles may perform unique intracellular tasks precisely because they involve less well-ordered (tertiary or quaternary) structures. The ensemble character we observe could thus represent a form of weak regulatory linkage among genes, ultimately being responsible for the remarkable capacity of MAPK networks to exhibit different but meaningful phenotypes when they are re-wired, either through synthetic modifications or naturally over the course of evolution [9], [25], [36], [60].
Since machines do indeed form in some signaling networks (e.g. the apoptosome), there is likely a spectrum of structural specificity in the formation of complexes during signal transduction [1], [6], [7]. Indeed, one could modify the machine model presented here to include a finite probability of “off-pathway” binding events (e.g. some chance that Ste7 will bind Ste5 even if Ste11 is not already bound). Such models could exhibit intermediate levels of both drift and combinatorial inhibition (Figs. 5B and 6B); future work on this and related systems will be necessary to understand the particular functional and evolutionary consequences of a particular degree of ensemble-like character in any given system. Nonetheless, our work clearly demonstrates that large, heterogeneous ensembles can indeed reliably transmit and interpret extracellular information [7], [21], [22]. This hints at the existence of a new paradigm for molecular computation, one in which the evolution or engineering of “local” interaction rules allows for robust information processing in the absence of “global” order (i.e. a stable, multi-subunit signaling machine) [1], [5]. Understanding the consequences of this paradigm for robustness [61], plasticity [9], [36] and crosstalk [8] in signaling networks represents a crucial task for the emerging field of systems biology.
The models in this work were simulated using KaSim, a stochastic simulator for rule-based models based on the Kappa language that is capable of stochastically sampling all possible species a given model can generate (Fig. 5B; Section 2.1 in Text S1) [23], [24]. The model is initialized with a set of (mostly) monomeric protein agents and simulated for 1000 seconds without pheromone to generate a steady-state population of N untreated “cells.” We treated the cells with pheromone, and generated a set of N′ independent hour-long simulations from each steady-state starting cell. All of the complexes present in the simulation were recorded at logarithmically spaced time intervals. Compositional drift calculations were performed using these “snapshots;” we only performed this calculation between simulations that started from exactly the same initial conditions (Fig. 3A). We performed similar simulations to determine both dose-response and the time course trends. Further simulation details may be found in Section 2.3 in Text S1.
Simulation data was fit to a set of exponential models using nonlinear least-squares regression. We found that a double exponential function was the best fit for the data upon analysis of the residuals and the statistical significance of the estimated model coefficients. The functional form of the model and the full statistical analysis can be found in Section 3.3 in Text S1.
We focused primarily on the scaffold-based species for the analysis of structural conservation and subsequent clustering. These were defined as any complex that included a Ste5 agent or that could bind a free Ste5 agent. We created a vector notation to uniquely identify any scaffold-based complex to simplify the calculation of the graph edit distance between any two complexes (Figs. S11, S12, and Section 3.4 in Text S1). We then implemented the clustroid-based hierarchical clustering approach described in the main text. Other clustering criteria, such as standard single- and complete-linkage, gave similar results (Section 3.4 in Text S1).
We extracted all the binary interactions from the set of complexes generated by our simulations, artificially creating “bait” and “prey” association data. This computational version of the TAP/MS experimental procedure was used to generate the SA scores [27]. The MCL clustering algorithm [52] was then employed to generate the “functional modules” generally associated with such data sets [51]. More information on the SA score calculation and clustering algorithm can be found in Section 3.6 in Text S1.
|
10.1371/journal.pbio.2004103 | The emergence of the visual word form: Longitudinal evolution of category-specific ventral visual areas during reading acquisition | How does education affect cortical organization? All literate adults possess a region specialized for letter strings, the visual word form area (VWFA), within the mosaic of ventral regions involved in processing other visual categories such as objects, places, faces, or body parts. Therefore, the acquisition of literacy may induce a reorientation of cortical maps towards letters at the expense of other categories such as faces. To test this cortical recycling hypothesis, we studied how the visual cortex of individual children changes during the first months of reading acquisition. Ten 6-year-old children were scanned longitudinally 6 or 7 times with functional magnetic resonance imaging (fMRI) before and throughout the first year of school. Subjects were exposed to a variety of pictures (words, numbers, tools, houses, faces, and bodies) while performing an unrelated target-detection task. Behavioral assessment indicated a sharp rise in grapheme–phoneme knowledge and reading speed in the first trimester of school. Concurrently, voxels specific to written words and digits emerged at the VWFA location. The responses to other categories remained largely stable, although right-hemispheric face-related activity increased in proportion to reading scores. Retrospective examination of the VWFA voxels prior to reading acquisition showed that reading encroaches on voxels that are initially weakly specialized for tools and close to but distinct from those responsive to faces. Remarkably, those voxels appear to keep their initial category selectivity while acquiring an additional and stronger responsivity to words. We propose a revised model of the neuronal recycling process in which new visual categories invade weakly specified cortex while leaving previously stabilized cortical responses unchanged.
| Reading acquisition is a major landmark in child development. We examined how it changes the child’s brain. Ten young children were scanned repeatedly, once every 2 months, before, during, and after their first year of school. In the scanner, they watched images of faces, tools, bodies, houses, numbers, and letters while searching for a picture of “Waldo.” As soon as they started to acquire reading skills, a specific region of the visual cortex of the left hemisphere—called the visual word form area (VWFA)—started to selectively respond to written words. In every child, it was then possible to go backward in time and ask what this region was doing prior to reading. We found that written words invaded a sector of visual cortex that was initially weakly specialized, slightly responsive to pictures of tools, and that lay next to a face-selective region. Reading acquisition did not displace those initial responses but blocked their development, such that face-selective responses became stronger in the right hemisphere. Those results provide direct evidence for how education recycles the human brain by repurposing some visual regions towards the shapes of letters.
| In both human and nonhuman primates, the ventral visual cortex comprises multiple specialized subregions that are involved in the visual recognition of image categories such as objects, faces, or places [1–5]. What is striking in humans, however, is that this mosaic of specialized regions is partially changed by the culture we live in. The acquisition of musical [6], mathematical [7,8], and reading abilities [9] leads to systematic changes in ventral visual organization.
In particular, in all adults who have learned to read—regardless of the writing system—a small region of the left ventral visual cortex within the depth of the left occipitotemporal cortex systematically activates in response to written words [9,10]. This region has been termed the visual word form area (VWFA). Although there is still debate about the exact function of the VWFA, which may vary according to a posterior-to-anterior gradient [11,12], it is widely accepted that the responsivity of this region is an excellent marker of reading acquisition [13]. For instance, a whole-brain comparison of brain activity evoked by letter strings in literate and illiterate adults isolates a specific site at the location of the VWFA, the activation of which is proportional to reading speed [14]. Similarly, the VWFA site appears when comparing functional magnetic resonance imaging (fMRI) images of children who have or have not learned to read, either as a group [15,16] or within the same individual [17–19]. In both children and adults, those changes are accompanied by a massive enhancement and left lateralization of the N170 component of event-related potentials evoked by written words [e.g. 17,20,21–23].
In the present work, we aimed to provide novel longitudinal data on the impact of the acquisition of reading on the representations of other visual categories in the ventral visual cortex. In adults, the VWFA systematically lands at a fixed location relative to a reproducible mosaic of regions partially specialized for objects, faces, bodies, and places [10,14, 24–26]. Little is known, however, about the development of this system in young children. In 9-year-old normal readers, the adult mesial-to-lateral gradient of preferred responses to houses, faces, and words is already present, with the expected left–right hemispheric asymmetries for words versus faces [15], but fMRI studies in younger children have underlined the protracted development of this mosaic of regions [27,28]. In particular, a selective response to faces is often difficult to isolate in the fusiform region in early childhood [27,28, but see 29] although a coarse mesial–lateral functional division at the level of the fusiform lobe can be suspected since infanthood [30]. The response to faces increases with age until late adolescence, while the activation for places and tools appears more stable along childhood [27,28,31,32]. A recent study indeed suggests that the face region is more plastic than the place region and continues to exhibit structural changes until adulthood [33].
In this context, it has been hypothesized that the acquisition of reading takes advantage of the preexisting organization and plasticity of this ventral visual cortex [34]. The theory of neuronal recycling proposes that cultural learning takes advantage of the prior organization of the cortex and repurposes some of its circuitry [35]. This could explain why expertise in reading primarily encroaches on the lateral sector of the ventral visual cortex, where late-childhood plasticity is maximal [33]. A combination of constraints, including preexisting connections to language areas [18,36], sensitivity to line junctions [37], and high-resolution representation of fovea shapes [26] would conspire to single out a specific cortical location as the most appropriate for the visual recognition of written words [38]. Reading acquisition would then displace the preexisting mosaic of visual categories in this region, leading to a reorganization that “makes space” for letter knowledge. In support of those ideas, the comparison of literate and illiterate adults [14,23] has revealed that the responses to written words overlap with those responding to objects, faces, and checkerboards and that as reading scores increase, face responses were slightly reduced in the left hemisphere and strongly increased in the right hemisphere. Similarly, in children, Monzalvo et al. [15] further showed that the right lateralization of the activation to faces increased with reading performance. Behavioral and event-related potentials have further supported the notion of a competition of words and faces for cortical space [23,39–41].
One problem, however, is that those studies relied on a comparison of distinct groups of subjects with variable ages and literacy scores. Such group comparisons are necessarily imprecise. Smoothing and intersubject averaging may lead to an apparent overlap between the cortical responses to different categories, even though those regions actually occupy well-delimited territories in individual subjects. Evidence on the development of reading within individual children is simply lacking. In the present study, our primary aim was therefore to obtain enough longitudinal data on a few individual children that they could be submitted to a single-subject analysis. To this aim, we scanned 10 children longitudinally at 6 different times, spread at approximately 2-month intervals before and throughout the first year of schooling (8 of them also came back for a seventh scan 1 year later). Furthermore, to better understand early cortical maps and their reorganization with reading, we presented the children with a broad array of age-appropriate pictures covering the 6 categories of letters, numbers, objects, faces, bodies, and places.
These longitudinal single-subject data depicting the evolution of ventral visual responses to words and other visual categories should allow us to clarify the topographical changes, dynamics, and cortical competition underlying reading development, and we aimed to clarify the following questions: how quickly does the VWFA emerge during reading acquisition? Does it immediately land at its usual adult location, or does it move during development? Do we observe a transient invasion of broader cortical territories followed by selective shrinking? Do other categories remain stable, or are they shifted away from the site that becomes specialized for letters? Can the VWFA site be predicted by a prior pattern of specialization for other categories, such as faces? Or, on the contrary, do word-specific voxels fall upon a sector that is initially poorly specified? And what is the relation between the development of face and word responses: are these categories competing for the same resources in a plastic region?
This study is part of the project "Etude multimodale en neuropsychologie et imagerie du développement cérébral et cognitif et de ses relations avec la variabilité génétique," approved on June 16, 2011, by the ethical committee (CPP Kremlin Bicêtre, N° 11–008).
We received ethical permission to scan 10 healthy children at approximately 2-month intervals, from the end of kindergarten through the first year of school (July, September, November, January, March, and June; the school year started 15 September). An extension allowed us to rescan 8 of them for a seventh time at the end of the next year of school (June). The 10 children (5 boys and 5 girls) were selected from an initial sample of 14 children who were scanned at session 1 (July, kindergarten). The purpose of that first session was 2-fold: (1) for the children and their parents to realize what a scanning session was and (2) for us to see which children had trouble staying quiet in the scanner. We then selected the first 10 children who seemed quiet enough and agreed to come back throughout the year. All children and both their parents signed a written consent at the first session. Then, at each session, they were asked whether they agreed to continue.
At the first scan, children were aged 6 years 2 months on average (range 5:7 to 6:7). We ensured that they possessed little or no reading ability (number of words read in 1 minute = 0 to 7), thus probably selecting children in the lower half of the normal range. None had any known risk factor of reading impairment in their family history: their development was judged normal by their teacher and parents, and they exhibited normal-range performance in verbal and perceptual intelligence quotient (IQ) subtests (WISC IV) and Raven’s Colored Progressive Matrices. We assessed their reading level with 2 tests: “L’alouette,” a classic standardized French reading test, which consists of reading as fast and accurately as possible a meaningless text of 265 words within 3 minutes [42], and “Lecture en une minute” (LUM), a standardized list of words that children are asked to read as fast as possible for 1 minute [43]. At the end of the first year of schooling, their reading age (“alouette”) was, on average, +1 month (range −9 to +15 months) relative to their civil age, indicating normal development, and this value was −2.4 months (range −17 to +39) at the end of the second year (Fig 1). The number of correct words read in 1 minute (LUM) was 33.4 at the end of the first year (range 16 to 54) and 59.25 at the end of the second year (range 33 to 89), which is within or above the standardized norm at the end of the second year of school, i.e., 36.7 (±15.8) words per minute. We recently received news from 9 of the 10 children. All 9, now in their first year of secondary school (sixth grade), have followed a normal school curriculum without reading difficulties.
Just before each fMRI session, the reading level was assessed with a new list of words similar to the standardized LUM list. We also probed knowledge of the grapheme–phoneme code (BATELEM test) and other abilities affected by reading acquisition, including the following: rapid automatic naming of pictures (RAN), verbal short-term memory (forward and backward digit span; sentence span: correct repetition of sentences of increasing length), phonological awareness (EVALEC test, comprising deletion of the first syllable in 10 trisyllabic words, then of the first phoneme in 12 CVC words and in 12 CCV words [44]), vocabulary level (DEN48 test of picture naming [45]), number reading, and number dictation. Finally, to study the effect of reading acquisition on face processing, the face recognition test from NEPSY [46] was evaluated. After having memorized 16 children’s faces, the child had to point to the studied face among 3 faces, across 16 trials. This test was done immediately after the encoding phase and after a 30-minute delay. Except on the first, sixth, and seventh sessions (1 year apart)—for which we used standardized tests—in the intermediate sessions, we constructed equivalent stimulus materials in order to avoid test–retest effects as much as possible.
For the fMRI paradigm, subjects were presented with stimuli belonging to the categories of houses, objects, faces, bodies, words, and numbers (see Fig 2 and S2 Fig). Two additional categories of high-frequency and low-frequency grids with variable orientations were also presented. Each category comprised 60 different exemplars, either using black and white pictures (for houses, objects, faces, and bodies) or 4-character strings (for letters and numbers). The words were frequent, regular words encountered by young readers, as specified in Manulex, a lexical database compiling the frequency of occurrence of words in 54 scholarly French reading books [47]. Faces were front views of male and female children’s faces. Bodies were standing male and female adult bodies; to avoid face responses, their head fell outside the picture frame (see an example in Fig 2). Objects were pictures of objects frequently encountered in a child’s daily life (scissors, spoons, shoes, etc.). Six subsets comprising 10 exemplars of each category were created, to be successively used in the 6 scanning sessions. The order of the subsets was different for each child.
We used a miniblock design. Blocks of 6 images belonging to the same category were randomly selected and presented during 1 second each, thus forming a 6-second block. Blocks were separated by a variable interblock interval of 2.4, 3.6, or 4.8 s (mean of 3.6 seconds). Thus, a new series of images was presented on average every 9.6 seconds. The order of the categories was randomly chosen, with the constraint that each category was presented 3 times in a functional run (8 categories × 3 repetitions = 24 blocks of 6 images each).
As in our previous work [14,15], we used an easy incidental target-detection task, the sole purpose of which was to maintain attention in all miniblocks, independently of reading or schooling level. Within each block, a target (the picture of the cartoon character Waldo) had a 33% chance to appear, replacing 1 of the 6 images (excluding the first 2 images of the block). Therefore, an average of 8 targets appeared during a run of 24 blocks. Children were instructed to press a button as soon as they detected Waldo. This task was used to keep the child’s attention focused toward the visual stimuli. The total run duration was 3’ 58”. In each fMRI session, 4 runs were acquired, except for the first and sixth sessions, in which only 3 runs were acquired due to the additional tests and sequences proposed to the children in these sessions. In the seventh session, 4 runs were acquired except for 2 children who asked to stop the acquisition after the third run.
After training to remain still in a mock scanner (only for the first session), children were brought to the 3T MRI scanner (Siemens Trio). They were protected with noise-protection earphones, and a mirror system above their head allowed them to see the visual stimuli presented on a screen at the end of the tunnel. The images were viewed from a distance of 120 cm with an approximate view angle of 6 degrees. Stimulus presentation and behavioral responses collection were performed using PsychToolbox (http://psychtoolbox.org), a free Matlab toolbox (MathWorks, Natick, MA, US). In each session, T1 images were first acquired (voxel size = 1 × 1 × 1 mm), then functional images were collected (100 EPI volumes with TR = 2.4 seconds, TE = 30 millisecond matrix 64 × 64 × 40, voxel size = 3 × 3 × 3 mm in each run). At the first, sixth, and seventh sessions, this sequence was completed by diffusion tensor imaging (DTI) and a second fMRI sequence studying audiovisual representations, which are not reported in the current paper. To reduce head motion, the quality of the MRI images was checked after each sequence acquisition and functional run, and verbal feedback was given to the child.
To correct for motion within each run, images were first realigned using the corresponding tool provided by SPM8 (http://www.fil.ion.ucl.ac.uk/spm/), including both estimation and reslicing steps. The target image was the mean of all images, except if movement during acquisition corrupted the mean image. In that case, the first image was used. S1 Fig shows the average amount of detected movement, computed as the maximum absolute value of the three translation and three rotation parameters provided by SPM. Average movement amounted to a few millimeters in translation and a fraction of a degree in rotation, often due to a sudden movement (e.g., cough), whereas the child was quiet most of the time. Each fMRI volume was then visually inspected one by one. Using the Matlab toolbox ArtRepair (http://cibsr.stanford.edu/tools/human-brain-project/artrepair-software.html), images affected by excessive intravolume movement artifacts (stripes, severe shape or size distortion) were replaced by linear interpolation of previous and subsequent images or by nearest-neighbor interpolation when the damaged volume was the first or the last or when several consecutive images were affected [48]. The percentage of rejected images remained low, with the majority of sessions requiring no correction at all (44/68 fMRI runs) or less than 4 images (61/68 = 90%). The mean rejection percentage across the 68 sessions was 0.77% (standard deviation = 1.1%; range 0%–16%).
Corrected images were then ready to undergo the rest of the preprocessing, i.e., slice timing, coregistration to the anatomy acquired on the same session, and normalization. For normalization, the T1-weighted anatomical images were first normalized to the standard European adult MNI template. This step segmented the images automatically into different tissue classes (grey matter, white matter, and nonbrain, i.e., cerebrospinal fluid and skull) using the “New Segmentation” option in SPM8. By averaging those segmented images across all 10 subjects and all 7 sessions, 3 tissue probability maps were obtained. The original T1 images were then normalized a second time, this time using as a target template those average images arising from our own cohort of children. The highly accurate alignment of the 6 or 7 anatomies obtained from the same child was visually verified using the CheckReg tool in SPM8. Finally, the normalization matrix was applied to all EPI images of the corresponding session, with a final resampled voxel size of 2 × 2 × 2 mm.
All behavioral variables assessing reading—which were collected on each fMRI session during the first year of school—except RAN and vocabulary showed a significant increase with days-at-school: reading speed (r2 = 64%, p < 10−13), knowledge of grapheme−phoneme relations (r2 = 85%, p < 10−25), metaphonological performances (r2 = 42%, p < 10−7), backward digit span (r2 = 18%, p = .0004), word span (r2 = 26%, p < 10−4), number and digit reading (respectively, r2 = 14%, p = 0.003 and 22%, p = 0.0001), and number and digit dictation (respectively, r2 = 15%, p = 0.002 and 28%, p < 10−4). Time spent at school affects forward digit span only when the seventh session was included (r2 = 36%, p < 10−7). Fig 1 illustrates this relationship for 2 key variables: knowledge of grapheme−phoneme relations and reading speed (number of words read in 1 minute). Both were low and stable for the first 2 sessions before school, then sharply increased. At the end of first grade, most grapheme-knowledge relationships were mastered by all children, but reading speed remained highly variable.
Two regression models were used to attempt to characterize this variability. First, a multiple regression with both days-at-school and age indicated, unsurprisingly, an effect of the former but not the latter (respectively, p < 10−5 and p = 0.14, r2 = 65% for 6 sessions; and p < 0.001 and p = 0.24, r2 = 69% when 7 sessions were considered). Second, another multiple regression examined whether this effect was mediated by the evolution of 4 key cognitive variables: knowledge of grapheme−phoneme relations, metaphonological capacities, RAN, and effortful short-term memory (backward digit span). The results indicated that days-at-school was no longer significant and that knowledge of grapheme−phoneme relations and RAN mostly contributed to reading speed during the first year of school (respectively, p < 10−6 and p = 0.03, r2 = 81%). Because reading speed thus appears to summarize many aspects of the evolution of reading ability both within and across subjects, this variable was selected as the main regressor to study the evolution of brain activity as in our previous work in adults and dyslexic children [14,15].
Face memory ability also increased with days-at-school during the same period of 2 years (r2 = 26%, p < .10−4). The same regressions as above indicated that (1) days-at-school was a better predictor when pitted against age (respectively, p = 0.04 and p = 0.70, r2 = 24%); and (2) none of the above 4 cognitive variables were better predictors than days-at-school although backward digit span approached significance (p = .07); and (3) the same conclusion was reached when reading speed was pitted against days-at-school (respectively, p = 0.88 and p = 0.02; r2 = 24%). Those results suggest that face recognition may be influenced by schooling yet independently of reading.
During fMRI, only a minimal detection task (detection of a specific picture of a cartoon character Waldo) was required. There were no misses, and the mean reaction time (RT) to the target was 978 milliseconds (882 to 1,004 milliseconds across subjects and 933 to 1,001 milliseconds across sessions). The mean RT was stable across sessions (F[6,52] = 1.7; p > 0.1).
We first examined category-specific activations when pooling across all 7 sessions (Fig 2 and S2 Fig for glass-brain views of the category-specific activations). When compared to the other categories, grids induced larger activation mainly in the calcarine scissures and areas of the dorsal occipitoparietal pathway, probably due to perceived movement induced by the constant change in line orientation (S2A Fig) across the successive images of the block. Because this category was so different from the others, we did not consider this condition further and only compared each category to the mean of the other “pictures” categories (i.e., tools, words, numbers, faces, bodies, and houses). For these visual categories, a mosaic of specific responses to each category was observed in ventral extrastriate areas, similar to that reported in adults. From medial to lateral, the preference shifted from houses to bodies and objects (Fig 2 and S2 and S3 Figs). Tools activated 2 separated ventral foci as described by Hasson et al. [50] in adults. Faces mainly activated both fusiform gyri and amygdala, plus the right superior temporal sulcus. Interestingly, bodies induced a much larger response than faces in the inferior temporal and posterior middle temporal regions, while the converse effect was seen in the medial occipital region (S2 Fig). Numbers did not elicit any larger response relative to the other visual categories. Words, however, elicited an extended pattern of activation, with significant clusters observed not only in the VWFA but also in posterior temporal sulcus, parietal, and inferior frontal regions, all in the left hemisphere (Fig 2).
We also pitted our two symbolic categories—letters and numbers—against each other (S2B Fig). Relative to numbers, words elicited stronger activations in the left frontal (240 vox, pcor < 0.001, z = 5.11 at [−52 10 24]), left parietal (115 vox, pcor < 0.001, z = 4.64 at [−32 −52 44]), and left fusiform regions (56 vox, pcor < 0.001, z = 4.06 at [−40 −56 −8] and z = 4.00 at [−48 −55 −8]), confirming the presence of an early specialization for letters over numbers at the left VWFA site even at this early age. No significant cluster was observed for the reverse comparison (numbers > words).
Second, we examined the changes in those activations across the 7 sessions, spread over the first year of reading acquisition, with an additional measurement at the end of the second year of schooling. Fig 3 shows an example of the evolution of the activation to words and to faces in an individual child (see S3 Fig for the other visual categories). In the group analysis, there was a linear increase with sessions of the activation to words relative to rest in several areas of the left hemisphere: in the anterior part of the fusiform region (VWFA), the occipital area, the posterior superior temporal sulcus, the precentral region, and both inferior parietal areas (see coordinates and Z scores in Table 1). A similar increase with sessions was seen for the activation to numbers relative to rest in the right inferior parietal region (77 vox, pcor < 0.001, z = 4.14 at [28 −60 36]) and for houses at [32 −78 32] (z = 4.16, 56 vox, pcor = 0.002). No significant linear effect was observed for the other categories. No area showed a significantly larger increase to one category than to the others. However, when compared to grids, a specific increase was seen for words in the VWFA (74 vox, pcor < 0.001, z = 4.72 at [−50 −58 −14]). Session-by-session comparisons (Fig 2) indicated that the difference between words and other categories was absent from sessions 1 and 2 (before or at the onset of schooling) and first became significant in session 3 for prefrontal and temporal activations or session 4 for the VWFA, i.e., about 2 or 4 months, respectively, after school onset.
We next used regressions to evaluate whether behavioral measures of reading would predict the evolution of brain activity across sessions, independently of age (see Methods and Fig 4). A significant linear effect of reading speed (variable LUM = number of words read in 1 minute) was found on activation to words relative to rest in left occipitotemporal cortex, at around the VWFA site (75 vox, pcor < 0.001, z = 4.66 at [−42 −66 −16]), at a more posterior occipital area (86 vox, pcor < 0.001, z = 4.72 at [−38 −80 −10]), and in the right cerebellum (101 vox, pcor < 0.001, z = 4.26 at [20 −50 −30]). This correlation was stronger for words than for other categories in left occipital (52 vox, pcor = 0.003, z = 4.58 at [−42 −78 −12]) and occipitotemporal areas (110 vox, pcor < 0.001, z = 5.25 at [−44 −64 −8]), indicating that these activations were reading specific.
In addition, interestingly, activation to categories other than words also correlated with reading speed. For faces relative to rest, an increase of activation with reading speed was found in the right-hemispheric fusiform gyrus, at or near the FFA site (48 vox, pcor = 0.005, z = 6.08 at [38 −52 −10]). Furthermore, this increase remained significant when contrasting faces versus other categories (78 vox, pcor < 0.001, z = 6.67 at [38 −52 −12]). This finding confirms the prior finding in adults (literate versus illiterate [14]) and in children (normal readers versus dyslexic [15]) that reading acquisition correlates with a right-hemispheric shift of face responses. For numbers relative to rest, an increase with reading speed was found in a left occipital area (46 vox, pcor = 0.01, z = 5.00 at [−36 −82 −6]), but no difference with other categories was seen. For other categories, there was no significant correlation. Also, in those regressions, no specific increase in activation was found with age or with the number of days at school, for any category relative to all others.
The temporal profiles of activation to words presented in Fig 2 (bottom row) suggested that, in addition to monotonic increases, there might also be transient activations in the first months of reading acquisition that later reduced or vanished. To evaluate this possibility, we examined a quadratic contrast across the 7 sessions for word activations. The quadratic effect was significant in several left-hemispheric clusters: inferior frontal gyrus pars opercularis (pcor = .043, z = 4.53, 33 vox at [−52 8 24]), anterior cingulate (36 vox, pcor = .026, z = 4.33 at [−6 12 48]), anterior occipitotemporal sulcus (42 vox, pcor = .012, z = 4.27 at [−48 −46 −8]), and anterior insula (34 vox, pcor = .037, z = 4.18 at [−30 18 8]). Left inferior parietal cortex was also present at a less stringent false discovery rate (FDR)–corrected threshold (26 vox, pFDRcor = .025, pFWEcor = .13, z = 4.50 at [−26 −66 32]). Only the left parietal cluster remained significant when compared to the other visual categories or to the grids.
Overall, the group analysis confirmed that ventral occipitotemporal organization in the present group of 6-year-old children was similar to previous studies of adults. Reading acquisition primarily impacted word and number activations in the left occipitotemporal region. Face- and body-selective responses were seen at the earliest age, and face selectivity increased with reading speed in the right hemisphere. To better understand how these responses were organized, we next examined the voxel-specific evolution of category specificity in each subject.
We asked whether the VWFA could be identified in each individual subject using the contrast for activation to words relative to other visual categories, once reading was established (we used sessions 6 and 7 or—in the 2 subjects who could not return for session 7—session 6 only). At our standard threshold (p < 0.001, cluster–FWE-corrected p < 0.05), 8 out of 10 subjects showed a significant cluster at, or near, the classical VWFA location in the left occipitotemporal area (see S4 and S5 Figs for individual brain maps and S1 Table for coordinates). Concerning the last 2 children, 1 had a cluster at the correct location but not significant when FWE-corrected (z = 5.67 at [−50 −62 −16], 40 vox, pcor = .092, p = .035 FDRcor), and the other, who was the second worst reader, showed only a significant occipital activation for the same contrast (83 vox, p cor <.001, z = 5.88 at [−16 −94 −10]) and a very small peak at the VWFA site (z = 3.39, 6 vox at [−46 −56 0]).
To further examine individual responses to the different visual categories and how they were modified by reading acquisition, we restricted the next analyses to subject-specific masks encompassing the fusiform region in the left and right hemisphere (Fig 5). First, we asked whether there was a systematic expansion in the number of category-specific voxels, both for reading and for other categories, during the course of reading acquisition. We therefore computed, for each subject and each session, the number of voxels significantly more activated by a given category than by the others (p < 0.001; hereafter called “specific voxels” for short). We entered these data into an ANOVA with factors of category (6 levels), sessions (1–6), and hemisphere (left and right). The results revealed a main effect of sessions (F[5, 45] = 5.6, p < .001), category (F[5, 45] = 9.0, p < .001), hemisphere (F[1, 9] = 10.1, p = .011), and an interaction session × category × hemisphere (F[25, 225] = 2.1, p = .002), which reflected a significant session × category interaction in the left but not the right fusiform region (Left: F[25, 225] = 1.7, p = .024; Right: F[25, 225] <1). Not surprisingly, only words showed a significant change in the number of selective voxels across sessions in the left hemisphere (F[5, 45] = 5.2, p < .001), reflecting a sudden increase after session 2, i.e., the start of school. No other visual categories showed a significant change of volume over the 6 sessions (Fig 5), in either hemisphere. The only other significant effects concerned hemispheric differences: there was a left-hemispheric asymmetry for words (F[1, 9] = 16.6, p = .003) and tools (F[1, 9] = 32, p < .001), whereas bodies (F[1, 9] = 48, p < .001), houses (F[1, 9] = 14, p = .005), and faces (F[1, 9] = 12.6, p = .006) induced larger volumes of activation in the right hemisphere. No effect of hemisphere was seen for numbers (F[1, 9] < 1).
We then asked whether the emergence of the VWFA modified the location of the responses to other categories. We therefore computed, for each image category and for each session, the distance (in mm) of its activation barycenter from the subject’s mean VWFA barycenter determined in sessions 6 and 7 (see Methods) and entered this distance into an ANOVA with category (4 levels, excluding words and numbers) and sessions (1–6). Only the main effect of category was significant (F[3, 27] = 6.61, p = .002), reflecting the obvious differences in locations of the activations for each visual category. There was no linear effect of session (F[1,9] < 1), no session × category interaction (F[3, 27] < 1), and no effect of session within each category indicating an absence of change in the position of activations relative to the VWFA. This was also the case when we considered the peaks of activation of houses, bodies, and tools relative to the left and right FFA barycenter, suggesting that the peaks of activation of the visual categories were not displaced by the development of the VWFA. Therefore, the development of the activation to words in the left fusiform region during the first year of school did not affect the volume or location of the activation to other categories.
We next asked whether the voxels that become specific for written words in sessions 6 and 7 could already be distinguished by a particular response profile in previous sessions (1 to 5) and notably before reading acquisition. Within our ventral mask, we therefore distinguished RR and nRR voxels, using a voxelwise threshold of p < 0.001 on the words > other visual categories in sessions 6 and 7. On average, we recovered 172 RR voxels (range 39–388 voxels), i.e., 1,376 mm3 representing an average 3.8% of the search volume (range 0.87%–8.57%). We could then go back in time and examine the properties that distinguished RR voxels from nRR voxels in the preceding sessions (Fig 6, left panel).
First, were these voxels already specific for words before school, or did a selectivity for words emerge only after the onset of reading acquisition? A comparison of the mean activation evoked by words versus other categories—in RR versus nRR voxels—revealed no significant difference in session 1, approximately 2 to 3 months before the start of school (t[9] < 1). There was a marginal effect in session 2, which occurred around school onset (t[9] = 2.23, p = .053). A clear difference emerged in session 3 (t[9] = 2.73, p = .023), session 4 (t[9] = 2.40, p = .040), and session 5 (t[9] = 3.46, p = .007). Similar findings were found when analyzing only the RR voxels for a difference in responsivity to words versus other categories (S7 Fig, left panel). Those findings show that, even with a sensitive analysis targeted precisely at subject-specific voxels that ultimately become RR, we could not identify an early responsivity to words prior to schooling.
In this respect, words behaved differently from other categories: when the same analysis was replicated with voxels specific for other categories of visual images in sessions 6 and 7 (294 voxels sensitive to tools, 167 to houses, 150 to faces, and 212 to bodies in the left hemisphere), these voxels were already specific in sessions 1 to 5 (Fig 6 and S6 Fig, all p < .003). Even for numbers—although a reduced set of voxels showed a preference for numbers than for other categories in sessions 6 and 7 (mean = 22 voxels, range = 1–63)—a difference in activation to numbers versus other categories was already found in all previous sessions except session 2 (p = 0.13, for all other sessions, p between 0.025 and <.001). Therefore, only for reading did we see the emergence of a novel cortical selectivity during the first year of schooling (which, in France, focuses almost entirely on reading acquisition).
Second, we asked whether the ultimate preference of RR voxels for words could be anticipated by an early preference of those voxels for another category—thus testing, at the single-subject level, previous suggestions that reading might specifically encroach, for instance, on face-related circuits [14, 15,51]. To this aim, we computed the average activity evoked by each of the 4 nonsymbolic image categories (tools, houses, faces, and bodies) relative to rest in the RR voxels and in each of the 5 initial sessions. We submitted these data to an ANOVA with categories and sessions as within-subject factors. A main effect of category was present (F[3, 27] = 6.91, p = .001), indicating that RR voxels responded to other categories in the following order: tools (mean activation = 0.51), bodies (0.32), houses (0.11), and faces (0.09). Posthoc paired t tests analyzing each possible pair using Holm correction for multiple comparisons revealed that all pairwise comparisons were significant (ps ≤ .01) except faces versus houses, which had a similar weak level of activation in RR voxels (p > .1), and bodies and tools (p = .097), which both produced a larger response in these voxels. Those results indicate that RR voxels showed an initial response to tools and bodies. Importantly, there was no interaction with sessions (F[3, 27] < 1), again indicating that those biases did not change during reading acquisition.
Another important question, however, is whether such preferences are sufficiently unique to RR voxels that they would suffice to determine which voxels ultimately become specialized for reading. We operationalized this question by asking whether RR voxels differed from nRR voxels in their profile of responsivity to nonletter categories (Fig 6). Interestingly, RR voxels, relative to nRR, preferred numbers over other categories early on (Fig 6, left panel, light blue curve). This difference was not significant in sessions 1 (t = 2.25) and 2 (t < 1) but became clear from sessions 3 to 7 (t[9] = 3.10, 3.67, 2.70, 2.6, and 4.62, respectively), suggesting that the development of letter responsivity was accompanied by an additional responsivity to numbers. With respect to other image categories, however, the responses of RR voxels did not differ from those of nRR voxels. Only a small preference for tools tended to be present in RR voxels more than in nRR voxels in session 1 and 5 (t[9] = 2.72, p = .024 and t[9] = 2.27, p = .049). Therefore, on average, RR voxels were not especially distinguished from nRR voxels in their initial commitment to a specific category, with the possible exception of a small bias for tools.
A related question is whether RR voxels are simply less specialized overall. According to this hypothesis, reading would “land” in voxels that are not already strongly committed to a particular visual category and are therefore available for learning. To evaluate this possibility, for each voxel, we also calculated the F-test quantifying any difference between the 4 nonsymbolic image categories (tools, houses, faces, and bodies). We then examined whether this F-test differed for RR versus nRR voxels. Again, no significant difference was found in any session (ps > .05), indicating that RR voxels were not particularly distinguished by a reduced initial commitment. We did find, however, a small but significant interaction with a linear contrast over sessions 1 through 7 (F[1, 9] = 7.26, p = .025), indicating another type of difference between RR and nRR voxels: nRR voxels exhibited a significant linear increase of the F-test, indicating that their selectivity for nonsymbolic image categories increased (F[1, 9] = 15.1, p = .004), while this was not true for RR voxels (F[1, 9] = 2.5, p = .15).
Those findings led us to ask, symmetrically, what were the initial preferences and temporal evolution of voxels that ultimately preferred nRR categories. When we selected voxels that ultimately preferred tools, faces, houses, bodies, or numbers over the other categories, we found that their preferences were temporally stable (unlike what was found for words): as shown in Fig 6, when going back in time, these voxels already exhibited a strong and significant selectivity for their preferred category in the first scanning session and a corresponding lack of responsivity to their nonpreferred categories. For instance, house-responsive voxels in sessions 6 and 7 showed a strong preference for houses in sessions 1 and 2 (ps < .001), accompanied by a mild response to tools (ps > .05) and a significantly smaller response to faces, bodies, words, and numbers (ps < .035) compared to the responses found in non–house-responsive voxels. These preferences were therefore entrenched early on and remained stable or slightly increased over time (Fig 6), as confirmed by the above F-test. Most importantly, voxels selective for tools, houses, faces, or bodies systematically showed an early negative responsivity to words relative to other categories (Fig 6, green curve, ps < .05). This finding implies that, if a voxel was strongly selective for a category other than words, it tended to be less responsive to written words than other voxels—and this antipreference remained stable over the time course of early reading acquisition.
The above conclusions remained when we performed a number of variants of the above analysis. Namely, the stability of visual preferences in nRR voxels was observed when we selected category-selective voxels in sessions 1 and 2 and evaluated their responses in sessions 3 through 7 (S7 Fig), when we considered the entire set of voxels within anatomically defined regions: fusiform gyrus (FG)1 and FG2 [52], and when we performed the same analyses in the right hemisphere.
Averaging over an entire set of RR voxels, as we did in the above analyses, may mask the presence of fine-grained activity patterns that are specific to a given subject or a given category [53,54]. In a final analysis, we therefore used multivariate pattern analyses in order to quantify the stability and the evolution of subject-specific activation patterns within the entire ventral mask and, more specifically, within the VWFA.
We first performed this analysis over the entire ventral mask and probed the reproducibility of the pattern of activation between one session and the next (i.e., over a delay of approximately 2 months) for each category relative to rest (S8 Fig). We performed an ANOVA on the correlation coefficients within category (e.g., correlation between the vectors of activity for faces in 2 successive sessions) versus across categories (e.g., correlation between the vector for faces at session n and for houses at session n + 1) over the 6 sessions of the first school year. As shown in Fig 7A, the within-category correlation was systematically positive (ranging around approximately 0.5–0.6) and systematically higher than the between-category correlation, indicating a robust replicability of the category-specific activation patterns across successive scanning sessions (left hemisphere: F[1, 9] = 58; right: F[1, 9] = 49, ps < .001). This result was observed for each category in both hemispheres (ps < .007), with the sole exception of words in the right hemisphere (F[1.9] = 3.35, p = .1). Interestingly, the activation pattern for bodies (p = .004) and faces (p = .02) was significantly more reproducible in the right hemisphere than in the left, whereas the reverse was true for words and tools (ps = .008). Therefore, this multivariate analysis revealed that activity patterns could be reliably measured in our young children for all categories.
We next examined how those patterns evolved over sessions (Fig 7B). For nonsymbolic images, this reliability was stable over time, whereas for words (and numbers), the reliability increased between sessions 3 and 4. This conclusion was confirmed by an interaction between category, correlation type, and a linear contrast for sessions (F[5, 45] = 2.6, p = .035) in the left ventral visual cortex, due to an interaction between sessions × correlation type for words (F[1, 9] = 6.9, p = .028) and numbers (F[1, 9] = 6.99, p = .027) but not for the other categories, bodies, houses, faces, and tools (all F[1, 9] < 1).
Armed with this analysis, we could next ask whether the VWFA itself is a site that is initially devoid of category-specific patterns or, on the contrary, whether it is already traversed by reproducible activations for visual categories other than words; and if the latter is true, whether those activations change over time. We therefore performed the same analyses as above on voxels restricted to the VWFA (RR voxels defined on sessions 6 and 7). Unsurprisingly, a reproducible pattern of activation was found for words (F[1, 9] = 15.1, p = .004) and numbers (F[1, 9] = 11.4, p = .008) but also, remarkably, for nonsymbolic images (F[3, 27] = 27.7, p < .001) mainly due to tools (F[1, 9] = 31, p < .001) and to a lesser degree bodies (F[1, 9] = 10.8, p = .009; no significant effect for houses and faces, ps > .1). Importantly, the pattern reliability for tools was present in sessions 1 and 2 (F[1, 9] = 12, p = .007) prior to reading and remained stable over time (interaction with sessions, F < 1). Therefore, RR-related voxels contain reliable multivariate activity patterns for visual categories other than words, and the emergence of reliable activation to words (Fig 7D) in the course of reading acquisition did not occur at the expense of other categories.
Similar results were observed for the other category-specific ROIs (defined on sessions 6 and 7). Most importantly, the pattern of activation for words was reproducible in the tool ROIs (left and right, ps < .001), house ROIs (left: p = .011; right: p = .005), left FFA (p < .001), and left-body ROI (p = .02) (Fig 7E and 7F). Therefore, activation patterns were distributed beyond the strict boundaries of their significant category-selective clusters [53]. As discussed below, those results are compatible with a superposition principle according to which visual categories are encoded by overlapping activity patterns over the same voxels.
In 10 individual children, we visualized the emergence of reading circuits by reproducibly scanning them with a battery of visual stimuli during, and 1 year after, the first year of schooling. Prior to schooling, the VWFA could not be detected, although selectivity for faces, houses, bodies, or tools was clearly present. Within the first 2 to 4 months of schooling, the VWFA emerged at the group level and in 8 out of 10 individual children (in the remaining 2, activation was present and significant at the voxel level but not after correction for multiple comparisons at the whole-brain level). The VWFA immediately appeared at its adult location, with a subject-specific topography that was stable over time. An inverted-U curve indicated that—in most of the reading circuit, including anterior VWFA and posterior parietal cortex—the onset of schooling was associated with a peak of activation that then slightly receded over the following months.
Most importantly for theories of reading acquisition, the data allowed us to retrospectively examine what the VWFA voxels were responding to prior to reading acquisition. Reading did not recruit voxels that were selective for faces but systematically encroached on slightly more lateral sectors of cortex, within the left occipitotemporal sulcus, in a region that was not strongly selective to any category but did respond more to tools than to other visual categories. The emergence of the VWFA occurred without radically altering the preferences or the topographical organization of ventral visual responses to faces, bodies, houses, or tools, although at the group level we detected a positive correlation between reading performance and the amount of right fusiform activation to faces.
Those results suggest that the VWFA emerges quickly, at a fixed and constrained location within a well-organized mosaic of preferences, and by superimposing itself onto existing preferences while minimally altering them. These points, which constrain theories of brain development, will now be discussed in turn.
The first question that our study aimed to answer is how quickly the VWFA emerges in children during the course of reading acquisition. In our subjects, the VWFA was undetectable prior to schooling: in our first 2 scans, we observed no specific response to letter strings relative to other visual categories. This is a negative result and should therefore be taken cautiously, but it is strengthened by the fact that we could easily detect reliable preferences for all nonsymbolic visual categories (faces, houses, bodies, and tools). The absence of the VWFA could be due to the fact that we selected children with very little prior knowledge of letters (Fig 1) in order to control their exposition to letters and reading when they entered school. Other studies show that preschoolers who possess some knowledge of letters already exhibit letter-specific steady-state responses that are detectable in a few minutes of electroencephalography (EEG) recording [22].
Our results suggest that, by 2 to 4 months after school onset, a change in the volume of selective activation to words is already detectable (Fig 5) and is associated with the establishment of a stable pattern of subject-specific activation for words relative to other categories (Fig 7). This finding is entirely coherent with another longitudinal study [17] in which left occipitotemporal responses emerged in fMRI and event-related potentials once preschoolers had been exposed to a few weeks of training with the GraphoGame, a software game that teaches grapheme−phoneme correspondences. Lochy et al. (2016) and Maurer et al. [21,55,56] likewise observed a rapid growth of left-lateralized event-related potentials evoked by letter strings (N170) in the course of reading acquisition, both in children and in adults acquiring a new script. Our results are compatible with those prior findings in as much as they show that VWFA responses emerge within the first few months of reading acquisition.
Although reading acquisition was necessarily confounded with age in the present within-subject study, we still observed an effect of reading speed on the VWFA activation even when the effect of age was taken into account (Fig 4A). In other between-subject studies, the 2 variables have been clearly decorrelated, and the results indicate that age alone does not suffice to induce the observed changes. For an equal age of approximately 6 years, children who have already learned to read by themselves prior to formal schooling already exhibit a specialized VWFA, contrary to those who have not [57]. Even adults, in the absence of schooling, do not develop a specific ventral visual response to letter strings relative to other visual categories [14,23].
The remarkable speed with which word-specific responses emerged in 6-year-old children fits with other findings indicating a high degree of plasticity in children’s ventral visual cortex. Indeed, the fusiform cytoarchitectonic area FG2, where the FFA and VWFA are located, shows a prolonged development and does not fully mature until late childhood, in contrast with the nearby area FG1 [33]. Its plasticity supports not only reading acquisition but also other forms of visual expertise. For instance, the acquisition of music reading, starting at 3 to 4 years of age, has been shown to induce a large activation of lateral ventral temporal cortex to printed music in professional musicians and to induce a displacement of the nearby VWFA [6]. In adults, fusiform plasticity is sufficient to acquire a novel expertise for birds and cars [58], greebles [59,60], children’s faces in teachers [61], or Braille reading in nonblind adults [62]. Its plasticity might be reduced in adults compared to 6-year-olds. In exilliterate subjects who learned to read as adults, the VWFA is present but only moderately activated in proportion to reading speed, and written words induce a coarser pattern of activation extending into bilateral ventral visual regions [14]. In a recent single-case study [63], we used longitudinal fMRI in a single illiterate subject to follow the trajectory of adult literacy acquisition over the course of 2 years. Unlike the steep nonlinear increase shown here in children, this person’s VWFA emerged slowly and with a continuous increase over a period of several months, paralleling a slow increase in behavioral reading ability. Therefore, the fast changes induced by reading acquisition that we observed here betray the intense plasticity of children’s immature ventral occipitotemporal cortex. This conclusion is also confirmed by a recent animal model of symbol recognition that shows that, while both juvenile and adult monkeys can acquire a behavioral ability to recognize Arabic numerals and letters, only the juveniles develop a category-specific response to such symbols in inferotemporal cortex [64].
Most previous studies of reading acquisition have either been cross-sectional [14,23,65–67] or longitudinal with few data points spread over a long time period, typically 6 months to several years [19,21,68,69]. Here, we aimed to clarify the developmental trajectory of the VWFA and other areas of the reading circuit by obtaining detailed single-subject measurements spaced every 2 months around the onset of literacy. Theories of child development have contrasted several accounts of the emergence of functional brain specialization (reviewed in [70]): according to a maturational account, the adult reading circuit should progressively emerge at its normal location, with increasing size and intensity as increasingly larger populations of neurons become tuned to written words. An alternative account, the interactive specialization model, predicts that the reading circuit should be initially disorganized, diffuse, distributed over broad cortical territories, and that it should dynamically change in a stochastic manner until it settles into a minimal set of regions the interactions of which optimally fit the task. Finally, the skill-learning model postulates an initial phase of effortful processing, relying on nonspecific prefrontal and parietal areas, followed by a progressive reduction of this activity and its transfer to specialized posterior areas.
Our data show features of both the maturational and the skill-learning models. On the one hand, the VWFA landed immediately at its adult-like location: just 2 months after school onset, it was already detectable, at its final location (e.g., Fig 3), and with a fine-grained topographic pattern that showed little or no change over the next 1.5 years (Fig 7D). Those findings argue against a stochastic search or dynamic change in interregional interactions and instead suggest the rapid maturation of a highly constrained brain circuit (the nature of those constraints is discussed below). On the other hand, several regions of the reading circuit showed an inverted-U activation as a function of time, peaking around the first year of schooling (Figs 4 and 5) and then decreasing steadily. In fact, activation at several parietal and inferior frontal sites eventually decreased down to an undetectable level (Fig 3). In the left occipitotemporal cortex, the volume of RR activation increased sharply and then was nearly halved (Fig 5C). This volume change arose because activation in the posterior part of the VWFA followed an inverted-U pattern (Fig 2), while activation at the peak of the VWFA proper steadily increased in parallel to behavior (Figs 2 and 6). In other words, there was a progressive concentration of activation in a specific ventral area and a progressive disengagement of parietoprefrontal areas, in agreement with the skill-learning model.
This developmental pattern fits with previous evidence that reading progressively switches from a slow, serial, effortful mode to a fast, parallel, efficient mode. The posterior parietal cortex, in particular, participates in a top-down attention circuit that has been implicated in letter-by-letter reading, both in adult patients with pure alexia following lesion at or near the VWFA [71] and in normal adult readers confronted with unusual formats such as vertical or s t r e t c h e d words [72]. Its strong activation early on in literacy acquisition is therefore congruent with the classic observation that young readers show a strong effect of word length on reading time [73,74]. This effect progressively vanishes as reading becomes automatized and may therefore explain the inverted-U profile of parietal activity that we observed.
The present data thus confirm the existence of a parietal circuit for effortful reading and underline its transient but important contribution during the early phase of reading acquisition. It fits with the suggestion that some children with developmental reading deficits suffer from an “attentional” subtype of dyslexia that prevents them from focusing on each letter in a sequential left-to-right order and avoiding illusions, conjunctions, and letter migrations from previous words [75–78]. In the future, it would be interesting to use the present fMRI task with subtypes of dyslexic children and examine whether the attentional subtype shows disorganized parietal activation while the more classical phonological subtype shows left temporal impairments, as shown in Peyrin et al. [79].
Another goal of our study was to examine whether a specific profile of functional brain activity predicts which voxels are going to become specialized for written words during reading acquisition. By identifying the VWFA approximately 1.5 years after the onset of schooling and then examining its activity in prior fMRI sessions, we could ask this question within individual subjects at a single-voxel resolution. This is an important improvement over our previous between-subject study of literacy [14], in which intersubject averaging could have caused an artificial overlap between unrelated areas.
The results indicated that, prior to schooling, the voxels that will ultimately become the VWFA are not very strongly specialized for faces, bodies, houses, or tools (Fig 6 and S6 Fig). They are clearly unresponsive to faces and show only a modest response to tools, clearly smaller than the peak response found at a slightly posterior site. Therefore, the VWFA lands at an initially relatively uncommitted cortical site, the main functional characteristic of which seems to be a low responsivity to pictures. This finding suggests that ventral visual cortex becomes progressively committed, first to visual categories such as faces and places [30] and later to cultural acquisitions such as letters and numbers and that the latter wave of cortical specialization is constrained to sites that were left partially or totally uncommitted by the first wave.
Additionally, the present findings confirm that the VWFA emerges at a systematic location relative to other functional landmarks, i.e., lateral to the fusiform face responses (as previously reported in [80]) and overlapping with the most anterior part of the lateral object-responsive cortex (Fig 2). Because the VWFA location is so reproducible, factors other than the mere lack of commitment to other categories are likely to play an important role in its spatial delineation [38]. Candidate factors that have been identified in other studies include (1) cytoarchitectony and extended plasticity [33], (2) preferred connectivity to a dedicated set of left-hemispheric targets that include classical spoken language areas [18,36,81], (3) high-resolution foveal bias [26], (4) preference for high spatial frequencies [82], (5) preference for line junctions that are characteristic of letter shapes [37], and (6) efficient detection of invariant shape features [38,83]. Animal studies in which juvenile monkeys were trained to recognize letters, cartoon faces, or tetris-like pentominos indicate that each of these categories landed at a dedicated site that was reproducible across monkeys, suggesting the existence of a biased proto-organization prior to learning [84]. Human adults are able to learn a new script based on human faces and activate the left, not the right, fusiform region for the facefont words, confirming the strong left bias for phonetic-based script, but they were less efficient than the group trained with Korean fonts, and the location of the activation was slightly displaced in an anteromedial position relative to the location of the VWFA [85].
A third goal of our study was to probe the organization of ventral visual cortex prior to reading and to examine how it is changed by literacy. As a selectivity for words emerges in occipitotemporal cortex, does it lead to a reorganization of the responses to other categories?
By 5 to 6 years of age, our results indicate that the mosaic of specialization for visual categories that has been reported in adults [4,86] is already largely in place and changes very little in the course of almost 2 years. There was an initial controversy as to whether the cortical representation of faces—as opposed to that of objects and houses/landscapes—develops relatively late in childhood [27,28] or, on the contrary, is present early on [29,30]. Our results concur with the latter studies in showing that, by the age of 5, category selectivity is already deeply entrenched. In fact, a recent study showed a clear difference in responses to faces versus places in 3- to 8-month-old infants [30], and this was corroborated by a longitudinal fMRI study of juvenile monkeys [87]. In both studies, however, the selectivity was initially imperfect, for instance, showing no clear selectivity for faces over objects in human infants, and the longitudinal study in monkeys indicated that highly selective face patches emerged during the first year of life [87].
Once these regions are in place, do they change with reading acquisition? Our results uncover that, in fact, they are remarkably stable over time. Word selectivity appears against a background of largely unchanged responsivity in nearby voxels specialized for faces, bodies, tools, or places (Figs 2 and 6). Even within the same voxels that ultimately become the VWFA, multivariate responses to tools stay essentially unchanged during literacy acquisition (Fig 7).
The results therefore strongly question the hypothesis that words and faces share the very same cortical circuitry [51,88]. There is, in fact, much evidence that face and word recognition engage nearby but systematically distinct cortical sectors in normal adult readers [14,80]. Neuropsychological dissociations have also been reported between both domains, both in adult subjects [89–92] and in developmental cases [93,94]. The present study is, we believe, the first to directly demonstrate that orthographic representations develop in ventral visual cortex without encroaching directly on preexisting face cortex but rather by invading nearby but more lateral and relatively uncommitted ventral visual cortex.
The results are, however, compatible with a broader view of cortical competition and neuronal recycling [35]. According to this model, word and face recognition emerge at distinct cortical locations, possibly because of their distinct connections to distant areas [18,36], but they both rely on a similar hierarchical architecture common to all ventral visual cortex and therefore compete for expansion into the same overall region of cortex during development [14,35]. During the acquisition of literacy, the VWFA emerges at a site just lateral to the FFA, thus blocking the further growth of face responses in the left hemisphere and forcing them to preferentially develop in a right-lateralized manner [14,15]. Several experimental results fit with this view. First, Golarai et al. [28] observed that, across development, the peak FFA activation evoked by faces does not change with age in either hemisphere. What increases with age is the activation evoked by faces in peripheral voxels within concentric shells that center on the FFA peak—the FFA slowly expands. Second, in a previous study of adult literacy [14], we found that, in adults with variable degrees of schooling and literacy, greater reading scores were associated with a small decrease of face responses at the usual left-hemispheric site of the VWFA. It is crucial to remember that this finding was based on a group comparison and therefore does not contradict the present within-subject results. Finer-grained analysis, inspired by Golarai et al. [28], in fact revealed that the peak response to faces was unaffected by literacy: it was only in voxels that lay at a distance of 8 or 12 mm from the FFA peak that a difference between literates and illiterates was observed (Figure S6 in [14]), again suggesting that cortical competition occurs at the periphery, where literacy shifts cortical boundaries. Third and finally, our adult study also showed a highly significant increase in right-lateralized fusiform activation to faces as the literacy score increased, suggesting that literacy competed with the development of face responses in the left hemisphere [14]. This finding was also observed in 9-year-old normal readers compared to dyslexics [15] and replicated in the present study (Fig 4). It is corroborated by several behavioral and event-related potential studies (for a review, see [13,23,40,41]).
Our study also probed the development of responses to written Arabic numerals. We expected to observe the emergence of number-related responses lateral to the VWFA, at a site known as the visual number form area (VNFA) and more responsive to digits than to letters in educated adults [7,8,95]. The results, however, were weak. There was no evidence for the emergence of any region specialized for numbers. Instead, we found that ventral visual voxels responsive to numbers were also responsive to letters (Fig 6, S3 and S6 Figs). Reading speed correlated with an increased fMRI response to numbers but only at an occipital site posterior to the VWFA (Fig 4). The fact that we observed no clear separation between number-form and letter-form areas, as it was also observed in younger 4-year-old children in Cantlon et al. [29], nor any intraparietal activation specific to numbers is compatible with the hypothesis that, at a young age, digits are primarily processed as readable letter-like symbols but not as meaningful quantities. At least, such quantity processing does not seem to be automatically engaged when the task does not require any calculation but a mere intruder detection (“find Waldo”) as was the case here. Furthermore, we used 4-digit numbers (matched in length to the words) that probably challenged children’s quantity comprehension. Our results are fully compatible with behavioral studies of the number-size interference effect, which indicates an absence of automatic quantity processing in first graders [96].
Finally, beyond the overlap of numbers and letters, our results on visual categories clearly indicate that, in our children, category selectivity is not absolute. Even within subject-specific voxels that were defined by their selective response to one category relative to all others, there were highly detectable differences between the other nonpreferred categories, both in average activation (Fig 6 and S6 Fig) and in multivariate activation pattern (Fig 7 and S8 Fig). This result, which was first observed in adults [53], suggests that at the relatively modest resolution used here (3 mm isotropic), neural responses to different visual categories are spatially intermixed. It does not, however, exclude that entirely selective patches would emerge in higher-resolution scans, particularly for faces [97].
Fig 8 presents a schematic model of the emergence of ventral occipitotemporal organization in relation to literacy. This model is meant as a graphic summary of our findings and an illustration capable of accounting for the present and several earlier observations within a single framework [14,28–30,53,98–101].
The model starts with the observation that, prior to schooling, category selectivity for faces, houses, and tools is already present (as well as responses to bodies, which are omitted from the Figure for simplicity). Such macroscopic category-selective fMRI responses probably arise from the existence of columns or patches of neurons responsive to each category and/or its visual features and clustered in a reproducible sector of cortex. Indeed, the existence of such dedicated cortical patches was directly demonstrated in monkeys (e.g., [97,99,101]). Developmental fMRI in humans [27–30] and monkeys [87] suggests that such category selectivity emerges within the first year of life and progressively expands in subsequent years. Therefore, we postulate that, prior to reading, the child’s ventral visual cortex comprises a mosaic of specialized sites as well as other more labile sites (respectively appearing in color and in gray in Fig 8).
The key result of the present study is that, with education to literacy, responses to written words emerge at a fixed location within this mosaic (the VWFA), at initially weakly specialized sites, and without altering their (weak) preexisting responsivity to other images. This is captured in Fig 8 by assuming that words invade the patches that were left labile at the time of schooling and lie near those already activated by objects and faces. Within the same fMRI voxel, we may therefore find patches of neurons selective for words and for other images, particularly objects. This “superposition principle” may explain why, throughout the emergence of the VWFA, across 7 scans, we continuously remain able to decode a stable representation of objects within the same voxels (Fig 7B and 7D).
For simplicity of illustration, the superposition principle is illustrated in Fig 8 at the level of adjacent patches, each specializing for a given category. However, superposition may also exist within the same populations of neurons. Neurophysiological studies have revealed neuronal vector codes whereby the same neurons may be engaged in the simultaneous coding of several features of the stimuli along orthogonal vector dimensions [102,103], compatible with the hypothesis that a given neural population may contain multiple superimposed or multiplexed codes in a high-dimensional space [86,104,105]. According to this view, word responsivity may emerge through the progressive differentiation of new principal axes within a preexisting neural population, without altering its pattern of responsivity to other categories. Future studies could use high-resolution fMRI (<1 mm) to attempt to separate the two models of cortical specialization (dedicated patches versus overlapping vectors).
Fig 8 also illustrates why, in the course of the development, superposition may give way to cortical competition. In literate children, the progressive dedication of an increasing number of neuronal patches to written words progressively prevents the expansion of the nearby object and face patches, which occurs in illiterates in the absence of schooling. The model nicely explains how (1) the VWFA emerges at a cortical site that is unresponsive to faces in young children; and yet (2) at a later age, this site shows a greater response to faces in illiterates than in literate subjects [14] and in dyslexics who are not able to acquire fluent reading than in normal 9-year-old readers [15].
Our data indicate a remarkable degree of stability and reproducible order in the organization of children’s ventral visual cortex in at least 2 respects. First, prior to reading, the ventral mosaic of preferences for faces, houses, tools, and bodies is already in place and reproducible across individuals; and second, during reading acquisition, word responses quickly emerge at a reproducible location distinct from those already committed to other preexisting categories. Such reproducibility implies that the architecture of human ventral visual cortex must be strongly preconstrained, and the precise identification of those constraints, whether cytoarchitectural or connectional, is an important goal for further research. Given such constraints, it should perhaps not be surprising that a novel visual category as different as letter strings, which involves specific computational and connectivity requirements, lands at its own distinct location. We may also reverse the reasoning and wonder whether the shapes of the most successful cultural objects, such as letters and numbers, were selected across cultural evolution in order to be quickly learnable [34]. If so, one constraint might have been to avoid direct cortical overlap with any preestablished categories such as faces or tools (see [85] for a nice example in adults trained to a new script). Therefore, biological constraints may explain why the evolution of scripts systematically involves a progressive abstraction away from iconicity and towards abstract shapes (e.g., from hieroglyphic to demotic in ancient Egypt). The success of education might also rely on the right timing to benefit from the highest neural plasticity. Our results might also explain why numerous academic curricula, even in ancient civilizations [106], propose to teach reading around 7 years.
|
10.1371/journal.pcbi.1004220 | Sharing and Specificity of Co-expression Networks across 35 Human Tissues | To understand the regulation of tissue-specific gene expression, the GTEx Consortium generated RNA-seq expression data for more than thirty distinct human tissues. This data provides an opportunity for deriving shared and tissue specific gene regulatory networks on the basis of co-expression between genes. However, a small number of samples are available for a majority of the tissues, and therefore statistical inference of networks in this setting is highly underpowered. To address this problem, we infer tissue-specific gene co-expression networks for 35 tissues in the GTEx dataset using a novel algorithm, GNAT, that uses a hierarchy of tissues to share data between related tissues. We show that this transfer learning approach increases the accuracy with which networks are learned. Analysis of these networks reveals that tissue-specific transcription factors are hubs that preferentially connect to genes with tissue specific functions. Additionally, we observe that genes with tissue-specific functions lie at the peripheries of our networks. We identify numerous modules enriched for Gene Ontology functions, and show that modules conserved across tissues are especially likely to have functions common to all tissues, while modules that are upregulated in a particular tissue are often instrumental to tissue-specific function. Finally, we provide a web tool, available at mostafavilab.stat.ubc.ca/GNAT, which allows exploration of gene function and regulation in a tissue-specific manner.
| Cells in different tissues perform very different functions with the same DNA. This requires tissue-specific gene expression and regulation; understanding this tissue-specificity is often instrumental to understanding complex diseases. Here, we use tissue-specific gene expression data to learn tissue-specific gene regulatory networks for 35 human tissues, where two genes are linked if their expression levels are correlated. Learning such networks accurately is difficult because of the large number of possible links between genes and small number of samples. We propose a novel algorithm that combats this problem by sharing data between similar tissues and show that this increases the accuracy with which networks are learned. We provide a web tool for exploring these networks, enabling users to pose diverse queries in a gene- or tissue-centric manner, and facilitating explorations into gene function and regulation.
| Tissue-specificity, in which cells perform different functions despite possessing identical DNA, is achieved partially through tissue-dependent mechanisms of gene regulation, including epigenetic modification and transcriptional and post-transcriptional regulation [1–3]. These complex programs of control produce different gene expression programs across tissues, with most genes showing statistically significant differential expression [4, 5]. These differences can have significant consequences: tissue-specific genes are especially likely to be drug targets [6] and tissue-specific transcription factors are especially likely to be implicated in complex diseases [2, 7, 8]. Understanding these differences is also essential for understanding pleiotropic genes, and for interpreting studies in which genomics data can only be collected for an accessible or a proxy tissue (such as use of blood in studying psychiatric disorders [9–11]).
Tissue-specific mechanisms of control may be captured by co-expression networks, in which two genes are connected if their expression levels are correlated across a set of individuals. In such a setting, genetic or environmental differences across individuals serve as small perturbations to the underlying regulatory network, resulting in correlation between genes’ expression levels that are consistent with regulatory relationships. Co-expression networks provide insight into cellular activity as genes that are co-expressed often share common functions [12], and such networks have been widely used to study disease [13–15].
The Genotype-Tissue Expression (GTEx) consortium dataset [16] provides an opportunity to study such co-expression networks for an unprecedented number of human tissues simultaneously. However, many of the profiled tissues have fewer than a dozen samples, too few to accurately infer the tens of millions of parameters that would define a co-expression or regulatory network. One solution would be to combine all available samples and learn a single consensus network for all tissues, but this would offer no insight into tissue-specificity. On the other hand, inferring each network independently ignores tissue commonalities: tissue networks share far more links than would be expected by chance, and learning links across multiple tissues is less noisy than learning links using a single tissue, even when using the same number of total samples [12].
Here, we use a novel algorithm, GNAT (Gene Network Analysis Tool), to simultaneously construct co-expression networks for 35 distinct human tissues. Using a hierarchy which encodes tissue similarity, our approach learns a network for each tissue, encouraging tissues that are nearby in the hierarchy to have similar networks. Hierarchical transfer learning has been shown to improve power and accuracy in previous work [5, 6, 17, 18]. We propose a novel hierarchical model along with a parameter optimization method designed for large-scale data, and apply it to the GTEx data. We show that our method infers networks with higher cross-validated likelihood than networks learned on each tissue independently or a single network learned on all tissues. Our method is applicable to any dataset in which sample relationships can be described by a hierarchy—for example, multiple cancer cell lines or species in a phylogenetic tree. The complete code for our method is available as S1 Data.
We analyze the resulting networks to make several novel observations regarding principles of tissue-specificity. We propose multiple metrics for identifying genes that are important in defining tissue identity, and demonstrate that such genes are disproportionately essential genes. We show that tissue-specific transcription factors, which are central hubs in our networks, link to genes with tissue-specific functions, which in turn display higher expression levels. We identify 1,789 gene modules that are enriched for Gene Ontology functions, and show that enriched modules that are upregulated within a tissue are often instrumental to tissue function. We also show that modules which occur across tissues are especially likely to be enriched for Gene Ontology functions, and that these functions tend to be those which are essential to all tissues. The results presented here, including all the networks and gene modules, can be interactively queried through our web tool [19]; the genes and modules identified provide a basis for future investigation.
The results we report here are based on application of the GNAT algorithm to 1,559 samples from 35 tissues in the GTEx dataset. In each sample, we analyzed expression levels for 9,998 genes (Methods; S1 Table).
The goal of our algorithm was to construct co-expression networks that captured both tissue-dependent and tissue-shared relationships between genes. In order to increase statistical power and accuracy when inferring such relationships in tissues with limited sample sizes, it used a two-stage transfer learning framework to construct networks for all tissues simultaneously. The first stage of the algorithm constructed a hierarchy over the tissues. The second stage optimized the network for each tissue using a method that encouraged fidelity to the expression data, sparsity in the networks, and similarity between networks that were nearby in the hierarchy.
1. Learning a hierarchy. A tissue hierarchy was constructed using agglomerative hierarchical clustering on the mean gene expression levels for the 35 tissues (Fig 1). Since the rest of the algorithm was independent of the construction of the hierarchy, the method would also work with a hierarchy based on prior knowledge or on some other measure of dataset similarity.
2. Learning networks based on the hierarchy. We modeled the network for each tissue in the hierarchy using a Gaussian Markov Random Field (GMRF), a standard model in computational biology and image processing [20–22]. GMRFs model gene expression with a multivariate Gaussian distribution; we projected the samples for each gene onto a Gaussian (Methods) so this modeling assumption was reasonable. GMRFs are parameterized by an inverse covariance matrix S(k) (where k denotes the kth tissue) whose zero entries indicate pairs of genes that have expression levels which are conditionally independent given the expression levels of the other genes. These entries correspond exactly to direct connections between genes in the GMRF; other genes may still be connected through longer paths in the network. To encourage zero entries and diminish the number of links in the network, GMRFs maximize the convex Gaussian log likelihood plus an L1 sparsity penalty:
n ( k ) 2 ( log det S ( k ) - t r ( S ( k ) Σ ( k ) ) ) - λ s ( k ) ‖ S ( k ) ‖ 1
where n(k) is the number of samples and Σ(k) the empirical covariance matrix for the genes in tissue k, and λ s ( k ) is a sparsity parameter. The sparsity makes the networks more interpretable and computationally tractable.
We extended this method by constraining the matrices S(k) in tissues that were nearby in the hierarchy to have similar entries, creating similar networks, using an L2 penalty that penalized differences between the S(k). We used an L2 penalty rather than an L1 penalty because it allowed us to develop a fast parallel algorithm for optimizing the objective function (Methods). This transfer learning framework proved especially valuable for tissues with very few samples, for which we would otherwise lacked sufficient statistical power to infer co-expression networks. For example, we had only about two dozen samples for each of the 13 brain tissues in the GTEx dataset—too few to learn networks with 50 million parameters—but because all the brain tissues were closely related in our hierarchy, by adaptively sharing samples for related brain tissues we were able to make more robust estimates of co-expression. We provide a schematic illustration of our algorithm in Fig 2.
Previous work suggests the promise of using transfer learning to learn multiple genetic networks [18, 20, 21, 23]; hierarchical models have also been used more broadly throughout biology, for example to study phylogenies [24]. [18] used prior knowledge of a hierarchy of cancer cell types to learn a network for each cell type. Their method, however, relied on a hand-specified hierarchy, which would only be feasible if the number of datasets was smaller than the 35 in the GTEx dataset, and though successful in simulation was never shown to improve on prior methods on real data. [20] and [21] learn networks for multiple datasets using shrinkage between precision matrices, although they do not use a hierarchy and simply use a single shrinkage parameter. Additionally, none of these methods were designed to work on the large number of tissues included in the GTEx dataset, because such data has not been previously available. Importantly, our choice of optimization objective allows parallel optimization of all 35 tissue networks, which is critical for scaling to a large number of tissues. In contrast, the methods described in [18] and [20] cannot be easily parallelized and thus will not scale to the GTEx dataset, as we confirmed by testing their code on simulations with 35 tissues but far fewer genes than we use in our analysis (n = 10 versus n = 9998). Adapting our algorithm to the scale of the GTEx data required several further methodological innovations (Methods). For example, selecting a sparsity parameter for each of the 35 datasets using cross validation would have been prohibitively slow, so we developed a faster heuristic.
We used 5-fold cross-validation to evaluate our algorithm: for each tissue, we randomly divided our samples into five groups, learned networks based on samples from four of the five groups, and measured the accuracy of each network (quantified by the log likelihood on the held out test data) using the remaining group. We compared the performance of our method to two baselines: learning a network for each tissue independently, or learning a single network for all tissues. We observed a higher log likelihood on the held out test set using our approach as compared to the two baselines on three different gene sets of increasing sizes (Fig 3), indicating that the transfer learning approach resulted in a more robust estimation of the networks.
We confirmed the accuracy of our learned networks in two ways. First, we evaluated agreement with two previous datasets. When we compared our networks to the co-expression database COEXPRESdb [25], pairs of genes we predicted to be linked had expression levels that were 2.6 times as correlated as genes we did not predict to be linked (p < 10−6, 2-sample KS test). To analyze tissue-specificity, we also compared our networks to TS-CoExp [12], which provides lists of tissue-specific co-expressed genes. Genes we predicted to be linked in a tissue were 10.5 times more likely to be linked in the corresponding TS-CoExp tissue than genes we did not predict to be linked (p < 10−6, χ2 test). Links in the TS-CoExp database that were specific to a tissue were 2.1 times more likely to appear in our networks for the tissue than links in the TS-CoExp database that were not specific to that tissue (p < 10−6, χ2 test). (We compared all these numbers to the baseline of the learning the networks independently, which yielded slightly higher agreement with TS-CoExp and virtually equivalent agreement with COEXPRESdb. We speculate that the higher agreement with TS-CoExp is due to the fact that the TS-CoExp networks were also learned on tissues independently.)
Second, using Gene Ontology [26], we found that genes that were linked in our networks were likely to represent functionally coherent interactions: across all tissues, genes that shared a Gene Ontology function were linked to each other 94% more often than were genes that did not share a function (p < 10−6, t-test). (Gene Ontology annotations were downloaded January 2012; for enrichment analysis, we only considered functional categories with 30–300 annotations.)
Tissue-specific transcription factors (tsTFs) are important in defining tissue-specific phenotypes and mutations affecting tsTFs are enriched in loci associated with disease [2, 27]. We used our networks to analyze the role tsTFs play in tissue specificity using a collection of 203 known tsTFs (S2 Table) and 88 general TFs (gTFs) defined in [8]. We provide a schematic illustration of important conclusions of our analysis in Fig 4 and a tabular summary in Table 1.
Well-connected genes (also known as “hubs”) are especially likely to be essential genes [28]. To quantify a measure of “hubness”, we computed the betweenness centrality [29] in our networks for each gene. Both general and tissue-specific TFs had higher average hubness scores than the average gene (p < .001, p = .023, respectively), highlighting the importance of TFs in our networks.
tsTFs were higher expressed in tissues they were specific to (p < .001, bootstrap; S1 Fig), and tsTFs that showed the largest expression increases in tissues they were specific to were especially likely to be essential genes as defined in [30] (16 of the top 20 tsTFs as compared to 115/203 tsTFs overall, p = .005, Fisher’s exact test; this enrichment was not sensitive to the choice of 20 as the cutoff). tsTFs which showed tissue-specific increases in expression tended to also show increases in hubness (Spearman p = 3 ⋅ 10−4) (S2 Fig).
To investigate how tsTFs interacted with genes with tissue-specific functions, we defined thirteen sets of tissue-specific function genes (tsFXNGs) using Gene Ontology annotations of gene function (S3 Table). Importantly, in our networks, tsTFs showed clear signs of preferentially connecting to and upregulating genes with tissue-specific functions. Across all tissues, tsTFs were 58% more likely to be linked to genes with tissue-specific functions than they were to be linked to other genes (p < 10−6, binomial test). Genes with tissue-specific functions that were connected to tsTFs were higher expressed on average than either a) genes with tissue-specific functions that were not connected to tsTFs or b) genes with non tissue-specific functions that were connected to tsTFs (p < 10−6, t-test). (For a list of the tsTFs linked to the largest numbers of tissue-specific genes, see S4 and S5 Tables). This underscores the important role that tfTFs play in upregulating genes with tissue-specific functions. Perhaps as a consequence of this upregulation, tsFXNGs were higher expressed in the tissues they were specific to than in the tissues they were not specific to (p < .001, bootstrap). (We note that because our analysis is correlative and our networks are undirected, further analysis is needed to conclusively establish directed regulatory relationships.)
Strikingly, in contrast to tsTFs, tsFXNGs were less hubby than the average gene. This was especially surprising given that, across all tissues, higher-expressed genes tended to be more hubby (p < .001, linear regression). However, our finding is consistent with prior research showing that tissue-specific proteins have fewer interactions than widely expressed proteins [31]. One possible explanation is that tsFXNGs lie at the periphery of our networks because they have specialized functions, acting as final nodes in pathways.
To gain further insights into genes that were important to tissue specificity, at each internal node in our tissue hierarchy (representing a point where one group of tissues split into two) we examined genes that differed in hubness most dramatically between the two tissue groups.
We first sorted all genes by the difference in their hubness in brain and non-brain tissues. The highest three scoring genes have all been previously shown to play important roles in the brain: ACTL6A, a chromatin remodeling factor which is required for the development of neural progenitors [32, 33]; VRK2, a gene implicated in schizophrenia [34]; and the Huntington’s gene, HTT. Notably, three of the four genes HTT was most often linked to in brain tissues are themselves associated with neurological disorders: RNF123 to major depression [35], MTHFR to neural tube defects [36] and dementia [37]; MECP2 to Rett syndrome [38]. HTT has been found to interact directly with MECP2 [39].
Several other tissue-specific hubs proved interesting (S6 Table). For example, the genes which increased most in hubness in the two skin tissues were APOE, which has been linked with skin lesions known as xanthomas [40] (although it is more famous because of its link with Alzheimer’s) and CERS3 [41], which when mutated causes congenital ichthyosis, a skin disease. Similarly, in the testis, the top-two ranked tissue-specific hubs were DDX3Y and KDM5D, both Y-chromosome linked genes which function in spermatogenesis [42–44].
To identify tissue-specific and tissue-shared gene modules, we used the affinity propagation algorithm [45] to group genes into modules for each of the tissue networks. The average number of genes per module was 18, with the largest module containing 56 genes; there were 548 modules per tissue on average. 1,789 modules were enriched for Gene Ontology functions (Fisher’s exact test with Bonferroni correction p < .05); all enriched modules can be viewed online [19].
Functionally enriched modules upregulated in a given tissue were often instrumental to tissue-specific function (S7 Table). In the blood, for example, the most upregulated enriched module (henceforth, the “top module”) was enriched for T cell receptor complex expression (Fig 5); in the skin, for epidermis development; in the testis, for chromosome segregation; in the muscle and heart for muscle-related functions; and in various brain tissues for glutamate receptor activity, chloride channel activity, and regulation of axonogenesis. Given the plausibility of these functions, these modules represent useful candidates for future investigation.
Curiously, genes that were members of enriched clusters were less hubby than genes that were not in every tissue (p < .001, t-test). This discrepancy was so pronounced that we originally noticed it by visual examination of the networks in our web tool. One explanation would be that these enriched modules, like tsFXNGs, lie at the peripheries of networks because they act as the final steps in functional pathways.
Top modules also revealed more complex relationships between tissues. For example, immune-related modules were found not only in the blood, but also in lung and digestive tissues. (We note that there is some possibility of sample contamination, with the collected lung tissue including some blood cells. On the other hand, previous research [5] has found that the lung has similar gene expression patterns to immune tissues like the spleen and thymus, perhaps indicating the importance of immune function in the lung.) The top module in suprapubic skin, enriched for mitosis, was also upregulated in other tissues where cells divide frequently, including the testis, the stomach, the esophagus, and the colon.
Our analysis also revealed upregulation of tissue-specific modules in “similar” tissues: the top module in one tissue was often upregulated in tissues nearby in the hierarchy. For all brain tissues, top modules were dramatically upregulated in all other brain tissues as well, but not in non-brain tissues (Fig 6). The top module in the heart atrium, related to “structural constituent of muscle” was unsurprisingly upregulated in the muscle and heart ventricle as well.
We also identified a number of modules that were conserved in most tissues, representing ubiquitous functions shared by all cells. For each module in each tissue, we measured the degree to which the module was conserved by calculating the average fraction of links that were present among its genes in other tissues: f = 1 K ∑ j = 1 K n k n, where K was the total number of tissues, nk was the number of links between genes in the module in the kth tissue, and n was the number of links had the module been fully connected. When we sorted modules by f (filtering out modules with fewer than 10 genes, which tended to have high interlink fractions) we found that the top 50 modules were much more likely than the average module to be significantly enriched for a Gene Ontology function (78% vs 11%), and were dominated by functions related to chromosome segregation or the cell cycle, capacities essential for almost every tissue. When we sorted functions by the degree to which their enriched modules were conserved, we found that 8 of the 10 most conserved functions were general to almost every tissue, relating to cell division or cell signaling: “phosphatidylinositol-mediated signaling”, “mitotic cell cycle spindle assembly checkpoint”, “chromosome segregation”, “cell cycle”, “transport”, “cytokinesis”, “M phase of mitotic cell cycle”, and “chromosome, centromeric region”.
We present an algorithm that infers genetic networks in a collection of tissues, using a hierarchy to share data between tissues with many samples and tissues with few, and show that this sharing increases the accuracy with which we infer the networks. We use an objective function that can be optimized over all tissues in parallel, allowing our algorithm to scale to the GTEx dataset, and propose several further innovations that increase scalability. Our algorithm has broad applicability to any dataset of hierarchically related samples: species in a phylogenetic tree or cell lineages in a tumor, for example.
We then conduct a detailed analysis of the genetic networks in 35 human tissues, searching for principles underlying both the unity and diversity of tissue function. We find that unity arises from modules that persist across tissues, which are not only disproportionately likely to be enriched for Gene Ontology functions, but for functions like mitosis that are shared across virtually every tissue. We show that previously discovered general transcription factors, which act across many tissues, tend to be hubs in our networks.
At the same time, we find strong evidence of functional specialization among tissues (Fig 4). tsTFs, which tend to be hubs in our networks, play instrumental roles: they preferentially connect to genes with tissue-specific functions, and these genes show higher expression levels. Strikingly, genes with tissue-specific functions lie at the peripheries of our networks, as do genes within enriched clusters; one explanation for this is that these genes act as the final steps in pathways instrumental to tissue-specific function. Finally, modules enriched for Gene Ontology functions that are upregulated within a tissue are often instrumental to tissue-specific function, and provide intriguing candidates for biological investigation. As the availability of biological data increases, statistical network analysis will continue to reveal both important general principles by which networks accomplish their functions, and specific hypotheses worth investigating.
Genome-wide gene expression data for 1,606 samples across 43 unique tissues was collected by the GTEx consortium using RNA-sequencing; we used version phs000424.v3.p1 of the data. We confined our analysis to tissues with expression data for at least ten samples, resulting in a total of 1,559 samples and 35 tissues (S1 Table).
GO annotations were downloaded from www.geneontology.org on January 28th, 2012. All IEA annotations were excluded, and then all remaining GO categories with 20–300 annotated genes (any annotation type except IEA) were included in the analysis. No filter was placed on the ontology.
For each read count ni in each sample, we computed the normalized read count ri = log2(2 + C ⋅ ni/n) where n was the total number of reads in the sample and C was the FPKM normalization constant, 5 ⋅ 107. Because GMRFs are designed for Gaussian data, we projected all samples for each transcript for each tissue onto a Gaussian with variance 1.
The GTEx dataset contained expression levels for 52,576 different transcripts, which would have produced a prohibitively large covariance matrix. We filtered down the set of transcripts to a more computationally tractable size. Since transcripts would have to show variation in expression levels to have meaningful patterns in correlation, we first filtered out all probes that were zero or constant across any tissue by requiring that genes show non-zero expression in at least 1/5 of samples in a tissue. We then selected a set of transcripts as follows: we repeatedly looped over all tissues, and for each tissue selected the transcript which corresponded to a gene which showed the highest relative expression in that tissue, was annotated in Gene Ontology, and was not already included in the genes selected. (We defined relative expression in a tissue to be the difference between the gene’s mean expression in that tissue and the gene’s mean expression across all tissues divided by the variance of the gene’s expression). We continued this process until we had obtained 9,998 genes. (This number was produced by choosing a threshold of 10,000 genes, which represented a compromise between representing the entire dataset and achieving computational tractability, and removing two genes which did not have unique names.) This process yielded a set of genes with diverse tissue-specific functions (since each tissue contributed many genes which showed high relative expression in that tissue).
We confirmed that our algorithm also produced improvements over the baseline algorithms in two smaller gene sets containing roughly 2,000 genes: one selected using the method described above, and one selected using the genes that showed the largest variance across tissues.
Given a hierarchy of K tissues, our algorithm learned a precision matrix for each node in the hierarchy, including the K leaf nodes S(1), S(2), …, S(K) (which corresponded directly to tissues) and the K − 1 internal nodes S(K+1), …, S(2K−1). Denote by S p k the parent of node k. Then the optimization objective was
max S ( k ) , k = 1 , … , 2 K - 1 ∑ k = 1 K ( n ( k ) 2 ( log det S ( k ) - t r ( S ( k ) Σ ( k ) ) ) - λ s ( k ) ‖ S ( k ) ‖ 1 ) - λ p ∑ k = 1 2 K - 2 ‖ S ( k ) - S p ( k ) ‖ 2 2
S ( k ) ⪰ 0 , k = 1 , 2 , … , K
where λ s ( k ) were the k L1 sparsity penalties (chosen for each dataset as described below) and λp was the L2 penalty that encouraged S(k) to be similar to its parent S p ( k ) (constant for all tissues). In other words, for the leaf nodes, our optimization objective included the Gaussian log likelihood term, a sparsity penalty on the off-diagonal elements, and an L2 parent similarity term; for the internal nodes, there was only an L2 similarity term. While this optimization objective was convex, the inverse precision matrices had tens of millions of entries and optimizing all 2K − 1 matrices simultaneously would have been very slow. Instead, we used an iterative algorithm: given a hierarchy, the full optimization procedure was as follows:
For each dataset k = 1, …, K, learn an initial S(k) by maximizing
n ( k ) 2 ( log det S ( k ) − t r ( S ( k ) Σ ( k ) ) ) − λ s ( k ) ‖ S ( k ) ‖ 1. In other words, initialize by solving the graphical lasso problem for each dataset independently.
Until convergence:
Optimize the internal matrices, S(k), k = K + 1, …, 2K − 1, holding the leaf matrices fixed; because all relevant terms of the objective were quadratic, this was analytic and essentially instantaneous. (We note that this would not be true if an L1 penalty were used rather than an L2 penalty.) Optimize the leaf matrices, S(k), k = 1, …, K, holding the internal matrices fixed; each leaf matrix was independent of the others given its parent, so this was done in parallel. Optimization was performed using the L1General [46] and glasso [47] packages.
To ensure that the size of the entries in S were comparable across tissues and between internal and external nodes, prior to each iteration we normalized each S such that all S had the same mean absolute value of diagonal elements and the same mean absolute value of nonzero off-diagonal elements.
To expedite this potentially lengthy process of choosing a sparsity parameter λ s ( k ) for each of 35 tissues, we used a heuristic rather than using the traditional cross-validation for every single tissue. We confirmed that our heuristic produced similar results to cross validation. [48] found the BIC penalty effective in selecting the sparsity parameter for graphical lasso: log(n)‖S(k)‖0, where ‖S(k)‖0 is the number of non-zero off-diagonal entries of S(k). This suggests setting λ s ( k ) to a value that makes the L1 penalty equal to the BIC penalty: λ s ( k ) = l o g ( n ( k ) ) / s ‾, where s ‾ is the mean absolute value of the nonzero off-diagonal entries in the optimized precision matrix. Substantiating this, we found that log(n(k)) was tightly correlated in both simulated and actual data with the optimal L1 penalty, and also outperformed the n ( k ) suggested by [49]. This appears to beg the question of how to estimate s ‾ without doing the actual optimization; however, we found that s ‾ was tightly correlated in both simulations and in the GTEx datasets with Σ ‾, the mean size of the entries in the empirical covariance matrix. Similarly, l o g ( n ( k ) ) / Σ ‾ ( k ) was tightly correlated in both simulations and actual data with λ s ( k ). Thus, we can select λ s ( k ) for all K datasets by using parameter search to select λ s ( 1 ) , λ s ( 2 ) , … λ s ( i ), where i is much smaller than K; we then do a regression of the optimized λ s ( k )s on l o g ( n ( k ) ) / Σ ‾ ( k ), and use that fit to compute the remaining λ s ( k ). We confirm that this method works on both simulated precision matrices and the GTEx dataset. For the GTEx dataset, using i = 5 yields λ(k) within 17% of the values selected by cross-validation on average; i = 3 yields values within 26%, acceptable discrepancies given the coarseness of parameter search.
Most algorithms for solving the graphical lasso problem with p genes are O(p3), making optimization intractable for 9,998 genes. If the optimal solution were block diagonal, with block sizes p1, …, pk, optimization could be performed in O ( ∑ i = 1 k p i 3 ), as noted in [50] and [51]. Unfortunately, we found that the criterion these papers provide for determining whether the problem decomposes requires too large a sparsity parameter to be practically useful. Instead, we used an approximate eigenvector-based diagonalization similar to that described in [52]: for each tissue, we computed a matrix C(k), with C i j ( k ) = m a x ( 0 , Σ i j ( k ) − λ s ( k ) ) 2. We then computed the weighted sum of the matrices: S = ∑ k = 1 K n ( k ) C ( k ), and partitioned S into approximate connected components using the principal eigenvector as described in [52]. (To ensure that all components had tractable size, we set a maximum component size of 500 genes and recursively partitioned components until they fell below this threshold.) We confirmed that this approximate solution had a higher test log likelihood than that obtained by choosing a sparsity parameter sufficiently large to make an exact solution tractable.
Because the L1 optimization algorithm and our initializations are stochastic, the final optimized networks may vary slightly from run to run. However, we verified that our results were not overly sensitive to repeated runs of the algorithm, to parameter settings, or to which samples we used by examining two modified networks: one optimized using a subset of 4/5 of the samples and one optimized using λp = 2 as opposed to λp = 4. We found that both modified networks were highly enriched for links in our actual network; links in the actual network were more than 100 times as likely as random links to be found in the modified networks. In modified networks, we tested a number of the network properties reported above. First, we verified that we still saw statistically significant correlations with the external datasets COEXPRESdb and TS-CoExp. Second, we verified that tissue-specific genes, and genes with shared functions, still showed statistically significant tendencies to be linked to each other. Finally, we verified that tsTFs still showed a statistically significant tendency to be linked to genes with tissue-specific functions. The robustness of all these conclusions made us confident that the conclusions reported above are unlikely to be due to which samples in the dataset are used, the values of the parameters, or variations in the initialization of the algorithm, although specific links in the networks may change.
We also analyzed the proportion of links that were conserved across different conditions. We compared networks calculated using our chosen value of λp = 4 to those learned with different values of λp (S9 Table); 89% of links were conserved between networks learned with λp = 4, λp = 2, and 98% between networks learned with λp = 4, λp = 8. A somewhat lower proportion (75%) of links were conserved between λp = 4, λp = 0, implying that the use of a similarity penalty may be more important than the exact size of the similarity penalty. We also compared the networks learned on all samples to the networks learned using a subset of 4/5 of the samples; 38% of the links were conserved in the average tissue. Given the sparsity of the networks, all these proportions are more than 100 times what random chance would predict. However, because specific links can change depending on which samples are used, the broad conclusions of our analysis are more robust than any particular link we predict.
|
10.1371/journal.pntd.0000625 | Molecular Evidence for a Functional Ecdysone Signaling System in Brugia malayi | Filarial nematodes, including Brugia malayi, the causative agent of lymphatic filariasis, undergo molting in both arthropod and mammalian hosts to complete their life cycles. An understanding of how these parasites cross developmental checkpoints may reveal potential targets for intervention. Pharmacological evidence suggests that ecdysteroids play a role in parasitic nematode molting and fertility although their specific function remains unknown. In insects, ecdysone triggers molting through the activation of the ecdysone receptor: a heterodimer of EcR (ecdysone receptor) and USP (Ultraspiracle).
We report the cloning and characterization of a B. malayi EcR homologue (Bma-EcR). Bma-EcR dimerizes with insect and nematode USP/RXRs and binds to DNA encoding a canonical ecdysone response element (EcRE). In support of the existence of an active ecdysone receptor in Brugia we also cloned a Brugia rxr (retinoid X receptor) homolog (Bma-RXR) and demonstrate that Bma-EcR and Bma-RXR interact to form an active heterodimer using a mammalian two-hybrid activation assay. The Bma-EcR ligand-binding domain (LBD) exhibits ligand-dependent transactivation via a GAL4 fusion protein combined with a chimeric RXR in mammalian cells treated with Ponasterone-A or a synthetic ecdysone agonist. Furthermore, we demonstrate specific up-regulation of reporter gene activity in transgenic B. malayi embryos transfected with a luciferase construct controlled by an EcRE engineered in a B. malayi promoter, in the presence of 20-hydroxy-ecdysone.
Our study identifies and characterizes the two components (Bma-EcR and Bma-RXR) necessary for constituting a functional ecdysteroid receptor in B. malayi. Importantly, the ligand binding domain of BmaEcR is shown to be capable of responding to ecdysteroid ligands, and conversely, ecdysteroids can activate transcription of genes downstream of an EcRE in live B. malayi embryos. These results together confirm that an ecdysone signaling system operates in B. malayi and strongly suggest that Bma-EcR plays a central role in it. Furthermore, our study proposes that existing compounds targeting the insect ecdysone signaling pathway should be considered as potential pharmacological agents against filarial parasites.
| Filarial parasites such as Brugia malayi and Onchocerca volvulus are the causative agents of the tropical diseases lymphatic filariasis and onchocerciasis, which infect 150 million people, mainly in Africa and Southeast Asia. Filarial nematodes have a complex life cycle that involves transmission and development within both mammalian and insect hosts. The successful completion of the life cycle includes four molts, two of which are triggered upon transmission from one host to the other, human and mosquito, respectively. Elucidation of the molecular mechanisms involved in the molting processes in filarial nematodes may yield a new set of targets for drug intervention. In insects and other arthropods molting transitions are regulated by the steroid hormone ecdysone that interacts with a specialized hormone receptor composed of two different proteins belonging to the family of nuclear receptors. We have cloned from B. malayi two members of the nuclear receptor family that show many sequence and biochemical properties consistent with the ecdysone receptor of insects. This finding represents the first report of a functional ecdysone receptor homolog in nematodes. We have also established a transgenic hormone induction assay in B. malayi that can be used to discover ecdysone responsive genes and potentially lead to screening assays for active compounds for pharmaceutical development.
| Human filarial parasitic nematodes are responsible for two chronic severely debilitating tropical diseases: lymphatic filariasis and onchocerciasis. The global efforts in the treatment and control of the spread of infection for both parasites so far have resulted in limited success. Also, the widespread use of the few available specific drugs for fighting these diseases raises the possibility of the development of drug resistance [1]. With 140 million cases of infection worldwide, and over a billion people at risk of contracting these debilitating diseases [2], the development of a wide range of therapeutic interventions and treatment options is urgent.
Filarial parasites spend portions of their life cycle in obligate mammalian and insect hosts. The completion of a successful life cycle requires the passage of the developing nematode through four molts, two in the mammalian host and two in the arthropod host. The transmission of the parasite from one host to the other initiates a rapid molt, indicating that the developmental cues that trigger molting are closely tied to the integration of the parasitic larva into a new host environment. Inhibition of molting would result in the arrest of the life cycle in either the mammalian or insect host and the prevention of both pathology and/or the infective cycle. Thus, the study of the molting process in filarial nematodes could point to specific targets for drug development.
Molting in ecdysozoans [3] has been best characterized in insects. 20-hydroxyecdysone (20E) acts as the temporal signal to initiate molting, regulates embryogenesis, and coordinates tissue-specific morphogenetic changes in insects [4]–[6]. Ecdysone signaling is regulated by the activity of a heterodimeric receptor composed of two nuclear receptor proteins EcR and USP, although the hormone binding function resides only within EcR [7]–[10]. After ligand binding, EcR/USP activates a cascade of gene expression whose end result is the execution of molting [11].
Three alternatively spliced mRNA isoforms of EcR have been identified in Drosophila [12]. Mutations in these different EcR mRNA isoforms result in a range of phenotypes that includes lethality at the embryonic, larval and pupal stages, disruption of salivary gland degeneration [13], aberrant neuronal remodeling during metamorphosis [14], and changes in female fecundity and vitellogenesis [15].
EcR and USP, as well as a number of the proteins involved in the ecdysone-signaling cascade, are members of the nuclear receptor (NR) superfamily [10],[16]–[18]. NRs are characterized by significant amino acid sequence similarities in two key functional domains: the DNA binding domain (DBD), which directs the sequence-specific DNA-binding of the receptor, and the ligand binding domain (LBD), which mediates dimerization, ligand binding and transcriptional activation [19]–[20]. Some nuclear receptors have been shown to interact with a number of small molecule ligands such as metabolites and hormones, and these interactions are important for regulation of their activity. Other NRs are considered orphan receptors and are either not ligand-regulated or their cognate ligands have yet to be identified [19]. Homologs of the insect NRs that function downstream of EcR and USP have also been identified in filarial parasites (21, 22; Egaña, Gissendanner and Maina, unpublished results) as well as in the free-living nematode C. elegans [23]–[26]. Surprisingly, however, homologs of EcR or RXR/USP are apparently absent in the exceptionally large C. elegans NR family [23].
In filarial nematodes the molecular triggers of molting remain largely unknown. As in insects, a possible candidate for a signal that controls molting in B. malayi, the causative agent of lymphatic filariasis, is the steroid hormone 20E. Both free and conjugated ecdysteroids have been identified in the larvae of several parasitic nematodes including Dirofilaria immitis and Onchocerca volvulus [27]–[29]. In addition, ecdysteroids have been shown to exert biological effects on several nematodes. For example, in Nematospiroides dubius [30] and Ascaris suum [31] molting can be stimulated in vitro by low concentrations of ecdysteroids. Also, molting of third stage larvae of D. immitis can be stimulated with 20E and RH5849, an ecdysone agonist [32],[33]. The arrest at the pachytene stage of meiosis is abrogated when D. immitis ovaries are cultured in vitro with ecdysone and B. pahangi adult females can be stimulated to release microfilaria when cultured in vitro with ecdysone [34].
There appears to be a physiological connection between the filarial parasite and its arthropod host that may involve ecdysteroid signaling. Uptake of microfilaria (L1) by a feeding female mosquito at the time of a bloodmeal coincides with an increase in the production of mosquito ecdysteroids that results in the initiation of mosquito oocyte maturation [35]. Concurrent with this increase in ecdysteroid concentration in the mosquito host, larvae initiate a molt transition from L1 to L2 and later from L2 to L3, the infectious stage of the parasite. These observations suggest a potential role for ecdysone in the regulation of molting and other developmental processes in filarial nematodes.
We previously identified an rxr homolog in the dog filarial parasitic nematode D. immitis and demonstrated its ability to dimerize with an insect EcR and function in Schneider S2 cells [36]. We extend this work here with the identification and characterization of EcR and rxr homologs from B. malayi. Bma-EcR and Bma-RXR share some of the biochemical properties of insect EcR and RXR and show differences that appear to be nematode specific.
Brugia malayi adult males, females or L1 larvae (TRS Labs, Athens, GA) were frozen in liquid nitrogen and ground with a pestle and mortar. Total RNA was purified from the pulverized tissue using RNAwiz (Ambion). RNA was quantified with a spectrophotometer and its quality assessed by gel electrophoresis. One µg of total RNA per isolation was reverse-transcribed using the ProtoScript first strand cDNA synthesis kit (New England Biolabs) following the manufacturer's protocol.
The genomic library from B. malayi in pBeloBAC vector gridded on Nylon filters (Filarial Genome Network, FGN, (http://www.nematodes.org/fgn/index.shtml) was screened using a cDNA fragment from a D. immitis EcR homolog (Di-EcR) (C. Shea, J. Richer and C. V. Maina, unpublished results) as a probe. Three positive BACs were identified and the individual corresponding bacterial clones were cultured. The inserts were confirmed to contain identical or overlapping sequences by restriction digestion analysis. One 11kb XbaI fragment identified by southern blot hybridization with the Di-EcR probe was subcloned into Litmus 28i and the insert was sequenced using GPS®-1 Genome Priming System (NEB) as directed by the manufacturer. PCR primers were designed to amplify Bma-EcR using sequence from the identified exons [37]. The primers used to amplify the full ORF were: 5′ – GGC GCT AGC ATG ACT ACA GCA ACA GTA ACA TAT CAT GAG TT – 3′ (Nco-MMT-5); 5′ – GGC CTC GAG CGA TTC TAT GGA TAG CCG GTT GAG GTT – 3′ (Xho-GYP-3). To determine the expression pattern and identify alternate isoforms of Bma-EcR, adult female, male, L1, L2, L3 cDNA libraries (FGN) were screened using the following primers: 5′ – GGG TAA TTC CTA CCA ACA GCT - 3′(GNS); 5′ – CAA GGG TCC AAT GAA TTC ACG AT – 3′ (GPL) corresponding to a fragment of the LBD from amino acids GNSYQQ to REFIGPL'. Additional sequence to extend the Bma-EcR isoforms identified was obtained by PCR, combining the latter two primers with the T3 and T7 promoter primers as their sequence is present in the library vector.
An O. volvulus L3 cDNA library (FGN) was screened by PCR using the following primers: 5′ – GAT CTT ATC TAT CTA TGC CGA GAA ,– 3′; 5′ – TAC TTT GAC ATT TGC GGT AAC GAC – 3′ corresponding to the amino acid sequence DLIYLCRE and RYRKCQSM of the conserved DNA binding domain of DiRXR-1 respectively. Additional Ov-rxr sequence was obtained by PCR using the same primers in combination with the T7 and T3 promoter primers. Candidate clones were identified by hybridization with a fragment of DiRXR-1 sequence. An amplified fragment from this library contained sequence corresponding to the Ov-rxr A/B and C domains.
Using BLAST, the Di-rxr-1 sequence [36] was used to screen the B. malayi genome sequence available from The Institute for Genomic Research (TIGR) parasites database (http://blast.jcvi.org/er-blast/index.cgi?project=bma1). This analysis resulted in the identification of several exons encoding an 1189 bp fragment of open reading frame (ORF) that corresponded to a putative homolog of rxr in Brugia (Bma-RXR). Based on the genomic sequence we designed PCR primers and used them to amplify the expected Bma-RXR mRNA using a nested PCR approach. One µl of a reverse transcription reaction from female total RNA was used as the template for the first round of PCR carried out with primers 5′ – CGA TCT ATG CCC ATC AGA TTG-3′ (LCP) and 5′ – CAC AAT GCA AGC TAA GAG ATC G – 3′(RSL) at 46°C annealing temperature. Six percent of the first round PCR reaction was used as a template in a second round of PCR with primers 5′ – CGA TTT AAC TCC AAA TGG AAG TCG – 3′ (DLT) and 5 ′– AGC AAA GCG TTG AGT TTG TGT TGG – 3′ (PTQ) at 47°C annealing temperature. Using the sequence obtained, primers were designed to extend the 5′ of the coding sequence using a semi-nested PCR approach in combination with the 5′ splice leader SL1 primer. As above, two rounds of PCR were employed, in the first round using the SL1 primer (5′ – GGT TTA ATT ACC CAA GTT TGA G - 3′) and primer 5′ – GAT GCT CGA TCA CCG CAT ATT GCA CAA ATG - 3′ (CAI) at 68°C annealing temperature and for the second round the SL1 primer and primer 5′- TGG CAT ACA GTG TCA TAT TTG GTG TTG TGC - 3′ (STT) at 66°C annealing temperature. The 3′ coding sequence was obtained by 3′ RACE using the First Choice RLM-RACE kit (Ambion) with Bma-RXR primers 5′ – GGC TCT AAT GCT ACC ATC ATT TAA TGA A - 3′ (ALM), and 5′ – GAA GAT CAA GCT CGA TTA ATA AGA TTT GGA - 3′ (EDQ) following the manufacturer's protocol. For each amplified fragment several clones were sequenced. The positions of the primers used are indicated by short arrows over the corresponding amino acid sequence in the alignment Figures.
Ten µg of total RNA from adult male, adult female and microfilaria were used to carry out northern blot analyses using the NorthernMax-Gly kit (Ambion). The Bma-EcR and Bma-RXR probes were 1kb DNA fragments from the respective coding regions labeled using random priming with the NEBlot kit (NEB) and 32P-dATP (NEN-Dupont). The sizes of the hybridizing RNA species were estimated using an RNA ladder that was run adjacent to the samples as a reference.
Predicted amino acid sequences of cloned cDNAs were aligned with all nuclear receptor sequences from Swissprot and GenBank, using Muscle [38] with default options. A complete phylogeny of all the nuclear receptor super-family was built with PhyML [39]. Subsequently, phylogenies of the relevant sub-families were constructed with 1000 bootstrap replicates. Well aligned sites were selected with GBLOCKS [40], with relaxed options to allow a few gaps per column of the alignment. In each case PhyML was run with rate heterogeneity with 4 classes, parameter alpha estimated from the data, BIONJ starting tree. Support for nodes was estimated by Approximate Likelihood-Ratio Test (aLRT) [41].
A cDNA fragment of Bma-EcR encoding aa 152–465 (upstream of the C-domain to the end of the predicted ORF) was amplified by PCR using primers 5′ – AGC TTC CAT GGC AGC TGA AGA AGG TCA ATC TAA TGG CGA CAG TGA GT – 3′ (536 to 557 of EF362469) and Xho-GYP-3 (See cloning of Bma-EcR). The fragment was cloned in frame with GST in the vector pGEX-KG [42]. The fusion protein was produced in E. coli BL21 by induction at 30°C with 0.1 mM IPTG and purified on Glutathione Sepharose beads (Pharmacia) as directed by the manufacturer. Recombinant Di-rxr-1 and Aausp (gift from A. Raikhel, University of California Riverside) in pcDNA-3 (Invitrogen) were transcribed and translated in vitro in rabbit reticulocyte lysates using the TNT T7 coupled transcription-translation system (Promega) in the presence of 35S-Methionine (Amersham Biosciences) as recommended by the manufacturer. Glutathione resin beads loaded with 1 µg of GST:Bma-EcR fusion protein were incubated for 1 h at 4°C with 5 µl of rabbit reticulocyte lysate containing labeled proteins (Di-RXR-1 or AaUSP) in a total volume of 10 µl of binding buffer (20 mM Tris, 1 mM EDTA, 1 mM DTT, 10% glycerol, 150 mM sodium chloride, 0.5 mg/ml of BSA, complete protease inhibitor cocktail (Sigma). The beads were washed twice with binding buffer and three times with buffer without BSA, then incubated with 10 mM reduced glutathione to elute the proteins, and centrifuged. Supernatants were mixed with loading buffer and analyzed by SDS-PAGE. Signals were detected by autoradiography of the dried gels.
The ecdysone response element (PAL-1) described by Hu et al. [43] was produced by annealing two synthetic oligonucleotides: 5′ – TTG GAC AAG GTC AGT GAC CTC CTT GTT CT – 3′ and its complement (with two overhanging Ts at each 3′ end). PAL-1 was labeled with 32P-dATP (NEN-Dupont) using Klenow polymerase (New England Biolabs) and purified by spin column G50 chromatography (Amersham Biosciences). A cDNA fragment containing the complete coding region of Bma-EcRA was cloned in pcDNA-3 (Invitrogen) using the NheI and XhoI restriction sites. Two additional constructs containing Bma-EcRB and C respectively were also cloned using the same strategy. The three Bma-EcR isoforms were transcribed and translated in vitro in rabbit reticulocyte lysates using the TNT T7 coupled transcription-translation system (Promega) following the manufacturer's protocol. The translation yield of each construct was assessed by labeling a portion of the reaction with 35S-Methionine and analyzing the products after gel electrophoresis and autoradiography. Binding reactions were performed at room temperature in 10mM Tris-HCl pH 7.5, 50 mM NaCl, 10 mM MgCl2, 0.5 mM DTT, 0.025 mM EDTA, 4% glycerol, 0.2 µg/µL poly dI-poly-dC, 0.13 µg/µL BSA, 0.05% NP40, with 13 fmol/µl labeled PAL-1 and 3.5 µL TNT reaction mixture containing the corresponding proteins (0.5 µLAaUSP and 1.5 or 2.5 µL Bma-EcR), in 15 µL total volume for 20 min before loading in a 6% native TBE gel (Invitrogen). Signals were detected by autoradiography of the dried gels.
To construct GAL4:Bma-EcR and VP16:Bma-RXR, the DEF domains of Bma-EcR and Bma-RXR were PCR amplified and cloned into pM and pVP16 vectors (EcR residues 259–565; RXR residues 191 to 464) respectively (Clontech). The construct VP16:Lm-HsRXREF (Chimera 9) has been previously described [44]. pFRLUC, encoding firefly luciferase under the control of the GAL4 response element (Stratagene Cloning Systems) was used as a reporter.
Fifty thousand NIH 3T3 cells per well in 12-well plates were transfected with 0.25 µg of receptor(s) and 1.0 µg of reporter constructs using 4 µl of SuperFect (Qiagen). After transfection, the cells were grown in medium containing ligands for 24–48 hours. A second reporter, Renilla luciferase (0.1 µg), expressed under a thymidine kinase constitutive promoter was cotransfected into cells and was used for normalization. The cells were harvested, lysed and the reporter activity was measured in an aliquot of lysate. Luciferase activity was measured using Dual-luciferaseTM reporter assay system from Promega Corporation (Madison, WI, USA). The results are reported as averages of normalized luciferase activity and the error bars correspond to the standard deviation from multiple assays. The ligands used were: RG-102240, a synthetic stable diacylhydrazine ecdysone agonist [N-(1,1-dimethylethyl)-N′-(2-ethyl-3-methoxybenzoyl)-3,5-dimethylbenzohydrazide] also known as GS-E or RSL1, (RheoGene, New England Biolabs) and Ponasterone-A (Invitrogen). The ligands were applied in DMSO at the indicated final concentrations and the final concentration of DMSO was maintained at <0.1%.
In order to construct an ecdysteroid response reporter for Brugia malayi the repeat domain of the B. malayi 12 kDa small ribosomal subunit gene promoter [45] (construct BmRPS12 (−641 to −1)/luc) was replaced (in both orientations) with the PAL-1 EcRE shown to be recognized by Bma-EcR in vitro. Previous studies have shown that the repeat acts as a transcriptional enhancer. Outward facing primers flanking the repeat domain containing synthetic SpeI sites at their 5′ ends were used in an inverse PCR reaction employing BmRPS12 (−641 to −1) as a template [46]. The resulting amplicons were purified using the QiaQuick PCR cleanup kit (Qiagen). The purified amplicons were digested with SpeI, gel purified, self-ligated and transformed into E. coli. The resulting construct was designated BmRPS12 -rep. A double stranded oligonucleotide consisting of five tandem repeats of the EcRE: ctag(GGACAAGGTCAGTGACCTCCTTGTTC) 5× with SpeI overhangs was then ligated into the SpeI site of BmRPS12 -rep. The insertions in the forward and reverse orientations were designated BmRPS12-EcRE and BmRPS12-EcRE-rev respectively.
Constructs were tested for promoter activity in transiently transfected B. malayi embryos essentially as previously described [47]. In brief, embryos were isolated from gravid female parasites and transfected with BmRPS12-EcRE (or BmRPS12-EcRE-rev) mixed with a constant amount of a transfection control, consisting of the BmHSP70 promoter fragment driving the expression of renilla luciferase (construct BmHSP70 (−659 to −1)/ren). Following a rest of five minutes, the transfected embryos were transferred to embryo culture media (RPMI tissue culture medium containing 25 mM HEPES, 20% fetal calf serum, 20 mM glucose, 24 mM sodium bicarbonate, 2.5 mg ml-1 amphotericin B, 10 units ml-1 penicillin, 10 units ml-1 streptomycin and 40 mg ml/L gentamycin), supplemented with 1 µM 20-OH ecdysone dissolved in 50% ethanol or solvent control. Transfected embryos were maintained in culture for 48 hours before being assayed for transgene activity. Firefly luciferase activity was normalized to renilla luciferase activity in each sample to control for variations in transfection efficiency. Firefly/renilla activity ratios for each sample were further normalized to the activity ratio from embryos transfected in parallel in each experiment with the parental construct BmRPS12 -rep. This permitted comparisons of data collected in experiments carried out on different days.
Each construct was tested in two independent experiments, with each experiment containing triplicate transfections of each construct to be analyzed. The statistical significance of differences noted between the activity in the control and experimental transfections was determined using Dunnett's test, as previously described [47].
The nucleotide sequences for Bma-EcR isoform A, Bma-EcR isoform C, Bma-RXR and Ovnhr-4 have been deposited in the GenBank database under GenBankAccession Numbers: EF362469, EF362470, EF362471, and EF362472.
A candidate EcR homolog was first identified from D. immitis using degenerate PCR primers based on insect EcRs (C. Shea, J. Richer and C.V. Maina, unpublished results). Using sequences from the D. immitis EcR homolog, genomic libraries from B. malayi available from the Filarial Genome Network (FGN) were screened. A strongly hybridizing BAC was identified and sequenced. This BAC contained a gene that encodes a protein with strong similarities to the EcR branch of nuclear receptors (see below). We designated this gene as Bma-EcR (to distinguish it from Bombyx mori EcR [37]). Using sequences corresponding to the predicted Bma-EcR exons, PCR primers were designed and used to screen larval and adult cDNA libraries. This library survey revealed Bma-EcR expression in L1, L3, and L4 larval stages, as well as in adult males and females (data not shown). In the microfilaria (L1) library, using primers from the putative ligand-binding domain (LBD) encoding region, two alternatively spliced mRNA isoforms of Bma-EcR (isoforms A and C) were identified (Fig. 1A). Bma-EcRA is the isoform containing the longest ORF (597 a.a.) with an intact LBD. Bma-EcRC contains exon 6 with a 29-nucleotide deletion that results in a reading frame shift that generates a premature stop codon and truncation of the LBD at helix 5 (Fig. 1). This is the result of an alternative splice site within exon 6. In addition to these confirmed isoforms, a splice site consensus sequence was identified within exon 5 at the end of the DNA-binding domain (DBD) that, if used, would result in the omission of ten amino acids from the C-terminal extension of the DBD (indicated by an arrow in Figure 1A). This type of spliced mRNA isoform has been identified in D. immitis (C. Shea, J. Richer and C. V. Maina, unpublished results). We were unable to clone such an isoform (EcRB) from B. malayi by RT-PCR. However, we cannot exclude the possibility that Bma-EcRB is expressed in specific tissues or developmental stages not represented in the libraries or RNA used.
Bma-EcR shows strongest similarity to the NR1H group of nuclear receptors typified by the insect EcRs and mammalian FXR and LXR receptors (Figs. 1B, 2). The strongest similarity is in the DBD which contains the canonical C4 zinc finger structure of nuclear receptors. This domain is 10 amino acids longer in Bma-EcR than in the homologous region of the other EcRs. However, as indicated above, exclusion of these 10 amino acids by alternative splicing of this site (isoform B) would result in a better alignment of Bma-EcR with the other EcRs (Fig. 1B). The LBD shows significant similarity in the regions encoding helices 3–10 [48]. An exception is found in the region of helix 11–12 (Fig. 1B). Helix 12 of insect EcRs contains the AF2 motif, responsible for ligand-dependent transcriptional activation. Although predictions of secondary structure of the Bma-EcR LBD protein sequence indicate helical folding of putative helices 3–10, no helical propensity is predicted in the region of helix 12 (data not shown). Immediately following the helix 12 region a glutamine-rich helical segment is present. Glutamine-rich sequences are often associated with transcription activation domains [49]. These differences make Bma-EcR an unusual member of the receptor family that perhaps uses a different mechanism for ligand-dependent activation.
Global phylogenetic analysis (see Supporting Information Fig. S1) places Bma-EcR with arthropod EcRs. The position of Bma-EcR is strongly supported (99% aLRT support) in a phylogenetic tree of the sub-family (Figures 2 and S2). The branch leading to Bma-EcR is long, indicating a relatively derived sequence, but not more derived than that of dipteran EcRs, for example. Separate analysis of the DBD and LBD produced similar tree topologies, especially concerning the position of Bma-EcR.
Northern blot analysis was used to establish the expression pattern of Bma-EcR in Brugia adult females, males, and L1 microfilaria. The fragment used as the probe encompassed the coding sequence common to both mRNA isoforms identified. A predominant species of approximately 3.75–4 Kb was present in all RNA samples tested (Fig. 3), which was consistent with our detection by RT-PCR of the Bma-EcR isoforms in libraries from those same stages, and implies the existence of longer 5′ and/or 3′ untranslated regions than are present in our cloned cDNA species. Shorter minor RNA species are detectable which may indicate the existence of additional isoforms (Fig. 3).
EcRs heterodimerize with Ultraspiracle (USP) proteins to form functional ecdysone receptors that bind to ecdysteroid ligands and ecdysone response elements (EcREs) [35]. In order to test whether Bma-EcR heterodimerizes with a canonical insect USP or its filarial homologue Di-RXR-1 [36], an in vitro binding assay was carried out. In vitro translated 35S-labeled Di-RXR-1 or Aedes aegypti USP (AaUSP) were incubated with GST or GST:Bma-EcR fusion proteins immobilized on glutathione-beads. Specific bands corresponding to the full length AaUSP and Di-RXR-1 were detected bound to GST:Bma-EcR (Fig. 4A). No binding to GST alone was detected with either protein bait. While the in vitro translation of AaUSP resulted in the production of a major protein species of the predicted full length AaUSP (Fig. 4A, lane 1), in vitro translation of Di-RXR-1 produced multiple protein species (Fig. 4 lane 4) including one corresponding to full-length Di-RXR-1 (ca 55 kD), which specifically bound to GST:Bma-EcR (Fig. 4A lane 6). These results indicate that Bma-EcR protein, like EcR, is capable of heterodimerization with USP protein in vitro.
Having established that Bma-EcR can dimerize with USP/RXRs, we investigated the DNA binding properties of the two isolated protein isoforms of Bma-EcR (Forms A and C) to a palindromic ecdysone response element (PAL-1 EcRE) based on the Drosophila hsp27 ecdysone response gene [43], using EMSA. In addition to the cloned Bma-EcRA and Bma-EcRC mRNA isoforms, a construct lacking the 10 amino acids downstream of the zinc finger domain was engineered (putative Bma-EcRB). The three Bma-EcR isoforms and AaUSP (as the heterodimerization partner) were produced in rabbit reticulocyte lysates and their relative amounts were estimated using 35S-met labeling and autoradiography (Fig S3). An equal amount of AaUSP-containing reticulocyte lysate was incubated with increasing amounts of each Bma-EcR isoform preparation and 32P-labeled EcRE prior to analysis by native polyacrylamide gel electrophoresis. AaUSP produces a specific band with the EcRE as has been shown before [50] (Fig. 4B, dot), which migrates faster than a nonspecific band produced by the reticulocyte lysate (Fig. 4B, asterisk). Both Bma-EcRA and -B produced an additional slower migrating band consistent with a heterodimer bound to the probe (Fig. 4B, lanes 2–5, arrow). In contrast, no additional band is detected with Bma-EcRC (Fig. 4B, lanes 6–7). This result is not unexpected given that Bma-EcRC, which contains a premature stop codon and encodes a protein with a truncated LBD, lacks essential structural features for heterodimerization. Neither Bma-EcRA nor Bma-EcRB bound substantially to the EcRE in the absence of AaUSP (Fig. 4B, lanes 8–9). This in vitro analysis of Bma-EcR heterodimerization with AaUSP and binding to an EcRE suggests that Bma-EcR has DNA-binding properties similar to those of ecdysone receptors.
The dimerization properties of Bma-EcR and the identification of rxr [36] and EcR homologs in the dog filarial parasite D. immitis pointed to the likelihood that an rxr homolog also exists in other filarial nematodes. Using degenerate PCR primers we were able to clone a fragment with high sequence similarity to Di-RXR-1 from O. volvulus cDNA (see Experimental Procedures; sequence deposited in GenBank ). In B. malayi, however, although we searched for an RXR type receptor in the genomic libraries available using Di-rxr-1 as a probe, no strongly hybridizing sequences were detected. While this work was in progress genomic data from the B. malayi genome project became available, which provided us with an alternative route to clone the B. malayi RXR/USP [51]. Using the sequence information from the other filarial species as well as the Brugia malayi genome project we designed a combined RT-PCR and RACE approach (described in detail in Experimental Procedures) that allowed us to obtain clones for the B. malayi homolog of RXR which we named Bma-RXR. The longest cDNA sequence identified for Bma-RXR is 1398 bp and encodes a 465 amino acid protein that has strong similarity to D. immitis Di-RXR-1 (100% amino acid identity in the DBD and 83% in the LBD). The amino acid sequence similarity between the B. malayi and D. immitis RXRs substantially deteriorates in the last exon. Interestingly, the last exon corresponds to the helix 12 region of the LBD where the activation function AF-2 usually resides (Fig. 5). This LBD region is also highly dissimilar between the filarial nematode RXRs and their homologs in other non-nematode species. Notably the motif LIRVL consistent with the RXR AF2 is found in Bma-RXR but not Di-RXR-1.
Similarly to Bma-EcR, global phylogenetic analysis places Bma-RXR together with USPs and RXRs (Supplementary material Figs. S1 and S2.). Using HNF4s as the outgroup, there is 100% aALRT (approximate Likelihood Ratio Test ) support to place Bma-RXR in the USP/RXR sub-family (Fig. 6). Relationships among arthropod USP/RXRs and Bma-RXR are not well resolved (aALRTs under 50%), but Bma-RXR groups strongly with Di-RXR-1. The grouping of Bma-RXR among USP/RXRs remains the same whether the Schistosoma mansoni sequences are included or not in the tree (data not shown). The Schistosoma sequences are extremely divergent, to the extent of not being phylogenetically informative [52],[53], and branch at the base of the tree. While it is known that dipteran and lepidopteran USPs evolve especially fast [53], Di-RXR-1 and Bma-RXR appear to have evolved even faster. Separate phylogenies of the DBD and LBD (not shown) indicate that this is entirely due to a very derived LBD. This observation is consistent with the alignment. The DBD on the other hand has evolved slowly, like the DBDs of its homologs in other species.
Expression of Bma-RXR was analyzed in adult females, males, and L1 larvae by Northen blot analysis (Fig. 7). A ∼5kb RNA species was clearly detected in female and male RNA samples. Low levels of the ∼5 kb Bma-RXR RNA species were also observed in the L1 RNA sample. Two additional Bma-RXR bands of approximately 3.75 kb and 3 kb were also detected in adult females. The presence of Bma-RXR mRNA in males and L1 larvae was in agreement with RT-PCR results (data not shown).
To further characterize the properties of Bma-EcR we tested whether Bma-EcR, by virtue of its LBD, is capable of forming a dimer with Bma-RXR, its putative native partner, to constitute a functional receptor and transduce the hormonal signal of ecdysteroids in a cellular context. The assay we employed takes advantage of the fact that the LBD of nuclear receptors can function in a modular fashion fused to heterologous DNA binding domains such as the GAL4 DBD [43],[44]. In order to test the ability of Bma-EcR LBD to activate transcription of a reporter gene in response to a particular hormone ligand, NIH 3T3 cells were co-transfected with GAL4:Bma-EcR(LBD) in combination with RXR LBDs fused to VP16. In addition to the Bma-RXR(LBD) we tested human HsRXR and Hs-LmRXR(LBD) (a chimeric human-locust LBD). The latter was selected because it shows no constitutive dimerization and high ligand-dependent activity when partnered with other ecdysone receptors [44] (and ). The transfected cells were tested for trans-activation in the absence or presence of either the ecdysteroid Ponasterone-A or the synthetic ecdysone agonist RSL1 by assaying luciferase activity.
Significant transactivation was detected when GAL4:Bma-EcR(LBD) was partnered with Bma-RXR(LBD) (Fig. 8A). The addition of RSL1 (Fig. 8B) or Ponasterone A (data not shown) had no further stimulatory effect on the detected activity. These data demonstrate that Bma-EcR and Bma-RXR are bona fide nuclear receptor partners and that, like their insect counterparts, they avidly dimerize in the absence of ligand. The ligands apparently cannot appreciably increase the heterodimer's ability to activate transcription above that of the VP16 activation domain in this assay.
Significant ligand-dependent transcriptional activation of luciferase was detected, however, when GAL4:Bma-EcR(LBD) was partnered with the chimeric VP16:Hs-LmRXR and treated either with Ponasterone-A or RSL1 (Fig. 8C). This is likely the result of ligand-dependent dimerization of the two receptor LBD fusions and subsequent trans-activation via the VP16 activation domain. This result clearly demonstrates the ability of Bma-EcR LBD to transduce the action of the ecdysteroid Ponasterone-A and the ecdysteroid agonist, RSL1 in the transfected cells.
The dimerization and transactivation studies presented here show that Bma-EcR is able to heterodimerize with Bma-RXR in a cellular context and capable of triggering a transcriptional response in an ecdysteroid-specific manner. These observations taken together along with their expression profile suggest that Bma-EcR and Bma-RXR have the prerequisite functional properties to constitute a functional Brugia malayi ecdysone receptor.
The existence of homologs for both protein components of Ecdysone Receptor in B. malayi which possess functional dimerization and DNA binding properties, and the earlier pharmacological observations by H. Rees [33],[34] suggest that ecdysone could function as a transcriptional regulatory ligand in B. malayi. To directly test this hypothesis, we employed a recently established transient transformation technique to explore whether ecdysteroids can activate transcription in B. malayi using a reporter assay. Recent studies have demonstrated that the 5′ UTR of the gene encoding the 12 kDa small subunit ribosomal protein of B. malayi (BmRPS12) was capable of acting as a promoter when used to drive the expression of a luciferase reporter gene in transiently transfected B. malayi embryos [45]. The BmRPS12 promoter contains 5 ¾ copies of an almost exact 44 nt repeat that acts as an enhancer element [45]. This promoter construct driving the expression of firefly luciferase (construct BmRPS12 (−641 to −1)/luc) was used to develop a reporter for B. malayi in which the enhancer repeat element was replaced with canonical ecdysone response elements (EcREs). We constructed the EcRE-BmRPS12-luciferase reporter (as described in Methods) using the PAL-1 element that Bma-ECR is capable of binding in vitro (Fig. 4B). This construct was tested for transcriptional activity in transfected B. malayi embryos, which were exposed to 20-OH ecdysone (20-E), or solvent alone, before being assayed for luciferase reporter activity. As shown in Fig. 9A ecdysteroid treatment resulted in a significant increase of reporter gene activity in cultures exposed to 20-E relative to control cultures (transfected in parallel with the same construct but exposed to solvent alone). This response to 20-E requires the presence of the EcRE sequence, since a construct lacking the EcRE did not exhibit any increase in luciferase activity in response to 20-E. Similarly, the response was strictly dependent on hormone, as constructs containing the EcRE produced levels of activity that were not significantly different from those obtained with the construct lacking the EcRE, in the absence of 20-E. Constructs containing the EcRE in both orientations were equally responsive to 20-E treatment, in keeping with previous studies demonstrating the symmetric nature of the binding of nuclear hormone receptors to their cognate response elements [43]. The response to the 20-OH ecdysone was dose-dependent, reaching a plateau at 5 µM (Fig. 9B). These results provide molecular evidence for the function of ecdysone in transcriptional responses of B. malayi and reveal the functional operation of a corresponding signaling system.
Molting in ecdyzosoans has been studied most extensively in insects. In insects EcR and USP initiate the transduction of the molt-triggering signal [4],[9]. Molting progression is mediated by the expression and activation of a number of well-characterized genes, including additional nuclear receptors [16],[17],[54,]. In contrast, in nematodes molting initiation and the molecular signaling responsible for its progression are only now starting to be understood. An RNAi screen in C. elegans for genes that are involved in molting has revealed a large number of “molting” genes, which encode proteins ranging from transcription factors and intercellular signaling molecules to proteases and protease inhibitors. However, no signal has been specifically identified as being the putative molting trigger [55]. Expression profiles of C. elegans “ecdysone cascade” nuclear receptors during molting cycles parallel the expression of their homologs in insects [56], and nhr-23, nhr-25, nhr-41, and nhr-85, the C. elegans orthologs of DHR3, Ftz-F1, DHR78, and E75, respectively, have been shown to be important for proper molting and/or dauer larva formation [26], [56]–[58]. The fact that the C. elegans genome contains no identifiable homologs of EcR or rxr [23] and that no ecdysteroids have been identified in this nematode, has led to the suggestion that ecdysone itself is unlikely to be the molting hormone in this free living nematode [55].
Our previous studies demonstrated the existence of an rxr homolog in the canine filarial nematode D. immitis [36]. The isolation of Di-rxr-1 indicated that, in contrast to C. elegans, filarial nematodes might contain different sets of NRs. The isolation of homologs of EcR and rxr in Brugia malayi presented here demonstrates that filarial nematodes express both components of the ecdysone receptor and these nuclear receptors show dimerization, DNA binding, and hormone-binding characteristics similar to those of the canonical insect ecdysone receptors. Our phylogenetic analyses place the two receptors in the corresponding branches of the superfamily tree. They also indicate a rapid evolution of the LBDs. The LBDs of nematode RXRs are extremely divergent, on a similar scale to that of Schistosoma RXR LBD. Subsequent to our identification of EcR and rxr homologs in Brugia, the sequencing of the genome was completed, identifying additional putative nuclear receptors in the ecdysone signaling cascade [51].
We cloned two Bma-EcR and one Bma-rxr mRNA isoforms. Northern blot analyses revealed Bma-EcR and Bma-rxr expression in adult males, females and L1s. In addition, RT-PCR analyses indicate that Bma-EcR is also present in L1, L2 and L3 larval stages. Since females contain developing embryos, it is not possible to differentiate between embryonic and female-specific expression of these two nuclear receptors in B. malayi. In insects EcR has been shown to be critical for both embryonic development and oogenesis [15],[59],[60] and in filarial nematodes ecdysone treatment releases meiotic arrest and stimulates microfilaria release [34]. Expression of EcR and rxr homologs in B. malayi females points to possible functions of the ecdysone receptor also in nematode oogenesis and/or embryogenesis.
The expression pattern of Bma-RXR differs somewhat from the expression pattern of the other filarial rxr identified to date, Di- rxr-1, which is expressed in males but not females [36]. In insects the rxr homologue “Ultraspiracle” (USP) is considered the main functional partner of EcR and as such its expression overlaps with that of EcR [5]. This also seems to be the case in B. malayi, where we observed that at least one isoform of Bma-rxr has an overlapping expression pattern with Bma-EcR. However, two other Bma-rxr isoforms appear to be specifically expressed only in females.
The sequence differences of B. malayi and D. immitis RXR may mirror differences in expression patterns of the two RXR homologues. Whether these differences in sequence and expression pattern correlate with differences in ligand interaction and/or function remains an open question.
Both Bma-EcRA and a putative isoform B are able to bind a canonical ecdysone response element (EcRE) when partnered with USP. The question of whether isoform B exists in B. malayi (as in D. immitis) remains unanswered. We have shown that such an isoform is biochemically active, being able to dimerize with an insect USP and bind EcRE in vitro. Furthermore, isoform B is the most similar to the insect EcRs. Bma-EcRB contains a shorter (i.e. canonical) “T-box” region than Bma-EcRA (Fig. 1). The “T-box” region has been described as being able to modulate DNA binding to extended hormone response elements [61]. The presence of possible sequence variation in the “T-box” region in these two Bma-EcR variants could point to the possibility of differences in isoform-specific interactions with DNA target sequences.
Bma-EcRC contains a truncated LBD, and it is similar in organization to the estrogen-alpha variant Delta-5, which displays dominant-negative activity [62]. As we have shown, isoform C is unable to dimerize with a bona fide USP to bind the palindromic EcRE. These data suggest that Bma-EcRC may carry out a novel function that is independent of any interactions with an RXR partner. Establishing the role of Bma-EcRC is the aim of future investigations.
The sequence in the region of helices 11–12 in the LBD of B. malayi and D. immitis EcR and RXR homologues is strikingly divergent when compared to each other and to other EcRs and RXRs respectively. The most prominent feature in Bma-EcR is the absence of conserved helix 12 residues. This difference raises the question of what constitutes a functional activation function corresponding to AF2 in these nematode members of the nuclear receptor family. Our transcriptional activation assay results clearly show that the two receptors can dimerize and that the LBD of Bma-EcR is capable of transducing an ecdysteroid signal in a cellular context. Even though our analysis was carried out in a heterologous system, this type of assay has been shown to be highly informative for LBD-ligand interactions [44]. In this system, however, strong constitutive dimerization of receptor partners can obscure possible transcriptional effects of the ligand. Our results obtained with the chimeric RXR-LBD (which confers low constitutive dimerization) as a partner, indicate that the Bma-EcR LBD does show an ecdysteroid response. Evidence of hormone binding from these transactivation assays and the absence of a recognizable AF2 motif in Bma-EcR suggest that this receptor utilizes different features to achieve equivalent transcriptional functions than its insect counterparts.
The identification of the putative ecdysone receptor components presented here provides strong support to the long standing hypothesis that ecdysteroids play a role in filarial nematode embryogenesis and molting similar to their role in insects.[4],[32]. Ecdysteroids have been detected in a number of nematodes (reviewed by Barker and Rees, [32]). When in vitro cultivation of Onchocerca volvulus microfilaria was attained, it was observed that the addition of 20E to the culture media resulted in L1 larva progressing to the infective L3 stage [63]. This observation is consistent with the fact that after the bloodmeal, mosquitoes raise their ecdysteroid level, which correlates with the subsequent rapid molting of the ingested L1 larvae to the L2 stage. We attempted to directly demonstrate that ecdysone can act as a transcriptional trigger in vivo using a transient transformation reporter assay. Indeed, significant activity was observed in response to ecdysone . Our transgenic Brugia experiments confirm the in vivo functionality of both a consensus EcRE and 20-hydroxyecdysone in measurable transcriptional activity. Although we present no data to establish that the observed activation is mediated by the receptor(s) we have cloned, our results in conjunction with previous studies on this subject confirm that filarial nematodes in particular, contain and express the gene components of a functional ecdysone signaling system that is quite similar to that of other ecdysozoa. The role of this signaling system in filarial development will be the subject of further studies. Furthermore, the existence of a functional ecdysone signaling pathway in filarial nematodes does point to the possibility of using a novel approach for the development of drugs to fight filariasis based on testing of pre-existing compounds that specifically target the ecdysone pathway [64].
|
10.1371/journal.pgen.1003255 | Transposon Variants and Their Effects on Gene Expression in Arabidopsis | Transposable elements (TEs) make up the majority of many plant genomes. Their transcription and transposition is controlled through siRNAs and epigenetic marks including DNA methylation. To dissect the interplay of siRNA–mediated regulation and TE evolution, and to examine how TE differences affect nearby gene expression, we investigated genome-wide differences in TEs, siRNAs, and gene expression among three Arabidopsis thaliana accessions. Both TE sequence polymorphisms and presence of linked TEs are positively correlated with intraspecific variation in gene expression. The expression of genes within 2 kb of conserved TEs is more stable than that of genes next to variant TEs harboring sequence polymorphisms. Polymorphism levels of TEs and closely linked adjacent genes are positively correlated as well. We also investigated the distribution of 24-nt-long siRNAs, which mediate TE repression. TEs targeted by uniquely mapping siRNAs are on average farther from coding genes, apparently because they more strongly suppress expression of adjacent genes. Furthermore, siRNAs, and especially uniquely mapping siRNAs, are enriched in TE regions missing in other accessions. Thus, targeting by uniquely mapping siRNAs appears to promote sequence deletions in TEs. Overall, our work indicates that siRNA–targeting of TEs may influence removal of sequences from the genome and hence evolution of gene expression in plants.
| Transposable elements (TEs) are selfish DNA sequences. Together with their immobilized derivatives, they account for a large fraction of eukaryotic genomes. TEs can affect nearby gene activity, either directly by disrupting regulatory sequences or indirectly through the host mechanisms used to prevent TE proliferation. A comparison of Arabidopsis thaliana genomes reveals rapid TE degeneration. We asked what drives TE degeneration and how often TE variation affects nearby gene expression. To answer these questions, we studied the interplay between TEs, DNA sequence variation, and short interfering RNAs (siRNAs) in three A. thaliana strains. We find sequence variation in genes and adjacent TEs to be correlated, from which we conclude either that TEs insert more often near polymorphic genes or that TEs next to polymorphic genes are less efficiently purged from the genome. We also noticed that processes that cause deletions within TEs and ones that silence TEs appear to be linked, because siRNA targeting is a predictor of sequence loss in accessions. Our work provides insight into the contribution of TEs to gene expression plasticity, and it links TE silencing mechanisms to the evolution of TE variation between genomes, thereby linking TE silencing mechanisms to expression plasticity.
| While transposable elements (TEs) constitute a large fraction of plant, animal and human genomes [1]–[3], their contribution to genome size can change rapidly during evolutionary time. In some taxa, TEs have been responsible for two-fold differences in genome size that arose over a few million years or less. These rapid fluctuations, which may be due to TEs being either more active or more efficiently deleted in certain species, indicate that control of TEs can differ greatly between closely related plant species [4]–[7]. The balance between TE transpositions and selection against TEs is influenced by factors ranging from mating system to silencing by short interfering RNAs (siRNAs) and chromatin modification. Therefore the control of TE activity and the removal of transposed copies can be considered key factors in the evolution of genomes.
TEs are often regarded as genomic parasites due to the potentially detrimental effects of insertional inactivation of genes and ectopic recombination of DNA [8]. Twenty-four nt long siRNAs are associated with most TEs as part of a ‘double-lock’ mechanism of siRNA-mediated DNA methylation that controls transposition via transcriptional repression, with a reinforcement loop between DNA methylation, histone methylation and siRNAs [reviewed in 9]. siRNAs are a robust proxy for DNA methylation at TEs, with unmethylated TEs generally lacking matching 24 nt siRNAs [10]–[13]. Most plant TEs have cytosine methylation at CG, CHG and CHH sites, but a quarter is unmethylated and a further 15% have atypical methylation patterns. In the TE-dense heterochromatin, DNA methylation can spread about 500 bp into neighboring unmethylated TEs [13]. In the euchromatin, methylation spreads from TEs to approximately 200 bp beyond the siRNA target sites [13], consistent with the effect of siRNAs on expression of proximal genes dissipating by 400 bp [14]. siRNA-targeted, methylated TEs are, on average, located farther away from expressed genes than TEs that are not strongly methylated or associated with siRNAs [13], [15]. As expected from this correlation, siRNA-targeted TEs have more effects on nearby gene expression than those without [14], [15].
Most poorly methylated TEs are short and have few CG dinucleotides [13]. This indicates a progression over evolutionary time from TEs that are active and targeted by siRNA-mediated DNA methylation, to inactive, degenerate relics that have changed through deletions and nucleotide substitutions initiated by deamination of methylated cytosines. These inactive TEs are then no longer targeted by siRNA-mediated DNA methylation.
Presumably because of interference with cis-regulatory elements, Arabidopsis TEs reduce the average expression levels of adjacent genes, although the distance over which these effects are noticeable varies between A. thaliana and A. lyrata [14]. Differences in TEs next to genes contribute to the divergence of gene expression levels between orthologs in these closely related species [14], and gene expression is negatively correlated with the number of nearby siRNA-targeted, methylated TEs [15].
In the selfing species A. thaliana, TEs account for only a fifth of the genome [7], [13], [16], making it relatively depauperate of TEs. Given that the A. thaliana genome is small relative to other members of the family and that its close relative A. lyrata, an outcrosser, contains approximately three times as many TEs [14], deletion of TEs in A. thaliana is likely an ongoing, active process. In accordance with this hypothesis, intraspecific polymorphisms and deletions in A. thaliana are disproportionately located within TEs and, to a lesser extent, intergenic regions [17]–[19].
A reference-guided assembly approach has been applied to accurately characterize complex sequence variation in several A. thaliana accessions [19]. Here, we exploit this information to examine TE variants and their effect on the expression of nearby genes in three divergent accessions. We report that TEs are more likely to be located in polymorphic regions of the genome. Where TEs are present in less polymorphic regions, they also tend to be less polymorphic themselves. Although polymorphic TE variants are less abundantly targeted by siRNAs, uniquely mapping siRNAs targeting polymorphic TE variants are strongly correlated with the TE regions that vary between accessions. These findings suggest a link between the ability to tolerate TE insertions, siRNA-mediated silencing and purging of TEs by deletion.
We annotated the sets of genes and TEs in three A. thaliana accessions: Col-0, Bur-0 and C24 [19], [20]. For reference accession Col-0, we used the TAIR9 annotation of TEs and protein-coding genes. Excluding centromeric sequences, 21,913 full-length and degenerate TEs and 26,541 genes were considered further. We built genome templates of Bur-0 and C24 from re-sequencing data using the SHORE pipeline [21]. The reference coordinates of TEs and genes were projected onto these genome templates, and variation in TEs and genes was determined based on single nucleotide polymorphisms (SNPs), 1 to 3 bp insertions/deletions (indels) and larger deletions of 4 to 11,464 bp (median 30 bp, mean 113 bp). Larger insertions were not included because of the high false-negative rate [17].
Comparison of polymorphism densities confirmed that coding regions were relatively depauperate of SNPs, indels and large deletions compared to intergenic regions and TEs (binomial test, p[Coding Regions/Intergenic Region] = 0 and p[Coding Regions/TE] = 0 for SNPs, indels or large deletions). Large deletions were significantly over-represented in TEs compared to intergenic regions, while SNPs and indels were not (Figure S1a; binomial test, p[TE/Intergenic Region] = 0 for large deletions). Over 6% of reference TEs differed by at least 10% of total length in each of the two accessions, Bur-0 and C24, compared to Col-0 (Figure 1a and Figure S2). Almost all of this variation, 93%, was due to large deletions (Figure S1b; for distribution of large deletion sizes see Figure S1c). We defined TEs with at least 10% variation by length (SNPs, indels and larger deletions combined), but not completely missing in Bur-0 or C24, as TE variants or VarTEs (please also see Figure S3 for abbreviation definitions). Close to 40% of VarTEs were shared between Bur-0 and C24 (Figure S4a).
TE density is highest in and next to the centromeres, where there are few genes. The fraction of VarTEs and the average level of TE variation were higher in the pericentromeric regions than on the gene-dense chromosome arms (Figure 1b; Mann-Whitney U [MWU] test, p<2×10−16 for Col-0 versus Bur-0/C24, Table S1 and Figures S5 and S6). To examine whether gene proximity biases TE variation across the chromosomes, we calculated the distance between TEs and protein-coding genes for Col-0. TEs were separated into two subsets: TEs within 2 kb of any gene, subsequently called proximal TEs, and TEs at least 2 kb away from the closest gene, called distal TEs. Distal TEs were on average more variable than proximal TEs (Figure 1c; Figures S7 and S8; MWU p[Col-0/Bur-0] = 0.001, p[Col-0/C24]<6×10−5). Proximity to protein-coding genes may therefore influence TE variation, consistent with TEs closer to genes likely being under stronger selective constraint [15], [22].
The correlation between TE variation and proximity to genes was compared among TE superfamilies [23], [24]. For non-centromeric TEs, LTR retrotransposons were more distal from genes, while no significant difference in distance to genes was observed for other TE superfamilies (Table S2). However, for proximal TEs there were differences among TE superfamilies in distance to genes and, as expected, TE superfamilies that are closer to genes (e.g. CACTA, MITE) were less variable than superfamilies located farther away from genes, e.g. non-LTR retrotransposons (Table S2).
To investigate the link between TE and proximal gene variation, we examined whether TE variation and location correlated with the polymorphism level of neighboring genes. We used the small-scale mutations to calculate the polymorphism level of non-centromeric genes. For each accession, genes were separated into two subsets; TE+ genes included genes within 2 kb of a TE and genes with TEs anywhere within the transcribed region, while TE- genes were at least 2 kb from the closest TE (Table S3). To be conservative, any TEs in Bur-0 or C24 with predicted deletions of at least 10% of the reference length were annotated as deleted. TE+ genes were on average more polymorphic than TE− genes in each accession (Figure 2a; MWU p<2×10−16 for Col-0, Bur-0 and C24). The same analysis was repeated for 80 resequenced A. thaliana accessions [17]; we could confirm the correlations observed with Bur-0 and C24 in these accessions.
Since polymorphism levels vary enormously among gene families, we further investigated whether there is a correlation of TE proximity with gene family using small-scale mutations from the 80 A. thaliana accessions (20, 61), and Col-0, C24 and Bur-0. Genes from highly polymorphic families such as those encoding NBS-LRR, F-box and Cytochrome P450s proteins were, on average, closer to TEs in all accessions (Figure S9; distance is negatively correlated with gene polymorphism, Spearman's ρ(Col-0) = −0.11, ρ(Bur-0) = −0.11, ρ(C24) = −0.10; p<2×10−16), including a higher proportion of genes having proximal TEs (Figure S10). TEs are therefore either more likely to insert into or near polymorphic genes, or are less efficiently purged from such regions.
To further examine the effects of TE variants on proximal genes, we divided TE+ genes into two subsets: genes where flanking TEs were <10% variant (Invariant TEs: InvTE) among the three accessions (InvTE+ genes), and genes where at least one flanking TE showed ≥10% sequence (VarTE) variation between accessions (VarTE+ genes; Table S3). Three quarters of VarTE+ genes were shared in comparisons between Col-0 and Bur-0 or Col-0 and C24 (Figure S4b). The VarTE+ genes were on average more polymorphic than InvTE+ genes (Figure 2b; MWU p = 0.005), also in the 80 accessions dataset [17]. We conclude that TEs close to genes are less polymorphic, while genes close to polymorphic TEs are themselves more polymorphic.
A correlation between polymorphism levels of TEs and nearby genes is insufficient to address whether this is a direct link as opposed to high directional selection pressure on the genomic region in general. To address this question, we therefore compared the polymorphism level of TEs, the flanking regions and nearby genes. TEs in highly polymorphic regions are themselves more polymorphic than TEs in regions of low divergence (Figure S11a; binomial test, p = 0), with the exception that TEs in highly polymorphic regions with nearby lowly polymorphic genes show a similar level of divergence as TEs in regions of low polymorphism with no coding genes. Moreover, TEs in gene-free regions show significantly higher divergence than TEs within 4 kb of a gene, especially if those genes are less polymorphic. TEs are generally more polymorphic than their flanking sequences (binomial test, p = 0), with the exception of TEs in highly polymorphic regions with lowly polymorphic gene. The results for large deletions (Figure S11b) are consistent with our observation from Figure S1 that large deletions are over-represented in TEs compared to intergenic regions. Notably, there is no significant difference in the level of small-scale mutations between TEs and flanking regions (Figure S11c). Taken together, TE variation through large deletions shows a positive correlation with flanking region polymorphism level, but is also strongly influenced by the conservation and presence/absence of nearby genes. The frequency of large deletions is however generally higher in TEs than in the flanking regions, indicating positive selection for large deletions within TEs.
Genes that are close to TEs (TE+ genes) tend to have a lower expression average than TE− genes in the Col-0 reference accession [15]. We set out to determine whether this was true for the accessions studied here as well. Gene expression was measured using Affymetrix tiling arrays and RNA extracted from floral tissue of each accession. We considered presence/absence of TEs in the flanking regions of genes, taking into account the number of linked TE insertions and the distance from each gene to the closest TE. We confirmed the reported pattern for Col-0 [15], and found that it applies to Bur-0 and C24 as well. In all three accessions, genes with proximal TEs (TE+ genes) were on average expressed at lower levels than those without proximal TEs (TE− genes; Figure 3a; MWU p<2×10−16 for Col-0, Bur-0 and C24). This effect was even stronger if TEs were located simultaneously within, upstream and downstream of the gene (Figure 3a; MWU p≤2×10−14 for Col-0, Bur-0 and C24). Moreover, the average expression level of neighboring genes was positively correlated with the distance to the nearest TE (Figure 3b; Spearman's ρ(Col-0) = 0.15, ρ(Bur-0) = 0.13, ρ(C24) = 0.13; p<2×10−16), and negatively correlated with the number of proximal TEs (Figure 3c; df = 55, chi-square sums 915, 588 and 553 for Col-0, Bur-0 and C24, respectively, p<2×10−16). Thus, gene expression is suppressed by proximal TEs, especially if they are close to the gene and numerous.
Since TE superfamilies may have different effects on proximal genes, we examined gene expression according to the TE superfamily of the closest proximal TE. TE+ genes are expressed differentially depending on the TE superfamily of the proximal TE. TE+ genes with DNA transposons are on average expressed at a higher level compared to TE+ genes surrounded by retrotransposons (Figure S12; MWU, p = 0.02 for Col-0, Bur-0 and C24). However, this is solely due to the higher expression level of genes proximal to CACTA elements. Indeed, we did not find evidence for CACTA TEs having any effect on gene expression (Figure S12, MWU, p(CACTA TE+ genes/TE− genes) = 0.7, 0.6 and 0.8 for Col-0, Bur-0 and C24, respectively), which may explain why they are on average closer to genes than TEs from other families. Within the retrotransposons, LTR retrotransposons are younger on average than non-LTR retrotransposons and have a greater suppressive effect on proximal genes (Table S2; [25]). Therefore TE superfamilies can differ considerably in their effects on proximal genes.
TEs suppress the expression of neighboring genes at least partially through DNA methylation, which in turn is linked to 24-nt long siRNAs [12], [15], [22], [26], [27]. To investigate the influence of siRNAs on TE silencing, we sequenced siRNAs from mixed inflorescence tissue (shoot meristem plus flowers, stages 1–14) of each accession and mapped the reads to all possible positions of the respective genomes without any mismatches. As expected from previous work, the density of siRNAs over TEs was about four times higher than the genome average (Table S4; Figure S13).
We have reported before that siRNA-targeted TEs are more effective in suppressing expression of neighboring genes than are non-siRNA-targeted TEs, and that they are farther from genes [15]. We determined whether this held true in the current, more comprehensive dataset. If at least one 24-nt siRNA mapped to a TE it was labeled as siRNA+ (Table S5). siRNA+ and siRNA− TEs were overall similar in number, but retrotransposons were targeted by siRNAs more frequently than DNA transposons (Figure S14; binomial test, p = 0 for Col-0, Bur-0 and C24). siRNA+ TEs were farther from genes (Figure 4a; Figure S15a; MWU p<2.2×10−16 for Col-0, Bur-0 and C24), and this bias was consistent among TE superfamilies (Figure S16). To examine the effects of siRNA-targeting on the expression of flanking genes, we classified genes by whether the nearest TE was siRNA+ or siRNA− (Table S5). In each accession, genes flanked by siRNA+ TEs had lower average expression levels than genes with adjacent siRNA− TEs (Figure 4b; Figure S15b; MWU p[Col-0] = 0.0001, p[Bur-0] = 0.002, p[C24] = 2×10−6). The effect of suppression was stronger if the closest siRNA+ TE was within 2 kb of the gene (Figure 4b; Figure S15b; MWU p<2×10−16 for Col-0, Bur-0 and C24). Therefore, as found previously for Col-0, siRNA-targeting of TEs represses nearby genes and TEs that are close to genes are less likely to be targeted by siRNAs, either due to stronger selection for deletion of siRNA-targeted TEs close to genes or selection against siRNA-targeting of these TEs.
Because siRNAs that map to unique positions in the genome (usiRNAs) correlate more closely with DNA methylation than siRNAs that map to multiple positions (msiRNAs; [12]), we investigated whether usiRNAs and msiRNAs target TEs differentially, and how usiRNA− and msiRNA-targeted TEs might affect the expression of nearby genes. All TEs with at least one usiRNA were labeled as usiRNA+ (Table S5). In both Bur-0 and C24, over 83% of siRNA+ TEs were usiRNA+, similar to what has been reported for Col-0 [14]. usiRNA+ TEs were farther away from genes than msiRNA+ TEs (Figure 4a; Figure S15a; MWU p[Col-0]<2×10−16, p[Bur-0] = 6×10−13 and p[C24] = 2×10−6). We also observed that the average expression level of genes within 2 kb of usiRNA+ TEs was lower than the expression of genes within 2 kb of msiRNA+ TEs (Figure 4b; Figure S15b; MWU p[Col-0] = 3×10−6, p[Bur-0] = 5×10−5, p[C24] = 0.01). Therefore, even though TEs targeted by usiRNAs and msiRNAs are on average farther from genes, they more strongly reduce expression of proximal genes compared to TEs targeted by only msiRNAs. Overall, we confirmed that siRNA+ TEs, especially usiRNA+ TEs, suppress neighboring gene expression, consistent with a trade-off between reduced TE mobility and deleterious effects on neighboring gene expression [14], [15].
If TEs suppress the expression of adjacent genes, presence of gene-proximal TEs in the different accessions should be associated with differences in expression levels of proximal genes. We found that expression of TE− genes varied less between accessions than TE+ genes, and further that expression varied less between genes proximal to invariant TEs (InvTE+ genes) than genes proximal to variant TEs (VarTE+ genes; Figure 5a; MWU p[TE−/TE+]<2×10−16, p[InvTE+/VarTE+] = 2×10−5). However, because TEs, and especially VarTEs, are found more often next to polymorphic genes, these conclusions could be confounded by correlated differences in genic polymorphisms. We therefore classified genes based on the extent of sequence variation (Table S6). Regardless of degree of genic polymorphism, VarTE+ genes were the ones that varied most in expression between accessions (Figure 5b), indicating that TE variation increases variance in gene expression.
We next determined whether differential siRNA-targeting influences gene expression. To remove the potentially confounding effects of variation in TEs themselves, we focused on InvTE+ genes and grouped these based on whether siRNAs for the adjacent TE could be detected in either all or none of the three accessions, or whether accessions differed in siRNA-targeting of the adjacent TE. We found that while variation in siRNA-targeting increased expression differences between accessions, this increase was not statistically significant (Figure 5a). It should be noted that in our analysis we could not distinguish between the effects of differential siRNA-targeting and any perturbations of cis-regulatory sequences.
Since each TE that differs in presence/absence or each siRNA-targeting variant between accessions represents a natural mutagenesis experiment, this offers an opportunity to study the effects on individual genes, to confirm the inferences drawn from averaging over all genes. We selected siRNA+ TE+ genes in Col-0 that are siRNA− TE+ or TE− in Bur-0 or C24 and tested for differential expression between Bur-0 or C24 and Col-0. To remove the potential confounding effect of genic polymorphism, we excluded genes with a polymorphism level greater than 2%. Overall 706 genes were retained for this analysis. The effect of siRNA-targeting on gene expression was further verified by comparing expression profiles among wild-type, rdr2-1 and a ddc (drm1drm2cmt3) DNA methyltransferase triple mutant [28]. Fifteen genes out of 706 showed significant up-regulation (top 5% ranking) in Bur-0 or C24 and in at least one of the RNA silencing mutants (Table S7). Although not statistically significant, this observation is consistent with siRNA-targeting and TE presence affecting gene expression. Moreover, it is likely an underestimate of TE effects on gene expression, given our stringent selection criteria.
Because siRNA+ TEs suppress neighboring gene expression particularly efficiently, we asked whether targeting of different regions of TEs was reflected in the expression of adjacent genes. We first investigated whether invariant and variant TEs (InvTEs and VarTEs) differed in siRNA-targeting, normalized by TE length, and whether there were differences between invariable and variable regions of VarTEs (Figure 6a; Table S8). Fewer siRNAs mapped to siRNA+ VarTEs than to siRNA+ InvTEs (Figure 6a; MWU p<2×10−16 for Col-0 versus Bur-0/C24), but there were more siRNAs in variable regions than invariable regions of siRNA+ VarTEs in Col-0 (Figure 6a; MWU p[Col-0/Bur-0] = 1×10−5, p[Col-0/C24]<2×10−16). Furthermore, usiRNAs were overrepresented in variable regions (binomial test, p[Col-0/Bur-0] = 7×10−18, p[Col-0/C24] = 0), while msiRNAs were biased towards invariable regions (p[Col-0/Bur-0] = 1×10−6, p[Col-0/C24] = 0). Therefore, usiRNAs strongly correlate with variability of TE sequences and are over-represented in the variable regions of variant TEs.
This finding raised the question whether TE regions that varied between accessions and were targeted by siRNAs had a particularly large effect on expression of adjacent genes. We therefore separated Col-0 genes within 2 kb of variable TEs into three subsets: genes next to siRNA− VarTEs (siRNA− VarTE+ genes); genes next to VarTEs with an siRNA-targeting bias towards invariable TE regions (InvsiRNA+ VarTE+ genes); and genes next to VarTEs with an siRNAs targeting bias towards variable TE regions (VarsiRNA+ VarTE+ genes; Table S8). As expected, siRNA− VarTE+ genes had a higher average expression level compared to InvsiRNA+ VarTE+ genes (Figure 6b; MWU p[Col-0/C24] = 0.01, p[Col-0/Bur-0] = 0.01) or VarsiRNA+ VarTE+ genes (MWU p[Col-0/C24] = 9×10−5, p[Col-0/Bur-0] = 0.003). The InvsiRNA+ VarTE+ genes, however, were expressed on average more highly than the VarsiRNA+ VarTE+ set (MWU p[Col-0/C24] = 0.01, p[Col-0/Bur-0] = 0.04). This indicates that gene suppression by neighboring TEs may not only be influenced by siRNA presence or absence at the TEs, but may also depend on which TE regions are targeted by siRNAs. We speculate that siRNA-targeting of particular TE regions suppresses the expression of nearby genes to such an extent that there is significantly higher selection pressure for these regions to be excised or mutated. Alternatively, due to the skew of usiRNA mapping towards variable regions, and the greater correlation between usiRNAs and TE methylation, the lower expression level of VarsiRNA+ VarTE+ genes may reflect a higher degree of epigenetic silencing of these elements compared to InvsiRNA+ VarTE+ genes.
TEs constitute the majority of DNA in many plant genomes [2], [3]. Evolutionary dynamics vary among TE types and they are affected, for example, by species demography and mating system [29]. A number of measures counteract the proliferation of TEs including TE silencing and removal. Because TE deletions via illegitimate recombination and unequal intra-strand homologous recombination are common [30]–[33], it is important to understand how changes in TE composition affect nearby gene expression. We have studied the interactions of TE variants, genic polymorphism, gene expression, and siRNA-targeting in Arabidopsis thaliana. We have shown that there is substantial variation in TEs between accessions primarily through large deletions, with invariant TEs on average closer to genes than variant TEs. We have confirmed that gene expression is positively correlated with distance to the nearest TE, and negatively correlated with the number of proximal TEs. While variation within a TE has some effect on the expression of adjacent genes, genes close to TEs are also on average more polymorphic than those that are not. Perhaps our most interesting observation is the increased usiRNA-targeting in TE regions that are variable between accessions compared to TE regions that are invariant.
TEs may be prevented from reaching fixation within a population through negative selection, especially for gene-proximal, methylated TEs [13], [15], [34]. Therefore, it is perhaps unsurprising that TEs are over-represented in analyses of structural variants among accessions and between species [17], [18], [35], [36], and that a recent comparison of 80 A. thaliana genomes reported evidence of structural variation in 80% of TEs [17]. Similarly, Hollister and Gaut [15] found that 44% of over 600 TE insertions were polymorphic among 48 accessions. Since most TEs in A. thaliana are relatively old [7], the simplest way to explain these patterns is ongoing deletion of TEs, which is also consistent with TEs in A. thaliana being on average farther from genes than in the closely related but outcrossing A. lyrata [7]. This may, however, be too simplistic an explanation as non-LTR retrotransposons are skewed towards an older insertion distribution than LTR retrotransposons [25], even though they are not significantly more variable (Table S2). While TE presence/absence polymorphisms in different accessions have been previously characterized [17], we have shown that there is substantial sequence variation in about 6% of TEs when comparing accessions (Figure 1a). These TE variants are equally distributed throughout the genome (Figure 1b).
TEs can affect the expression of proximal genes via mechanisms including disruption of promoter sequences, reduction of transcription through the spread of epigenetic silencing [13], or read-though antisense transcription [37]. Often TEs suppress the expression of proximal coding genes [15], [22], [38] however, TEs can also introduce new promoter sequences, leading to up-regulation of proximal genes [37]. In both plants and animals, TE-derived sequences have been recruited to form regulatory sequences and have contributed to coding regions [8], [39]–[42].
Methylated TEs suppress expression of proximal genes in A. thaliana, regardless of insertion upstream or downstream of the coding region. Purifying selection is therefore greatest for methylated TEs proximal to genes [15]. Notably, the effects of siRNAs on expression of proximal genes can only be detected up to 400 bp [14], while measurable TE effects extend to 2 kb [14]. This supports the assertion that TEs either directly affect gene expression by disruption of positive regulatory sequences, or otherwise act through DNA structure and epigenetic marks to affect genes over longer distances.
We found that TEs that with variable siRNA-targeting do not affect proximal genes more strongly than TEs that are targeted in all three accessions (Figure 5). It is possible that siRNA-targeting varies independently of TE sequence variation, as observed recently for DNA methylation [43], and that such TEs mask more subtle differences between the TE classes examined. However, the region of the TE targeted by siRNAs does seem to matter, with siRNA-targeting of TE sequences within an accession that are variant/absent in other accessions showing a greater suppression of proximal genes (Figure 6). This agrees with the observation that genes close to usiRNA-targeted TEs have a lower expression average than those close to msiRNA-targeted TEs, and that usiRNAs are over-represented in the variable regions of transposons. A recent study of hybrids between parents of different ploidy found that a reduction in 24 nt siRNAs is associated with up-regulation of more TE-associated genes than when there is no significant change in siRNA levels [44]. This result supports the hypothesis that siRNAs, or linked epigenetic changes, can affect the expression of nearby genes, with deletion of the siRNA-targeted regions alleviating repression of adjacent genes.
While TEs in the euchromatin are often found close to genes, methylated TEs are underrepresented upstream of genes, likely because changes in the promoter more easily affect gene expression than variation in the 3′ region [13]. In agreement, methylated TEs have a skewed distribution, with older elements farther from genes, but unmethylated TEs do not show such a bias [15]. In a comparison of humans and chimpanzees, TE insertion site preference appears to be the main cause for TEs being found more often in the vicinity of genes with increased interspecific expression variation [45]. This is reminiscent of what we have observed, with additive effects of polymorphism, TE presence and TE variance on the variability of orthologous gene expression (Figure 2 and Figure 4). In a comparison of two rice subspecies, TE presence/absence polymorphisms were also found to be underrepresented in SNP deserts [35]. There are several possible explanations for these observations: some genomic regions may suffer from generally elevated mutation rates TEs near highly conserved genes are more efficiently purged; or TE integration into more mutable genomic regions is favored. In the latter case, new mutations may destabilize DNA packing and facilitate TE insertions, similar to the TE insertion preference for transcribed genomic regions [42].
With our observation of TE deletions correlating with siRNA-targeting, we can expand the current model for TE evolution [15]. Our model starts with the duplication of a TE that is already present and targeted by siRNAs within the genome (Figure 7a and 7b), leading to all siRNAs produced by and targeting the original TE now being multiply-mapping siRNAs (msiRNAs). As the two copies of the duplicated TE gain mutations (enhanced by deamination of methylated cytosines), uniquely-mapping siRNAs (usiRNAs) are produced in addition to msiRNAs (Figure 7c). Hollister and colleagues [14] noted that usiRNA-targeting increases with TE age, while msiRNA-targeting decreases, and that TEs are expressed at lower levels when also targeted by usiRNAs. Furthermore, usiRNAs are more closely correlated with DNA methylation than are msiRNAs [12] and they are expressed at higher levels than msiRNAs [14]. With usiRNAs, the duplicated TEs will therefore be more effectively silenced, probably with a concurrent increase in methylation, a further reduction in the expression level of proximal genes, and thus increased selection against the TEs.
usiRNA-targeting may then facilitate TE inactivation through preferential deletion of usiRNA-targeted regions (Figure 6 and Figure 7d). This may be actively promoted by the usiRNAs and attendant epigenetic marks, in a mechanism analogous to the siRNA-guided removal of “internal eliminated sequences” including TEs in Tetrahymena [46], [47]. In favor of such a scenario, small deletions within TEs have been shown to occur more frequently than ectopic recombination events at the LTRs [31], [48]. Ectopic recombination appears to be less important for TE elimination in A. thaliana, as TE density and recombination rate are not correlated in this species [48], and because ectopic recombination is lower in homozygotes [49]. No matter what the mechanism, deletions within TEs would reduce selection pressure by removing usiRNA target sites, inactivating TEs so they are no longer transposition-competent, and relieving proximal gene repression.
In apparent contrast to the majority of TEs, some are under positive selection [50], [51], and TEs can also contribute to new regulatory networks [52]. Our model is only appropriate for TEs under neutral or negative selection. Modeling of TE dynamics suggests that transposition events occur in a cyclical manner [53], [54], with some activation events creating new favorable genetic variants. One such example is provided by transposition of a TE that is induced upon heat stress in genetic backgrounds impaired in siRNA biogenesis confers heat-responsiveness to proximal genes [55].
We have exploited high-quality genome information from multiple accessions of a single species to study the effects of TE variation on proximal gene expression. We discovered a link between siRNA-targeting and TE variation that illuminates how epigenetic mechanisms may help to shape genomes, but several questions remain: Do usiRNAs directly facilitate TE deletions or do they act indirectly through differences in selection for deletions? Are TE deletions in other species also associated with regions of increased usiRNA-targeting? And do species differ in the rate of TE deletion via this mechanism? Because of the rarity of TE deletions, this is a challenging process to dissect. Genomes with a large fraction of TEs, such as those of many crop plants, might therefore prove more tractable systems for studying mechanism of TE removal than the TE poor A. thaliana genome.
We extracted positions of genes and TEs from the A. thaliana Col-0 genome sequence TAIR version 9 from http://www.arabidopsis.org. We excluded genes and TEs within the centromeric regions [56]. To define gene and TE sets in Bur-0 and C24, we built genome templates using published Illumina paired-end reads of Bur-0 and C24 [19]. We used the SHORE pipeline [21] to align the reads to the Col-0 reference genome and extracted the consensus sequences as genome templates by calling bases with quality>24, support>6, concordance>0.7 and average hits = 1. We then applied a naïve projection of the coordinates of genes and TEs from Col-0 onto the genome templates to define the gene and TE sets of Bur-0 and C24. SHORE was also used to detect genomic variations by calling SNPs, small (1–3 bp) insertions/deletions and larger deletions from the genome templates of Bur-0 and C24 compared to the Col-0 genome using the same parameters for quality control. The distance between TEs and genes in Bur-0 and C24 was estimated from Col-0 using the annotated TE and gene coordinates, and adjusted to account for insertions and deletions between TEs and genes.
For each polymorphism type (i.e., SNPs, small indels, and large deletions), we compared the densities pairwise across coding regions, intergenic regions and TEs. To test whether a higher density was significant in a particular genomic region (e.g. TE) compared to others (e.g. coding region), a cumulative binomial probability distribution was applied:p is the polymorphism density in coding regions, and k and n are the total number of polymorphic sites in TEs and the total length of TEs, respectively.
We calculated gene polymorphism levels as the fraction of genic region containing small-scale variations in at least C24 or Bur-0, or one of the 80 A. thaliana accessions [17]. Genes with more than 20% zero sequencing coverage or no base calls among 80 accessions were excluded from the analysis.
4 kb 5′ and 3′ flanking regions for each TE were extracted. For each flanking region (FR), or genic regions (GR) within the FR, small-scale mutations and large deletion polymorphisms between Col-0 and Bur-0/C24 were calculated. Using all mutations, the polymorphism levels of TEs, FRs and GRs were ranked. A threshold of 50% was used to split FRs and GRs into high or low polymorphism datasets and thereby classify the TEs by genomic environment. The polymorphism levels of the FRs were calculated in 200 bp bins for each group of TEs, with binomial tests to compare polymorphism levels between TEs and FRs, and between different TE groups.
Inflorescences (meristem and flowers up to stage 14) were pooled from five plants of each accession grown at 23°C. Triplicate samples were collected between 7 and 8 hours into a 16 hour light cycle. RNA was extracted using the Qiagen (Hilden, Germany) Plant RNeasy Mini kit. Each biological replicate was analyzed with Affymetrix (Santa Clara, CA, USA) tiling 1.0R arrays and the data were processed according to published methods [57], [58]. Tiling array probes that were polymorphic for C24 or Bur-0 were removed from the dataset for the affected accession(s). For gene expression estimates, ≥70% and at least 3 probes had to be present; all other genes were not considered.
Tilling array data from Arabidopsis Col-0 and the RNA silencing mutants rdr2-1 and ddc (drm1-1;drm2-2;cmt3-11) mutants were downloaded from GEO (GSE12549; [28]) and processed according to published methods [57], [58]. Expression level changes for each dataset were estimated by fold-change differences between Bur-0/C24 and Col-0, and between the RNA silencing mutants and wild type Col-0. Background distributions of fold-change were calculated and genes, with a fold-change exceeding a one-sided 95% quantile in each dataset were considered as significantly up-regulated in Bur-0/C24 or the mutants.
The siRNA datasets have been published [19] (GEO accession number GSE24569). We mapped the 24-nt siRNA reads onto both strands of the genome templates (see below) and the TEs of Col-0, Bur-0 and C24, respectively, using the Vmatch package (http://www.vmatch.de). Only reads with perfect matches were considered.
The statistical significance of over-representation of usiRNAs or msiRNAs within the variable regions of siRNA+ VarTEs in comparison to all siRNAs was tested using the cumulative binomial probability distribution given above. p, expected frequency, is the ratio between the number of siRNAs mapped to the variable regions the total number of siRNAs mapped to any region of siRNA+ VarTEs, and n and k are the total number of usiRNAs/msiRNAs mapped to any region and the number of usiRNAs/msiRNAs mapped to the variable regions, respectively.
We defined an siRNA+ VarTE as either InvsiRNA+ or VarsiRNA+ if siRNAs are overrepresented in the invariable regions and variable regions, respectively. For siRNA+ VarTEs that contain siRNAs in both variable and invariable regions, we employed the cumulative binomial probability distribution described above to test whether siRNA-targeting shows statistically significant bias towards variable or invariable regions. For each siRNA+ VarTE, p in the formula above is the abundance of siRNA-targeting at the TE. To test the bias towards variable regions, n and k represent the genomic length of variable regions and the number of siRNAs targeting variable regions, respectively. Similarly, to test the bias towards invariable regions, n and k represent the genomic length of invariable regions and the number of siRNAs targeting invariable regions, respectively. P-values were adjusted for multiple hypothesis testing with the Benjamini-Hochberg method to control for a false discovery rate of 5% [59].
The siRNA and microarray data reported in this paper have been deposited in the National Center for Biotechnology Information Gene Expression Omnibus (NCBI GEO) (http://www.ncbi.nlm.nih.gov/geo/) under accession numbers GSE24569 and GSE24669. The genome assemblies are available from http://1001genomes.org/projects/MPIWang2012/ while the transposable element annotations for Bur-0 and C24 are available from Dryad under doi 10.5061/dryad.8674d.
|
10.1371/journal.ppat.1004975 | Border Patrol Gone Awry: Lung NKT Cell Activation by Francisella tularensis Exacerbates Tularemia-Like Disease | The respiratory mucosa is a major site for pathogen invasion and, hence, a site requiring constant immune surveillance. The type I, semi-invariant natural killer T (NKT) cells are enriched within the lung vasculature. Despite optimal positioning, the role of NKT cells in respiratory infectious diseases remains poorly understood. Hence, we assessed their function in a murine model of pulmonary tularemia—because tularemia is a sepsis-like proinflammatory disease and NKT cells are known to control the cellular and humoral responses underlying sepsis. Here we show for the first time that respiratory infection with Francisella tularensis live vaccine strain resulted in rapid accumulation of NKT cells within the lung interstitium. Activated NKT cells produced interferon-γ and promoted both local and systemic proinflammatory responses. Consistent with these results, NKT cell-deficient mice showed reduced inflammatory cytokine and chemokine response yet they survived the infection better than their wild type counterparts. Strikingly, NKT cell-deficient mice had increased lymphocytic infiltration in the lungs that organized into tertiary lymphoid structures resembling induced bronchus-associated lymphoid tissue (iBALT) at the peak of infection. Thus, NKT cell activation by F. tularensis infection hampers iBALT formation and promotes a systemic proinflammatory response, which exacerbates severe pulmonary tularemia-like disease in mice.
| NKT cells are innate-like lymphocytes with a demonstrated role in a wide range of diseases. Often cited for their ability to rapidly produce a variety of cytokines upon activation, they have long been appreciated for their ability to “jump-start” the immune system and to shape the quality of both the innate and adaptive response. This understanding of their function has been deduced from in vitro experiments or through the in vivo administration of highly potent, chemically synthesized lipid ligands, which may not necessarily reflect a physiologically relevant response as observed in a natural infection. Using a mouse model of pulmonary tularemia, we report that intranasal infection with the live vaccine strain of F. tularensis rapidly activates NKT cells and promotes systemic inflammation, increased tissue damage, and a dysregulated immune response resulting in increased morbidity and mortality in infected mice. Our data highlight the detrimental effects of NKT cell activation and identify a potential new target for therapies against pulmonary tularemia.
| The respiratory mucosa is a major site for pathogen entry and hence, requires constant immune surveillance. Like other mucosal surfaces, the lungs are populated by a variety of innate cells and innate-like lymphocytes. One such cell type, the type I, semi-invariant natural killer T (NKT) cells, are enriched within the lung vasculature where they are optimally positioned for early antigen encounter [1]. These pulmonary NKT cells exert diverse functions dependent upon experimental settings [2].
NKT cells express an invariant TCR α-chain (Vα14-Jα18 in mice and Vα24-Jα18 in humans) and one of a restricted set of TCR β-chains and, hence, called semi-invariant. Their name also reflects their hybrid nature, in that they co-express markers of both NK cells and conventional T cells. Their innate-like character is reflected in their ability to rapidly respond to stimulation by producing a wide variety of cytokines [3]. Several subsets of NKT cells have been identified, each of which may have distinct functional consequences in disease conditions where NKT cells are known to play a role [3–7]. NKT cell functions are controlled by microbial or self-glycolipids presented by CD1d molecules or by pro-inflammatory cytokines produced by activated antigen presenting cells (APCs). The quality and magnitude of the NKT cell response is determined by the mode of activation and the chemical nature of the activating lipid(s) [4]. Activated NKT cells can stimulate APCs, natural killer cells, and other leukocytes through the expression of cytokines and costimulatory molecules, thus functioning at the interface between innate and adaptive immunity [4]. Consequently, NKT cells control microbial and tumour immunity as well as autoimmune diseases [8–10].
In the lungs, NKT cells promote inflammation in models of airway hyperreactivity (AHR), acute lung injury (ALI), and chronic obstructive pulmonary disease (COPD) [2]. NKT cells may also contribute to the inflammatory cascade accompanying sepsis, which is often a complication of bacterial infections of the lung [11,12]. In general, pulmonary NKT cells are thought to play a protective role in microbial infections, but in some cases, may also exacerbate disease [2]. However, the mechanisms by which pulmonary NKT cells contribute to disease pathology remain poorly defined. The disparate results encountered in the literature are likely due to the differential function of individual NKT cell subsets, the various means by which NKT cells are activated in different disease settings, and the use of different NKT-deficient mouse models [13,14].
To probe the function of lung NKT cells, we chose a respiratory Francisella tularensis infection model as this infection causes lethal pulmonary tularemia. F. tularensis is a gram-negative facultative intracellular bacterium, which infects multiple cell types including macrophages, dendritic cells, hepatocytes, neutrophils, and epithelial cells [15,16]. The resulting disease targets multiple organs and manifests itself in several forms of differing severity depending on the inciting bacterial strain, dose, and route of infection. Of these, the respiratory route is the most deadly, and the most likely route of infection by weaponized F. tularensis [17]. After inhalation, patients typically show signs of systemic illness, which may be accompanied by immediate signs of respiratory disease and can result in death in 30–60% of cases if left untreated [18–21]. Although the exact cause of death is unclear, it is likely due to an overwhelming systemic inflammatory response [22]. Mice inoculated intranasally (i.n.) with F. tularensis fail to mount an effective immune response for the first 48–72h. After this initial immune latency, the response to F. tularensis is characterized by a robust local and systemic “cytokine storm” reminiscent of sepsis [23,24]. Little is known about the role of NKT cells in pulmonary tularemia—in either humans or mice, due in part to the difficulties in distinguishing them from NK cells, which protect mice from tularemia-like disease-but a beneficial role has been implied [25]. Each of the cell types known to be susceptible to F. tularensis infection has been shown to activate NKT cells [4]. Therefore, we reasoned that NKT cells might be activated very early after infection and could function in shaping the quality of both the innate and adaptive response.
The results emerging from testing the afore hypothesis revealed that indeed respiratory infection with F. tularensis activated iNKT cells which produced interferon (IFN)-γ and propagated a sepsis-like proinflammatory response that led to a lethal phenotype in wild type mice. This proinflammatory response was much tempered in CD1d-deficient mice that lacked NKT cells. Strikingly, however, the mutant mice had increased lymphocytic infiltration in the lungs that organized into structures resembling induced bronchus-associated lymphoid tissue (iBALT) at the peak of infection. Hence, NKT cell activation by F tularensis infection hampers iBALT formation, which in conjunction with an NKT cell-dependent proinflammatory response, exacerbates severe pulmonary tularemia-like disease in mice.
IFN-γ is critically required for murine resistance to primary F. tularensis LVS infection. The early production of IFN-γ after pulmonary LVS infection is primarily attributed to NK cells and double negative T cells [26]. NKT cells are also a source of early IFN-γ and hence, we reasoned that they might contribute to resistance to i.n. LVS infection. We therefore investigated whether NKT cell deficiency would alter disease outcome. Studies of immune function in infectious disease rely on one of two methods—genetic deletion or in vivo depletion—by which to assess their contribution to resistance or pathology. Because no unique NKT cell-specific marker has been identified, the available Ab-mediated depletion methods deplete both NK and NKT cells as these two cell types have significant overlap in surface marker expression, making interpretation difficult. Mice made genetically deficient in NKT cells are therefore a better experimental model and were generated previously either by deletion of one of the TCR α-chain gene segments (Jα18-/-) or by mutation of the restriction element required for thymic NKT cell selection and antigen presentation in peripheral tissues (CD1d-/-) [27–30].
To determine how NKT cell deficiency affects the outcome of pulmonary F. tularensis infection, we first determined the LD50 to i.n. inoculation of the live vaccine strain (LVS) in C57BL/6 (B6) mice using an established method [31,32]. During our preliminary experiments, we found that infected mice lost up to 30% of their initial weight beginning d4 p.i.. Although weight loss was observed at all doses tested (500, 2,000, 8,000 and 30,000 cfu), the degree of disease severity was dependent on the initial inoculum dose. It was only at a dose of 8,000 CFU where we consistently observed moribund mice. At lower doses the majority of mice recovered quickly with few outward signs of disease beyond slightly ruffled fur, which was not consistent between animals. Even at the higher dose, those WT mice that did not succumb by day 12 quickly began to regain weight and appeared otherwise healthy (S1 Fig). Other clinical manifestations included ruffled fur, hunched posture, labored breathing, and reduced mobility. Because we found that weight loss alone was not always an accurate predictor of disease severity, a clinical score based on physical appearance was also included in our endpoint criteria (see Materials and Methods). The resulting LD50 for i.n. LVS infection in B6 mice was ~6,000–8,000 cfu (S1 Fig), which was consistent with that previously published by others [33–35].
To examine the contribution of NKT cells in resistance to primary pneumonic tularemia, we first tested Jα18-/- mice. Hence, B6 and Jα18-/- mice were inoculated i.n. with ~8,000 cfu LVS and monitored for weight loss and signs of morbidity as described in Materials and Methods. When compared to B6, Jα18-/- mice showed a significant increase in susceptibility to i.n. LVS infection (Fig 1A), initially suggesting a protective role for iNKT cells. Although Jα18-/- mice have historically been used as a model for type I NKT cell deficiency, a recent study demonstrated that they also have a profound defect in conventional αβ T cells, with the loss of an estimated 60% of total TCRα repertoire diversity [36]. This report suggests that the increased susceptibility observed in these mice might be due to this more global T cell deficiency rather than the loss of iNKT cells, since conventional T cells were previously shown to mediate protective immunity to LVS [26,34].
We therefore ascertained the outcome of pulmonary LVS infection in CD1d-/- mice, which have a normal complement of αβ T cells but lack NKT cells. By d4 post-inoculation (p.i.), both B6 and CD1d-/- mice began to show signs of disease, including weight loss and ruffled fur. By d5-6, B6 mice continued to lose more weight and showed more severe outward signs of disease as indicated by clinical score (Fig 1B and 1C). In striking contrast, by d7, nearly all NKT deficient CD1d-/- mice began to recover and a significantly lower percentage of CD1d-/- mice succumbed to i.n. LVS infection (Fig 1D). By d14, fewer than 50% B6 and almost all CD1d-/- mice regained weight and showed no outward signs of disease, surviving the infection (Fig 1D). These data suggest that NKT cells serve a deleterious role in pneumonic tularemia. We therefore reasoned that increased iNKT cell number might further exacerbate disease. Thus, Vα14tg mice, which have increased numbers of iNKT cells [37], were infected intranasally and found to have increased susceptibility as predicted (S2 Fig). Taken together, these data strongly support a detrimental function of NKT cells in pneumonic tularemia and show that CD1d-/- mice are better model than Jα18-/- mice to study the role of NKT cells in disease. Hence, CD1d-/- mice were used for all subsequent experiments.
NKT cells are overrepresented among T cells in the healthy lung compared to their frequency in other organs. These tissue-resident NKT cells do not recirculate but rapidly extravasate into interstitial spaces upon recognition of lipid antigens such as the potent CD1d-restricted agonist α-galactosylceramide (αGC) [1,38]. Whether NKT cells behave similarly in response to bacterial infection, where lipid ligands are presumably of lower affinity [4], is unknown. To visualize the location of NKT cells in naïve lungs, we injected B6 mice with anti-CD45 antibody (αCD45 Ab) i.v., allowed it to circulate through blood for several minutes, and tracked the anatomic localization of lung mononuclear cells by flow cytometry after labeling lung leukocytes with Abs against lineage-specific markers. This quick procedure allows circulating αCD45 Ab to label all intravascular cells (αCD45POS), while cells that are in the tissue interstitial space would be αCD45NEG. We confirmed that the largest percentage of lymphocytes (T, B, and NK cells) was in the αCD45POS intravascular population as reported previously [39]. Interestingly, we found that a proportionately higher frequency of pulmonary NKT cells were present in the interstitium (αCD45NEG) when compared to other lymphocytes (Fig 2A and S3 Fig). Such localization suggests that pulmonary NKT cells occupy a frontline niche that allows them to rapidly sense and respond to inhaled antigens.
To test the hypothesis that pulmonary NKT cells rapidly engage in host defense at frontline niches, we inoculated B6 mice i.n. with ~LD50 of LVS. On d3 p.i., we injected αCD45 Ab and, after several minutes, lungs were analyzed by flow cytometry for NKT cell localization. We found a significant increase in the frequency of αCD45NEG interstitial NKT cells that was equivalent to the frequency observed after i.n. αGC treatment (Fig 2B and S3 Fig). Furthermore, beginning d3 after i.n. LVS inoculation, we found a significant decrease in the frequency of NKT cells in the blood when compared to uninfected controls (Fig 2C and S3 Fig). This reduction in circulating NKT cells was coincident with their increased numbers in the lungs (Fig 2C and 2D), more specifically and significantly in the lung interstitium (αCD45NEG lung NKT). Thus, i.n. LVS infection rapidly recruits NKT cells from the vasculature into the infected lung interstitium. This finding is consistent with a previous report showing NKT cell extravasation into the lung interstitium after respiratory αGC administration [1].
NKT cells can be activated by recognition of CD1d-bound microbial lipids or self-lipids presented by activated APCs, or by cytokines which can activate these cells independent of TCR stimulation [4]. The mechanism by which NKT cells are activated can influence their function. To determine whether the accumulating NKT cells were activated by LVS infection, we monitored expression of the activation marker CD69 which reports on NKT activation in general, and we employed Nur77gfp reporter mice, which increase GFP expression only upon NKT cell activation via CD1d-restricted microbial glycolipids but not self-lipids [40]. On d3 post i.n. LVS inoculation, we found that both interstitial and intravascular NKT cells increased CD69 expression (Fig 3A). Strikingly, however, only the interstitial population showed increased GFP fluorescence (Fig 3A). This result indicated that local LVS antigen presentation occurred within the interstitium to activate LVS-recruited NKT cells in situ.
To determine the consequence of interstitial NKT cell activation, intracellular cytokine staining was conducted directly ex vivo without any restimulation. We found that the GFPhi interstitial NKT cells but not GFPlo intravascular cells produced IFN-γ (Fig 3A and 3B), demonstrating that this cytokine production was a result of LVS infection in vivo. Taken together, these data suggest that upon i.n. F. tularensis LVS infection, NKT cells migrate from the periphery and accumulate within the infected lung where they are activated through their TCR to produce IFN-γ.
Two factors can affect overall fitness after infection: pathogen burden and the level of organ pathology. To determine whether the decreased susceptibility of CD1d-/- mice to LVS was due to differences in bacterial burden, we measured LVS burden in lung homogenates at various time points post i.n. LVS inoculation. Lung burden was similar in both groups, with a modest but statistically significant difference (two-fold) seen only at d7 and d9 in infected CD1d-/- mice (Fig 4A), which was concomitant with the observed differences in clinical score and weight loss (Fig 1B and 1C).
Since the observed differences in illness began to appear at d7 p.i., we focused on this time point for subsequent analyses. Differing degrees of lung pathology can result in different animal fitness despite similar bacterial burden [35,41,42]. To determine whether the difference in morbidity was due to differential lung damage, we measured blood oxygen saturation (SpO2), which was suggested as an accurate measure of lung pathology [43]. While both groups of mice showed decreased SpO2 levels, there were no significant differences observed between groups (Fig 4B). To more directly assess tissue damage, H&E-stained lungs were evaluated at d7 and d9 p.i. As predicted by pulse oximetry data, the findings in the lungs of B6 and CD1d-/- mice were similar. Lungs from both groups of mice displayed focally extensive interstitial pneumonia with perivascular cuffs of mononuclear cells. Inflammatory infiltrate consisting of macrophages and neutrophils filled the alveoli, which also contained necrotic debris and edema with focal areas of necrosis on the alveolar walls (Fig 4C).
After i.n. infection, F. tularensis rapidly disseminates to the periphery [44]. The kinetics and extent of dissemination are suggested as determinants of disease severity [45–48]. Hence, we measured burden in blood, liver, and spleen. LVS was only transiently detectable in the blood, where levels peaked at d3 p.i., but there were no differences in bacteremia between groups (Fig 5A). Liver burden was similar in both groups, but CD1d-/- mice had significantly lower splenic burden d3–7 p.i. (Fig 5A). Further analysis showed that only lung burden—but not liver or spleen—correlated with weight loss at d7 p.i. (S4 Fig). Consistent with these findings, histopathological analysis failed to identify any striking differences in either liver or spleen pathology after intranasal inoculation (Fig 5B). The extent of hepatic granuloma formation did not differ between groups (Fig 5B). Contrary to previous reports in BALB/c mice [49,50], splenic architecture was mostly intact with some evidence of apoptosis and extramedullary hematopoiesis that did not seem to differ between groups (Fig 5B).
In summary, the above data indicate that the modest differences in bacterial burden in the lungs and peripheral organs, or another mechanism, but not differential pulmonary, hepatic, or splenic histopathology could explain the milder disease observed in NKT cell deficient mice.
The differences described thus far become most pronounced at d7 p.i, suggesting that the quality of the adaptive response may be the principal underlying cause of the reduced susceptibility observed in CD1d-/- mice. We therefore analyzed lymphocyte numbers in the lungs after LVS infection. Lungs of CD1d-/- mice had consistently higher numbers of both B and T lymphocytes present at d7 p.i. (Fig 6A). To directly visualize the localization of these cells—whether the lymphocytic infiltration was in the pulmonary vasculature or had extravasated into the lung tissue—fixed frozen sections of infected lung were stained and examined by confocal microscopy. Consistent with earlier analysis of H&E stained sections, both groups of mice showed extensive perivascular and peribronchiolar infiltration of leukocytes. Significantly however, in addition to an increased cellularity in CD1d-/- mice, there was a striking difference in the degree of organization of the infiltrating immune cells. We found evidence for the formation of tertiary lymphoid structures, iBALT, within the infected lungs of CD1d-/- mice (Fig 6B). These structures were marked by the formation of B cell follicles, which were surrounded by T cell congregates. In contrast, fewer iBALT structures were formed in B6 mice by d7 p.i. (Fig 6C) and those that developed were small and appeared rudimentary in that T and B cells were scattered or at best loosely packed together (S5 Fig). Such tertiary lymphoid structures were not observed in uninfected CD1d-/- or B6 mouse lungs (S6 Fig) suggesting that iBALTs formed in response to pulmonary LVS infection.
Dendritic cells (DCs) are both necessary and sufficient for the induction of iBALT within the lungs [51]. Consistent with previous reports, concentrations of CD11c+ cells were also observed in and around the B and T cell zones of iBALT formed in CD1d-/- lungs (Fig 6B). Further characterization of CD11c+ cells within the infected lungs by flow cytometry revealed that all DC subsets identified were increased in the LVS infected lungs of CD1d-/- mice (Fig 6D). Considering that DCs have been implicated as a primary vehicle for dissemination of F. tularensis from the infected lung [45], this finding may also partially explain the reduced splenic burden in CD1d-/- mice. Hence, iBALT formation in response to LVS infection of the lungs is associated with milder tularemia-like disease in mice and prognosticates recovery.
A previous study showed that increased disease severity in a murine tularemia model was associated with hepatic damage and neutrophilia [52]. NKT cells activated by LVS could cause liver damage through direct lysis of infected hepatocytes [53,54]. They might also promote neutrophilia through production of granulocyte colony-stimulating factor (G-CSF), the major neutrophil survival and proliferation factor [55]. Thus we monitored these two disease parameters in CD1d-/- mice. Although histological analysis failed to reveal any gross differences in liver pathology (Fig 5B), we measured AST and ALT in the serum of infected animals as a more sensitive indicator of hepatic injury (Fig 7A). CD1d-/- mice had serum ALT levels that were slightly, yet significantly, lower than those measured in B6 mice, which implied slightly more severe liver damage in the latter strain. However, when taken together with the absence of widespread necrosis, these mild-to-moderate elevations in circulating hepatic enzymes are not consistent with death from liver failure, particularly considering the similar liver burden in the two groups of mice (Fig 5). Such slightly elevated ALT levels need not be a result of the death of infected hepatocytes, but might rather be indicative of increased systemic inflammation [56].
To monitor the numbers of circulating neutrophils we performed complete blood counts (CBC) at various time points post inoculation. B6 mice did indeed show more pronounced neutrophilia d7 p.i. when compared to CD1d-/- mice (Fig 7B). This difference was not observed in naïve mice and did not appear until d7 p.i. (Fig 7C). We found lower G-CSF levels in the serum but not the lungs of CD1d-/- mice by d7 (Fig 7D), which was consistent with the lower frequency of neutrophils in the blood (Fig 7B). Thus, decreased neutrophilia incited by LVS infection in CD1d-/- mice could be one cause for their increased resistance to tularemia.
Although IFN-γ is necessary for resistance to LVS, excessive production of proinflammatory cytokines can be detrimental, particularly after intranasal infection [24,41]. Because NKT cells are known to induce IFN-γ production by NK cells [57,58], we therefore asked whether CD1d-/- mice—which have normal numbers of NK cells—might produce less IFN-γ in response to LVS. We found that although the early IFN-γ response was comparable in both groups of mice (S7 Fig), by d7 p.i. CD1d-/- mice had significantly lower levels in both the serum and lung (Fig 8A and 8B).
Many cytokines produced by NKT cells were suggested to promote sepsis-like inflammatory disease [11,59] and, hence, we measured the levels of those cytokines previously shown to be induced by LVS infection that have also been associated with severe sepsis [49,60]. We found that CD1d-/- mice had consistently lower levels of IFN-γ, IL-6, and TNF-α in both the lungs and serum, with lower levels of MCP-1 and KC in the serum at d7 p.i. (Fig 8A and 8B) which coincided with modestly reduced burden, reduced weight loss, and less severe outward signs of disease (Fig 1). Hence, NKT deficiency results in a less severe “cytokine storm” in response to i.n. LVS infection.
Given the observed differences in susceptibility among various mouse strains [61–64], we next ascertained whether CD1d-/- mice in the BALB/c background would exhibit the same phenotype. Consistent with the results in C57BL6 background, we found that CD1d-/- mice were less susceptible to LVS infection in the BALB/c background as well (Fig 9A). As observed with C57BL6 mice, higher doses caused severe disease in both groups, although death was slightly delayed in CD1d-deficient mice.
Mice are highly susceptible to several subspecies of Francisella and the extreme virulence of type A (subspecies tularensis) or B (subspecies holarctica) strains complicates the study of effective immune responses, particularly when administered intranasally (LD50<10) [26]. Although LVS has long been used in murine models of experimental tularemia [31,65], it is an attenuated type B strain that fails to cause disease in humans. We therefore sought to determine whether CD1d-/- would have a similar resistance to the more virulent type A strain Schu S4. Not surprisingly however, both wild-type and CD1d-/- mice were extremely susceptible to a low-dose (<10 CFU) pulmonary infection with Schu S4 (Fig 9B), similar to the results obtained with higher doses of LVS in both C57BL6 (S1 Fig) and BALB/c (Fig 9A) mice. These data are consistent with previous reports showing reduced efficacy of therapies or mutations that confer resistance to LVS when challenged with Schu S4 [35,66–68].
Once infected, the host has two strategies to deal with invasive pathogens: resistance and tolerance. Resistance implies the ability of the host to clear a pathogen or limit its spread whilst tolerance indicates the ability of the host to bear the ensuing pathology [69]. Surviving infection from a pathogen such as F. tularensis likely requires a balance of both. Previous studies of infection in immunodeficient mice revealed that immunity to F. tularensis is mediated by the concerted effort of the innate and adaptive humoral and cellular immune responses. Because the host deploys a wide array of effector mediators in its inflammatory arsenal against the infection a sepsis-like disease ensues [22,26]. Hence, fatalities in this infectious disease appear to be caused by an inability to control bacterial growth and dissemination, which precipitates inflammation and causes irreparable collateral tissue damage. Accordingly, if diagnosed early, antibiotic treatment limits bacterial burden and tempers inflammation thereby controlling morbidity and mortality underlying tularemia. Herein we report that NKT cells respond to intranasal infection by migrating to the infected lung to produce IFN-γ, which promotes clearance and killing of the pathogen. As a trade-off however, the early activation of NKT cells causes excessive systemic inflammation, increases neutrophil mobilization, and delays infiltration of lymphocytes into the lungs and formation of protective tertiary lymphoid structures. Hence, CD1d-deficient animals have a tempered inflammatory response and perform clinically better in response to intranasal LVS infection. This advantage, however, can be overcome with higher doses of bacteria or infection with more virulent subspecies. The discordant results obtained using Jα18-/- and CD1d-/- mouse models likely reflect a more global T cell deficiency in Jα18-/- mice and underscore the difficulties in deciphering the role of NKT cells using currently available methods [14].
Akin to CD1d-deficient animals, F. tularensis infection of metalloproteinase-9 null as well as IL-17 and IL-10 double-deficient mice showed low disease incidence despite similar bacterial burden [35,41]. In conjunction with a tempered inflammatory response, these immunodeficient mice were able to limit the damage caused by the pathogen or the host immune system and were therefore less susceptible to i.n. F. tularensis infection. These findings and our data reported herein, thus enforce the notion that the host response to F. tularensis is a balance between resistance to pathogen and tolerance to pathology.
LVS infection of endothelial cells induces the transmigration of neutrophils into infected tissues [70,71]. Rather than contributing to host defense, neutrophils may instead increase disease severity. F. tularensis-infected neutrophils assume an extended proinflammatory phenotype, which is thought to contribute to increased tissue destruction [72]. Because F. tularensis is able to infect and replicate within neutrophils, these cells likely serve as a vehicle for dissemination from the primary site of infection, which might contribute to the reduced peripheral burden observed in CD1d-/- mice [15,73]. Direct interaction with NKT cells has been shown to influence the function of neutrophils [74] and to promote hematopoiesis in both humans and mice [75]. NKT cell-neutrophil interactions were also recently shown to exacerbate polymicrobial sepsis [59]. Although we found no differences in differential cell counts from peripheral blood of naive wild-type and CD1d-/- mice, CD1d-/- mice displayed less severe neutrophilia upon LVS infection. We also found that slightly decreased hepatotoxicity, reduced peripheral burden, and less severe inflammation in NKT cell-deficient CD1d-/- mice corresponded with less severe neutrophilia.
Our studies revealed that acute LVS infection of CD1d-/- mice resulted in increased lymphocytic infiltration in the lung interstitium when compared to B6 mice. The cellular infiltration organized into iBALT at the peak of infection. Although typically associated with chronic inflammatory conditions, iBALT has been shown to form in the lung in response to viral infection [51]. However, the formation of such structures in response to acute bacterial infection is less well-studied. In the context of pulmonary F. tularensis infection, lymphoid aggregates have been described in wild-type mice that survived F. tularensis infection after >50d p.i. whilst bone fide iBALTs were described only in LPS/rPorB vaccinated mice challenged with a high dose of LVS [49,76]. The post-vaccination development of these structures in the lungs was associated with better survival upon i.n. challenge with LVS or the more virulent F. tularensis subspp. tularensis [76,77]. Data from these studies suggest that the T and B cells found in iBALT limit the spread of the bacteria from the lung thereby reducing immune-mediated damage to peripheral organs. Contrary to the results of these studies with wild-type animals, our data demonstrate the development of iBALT in the lungs of B6 mice is hampered perhaps by the presence and activation of NKT cells as more robust iBALT formation is observed in CD1d-/- mice during the peak of a primary LVS infection. Such early iBALT formation is suggestive of a more vigorous and effective adaptive immune response resulting in reduced systemic inflammation and therefore less severe disease. These findings suggest a suppressive/regulatory role for pulmonary NKT cells. The regulatory function of NKT cells is just beginning to be elucidated and much of what we know derives from studies utilizing model antigens [7]. Thus, our findings provide a model in which to study the potentially detrimental functions of iNKT cells in a natural infection.
In the context of intranasal LVS infection, the inability to form iBALT proves to be detrimental, but there may be cases where this outcome is desirable. For example, iBALT formation has been associated with chronic inflammatory conditions such as asthma [78]. Therefore, the discovery of an NKT cell-activating ligand(s) derived from F. tularensis that can prevent iBALT formation might be of therapeutic benefit in such instances. Similarly, a better understanding of the regulatory function of NKT cells may lead to advances toward the goals of rational vaccine design.
All procedures on mice were performed by approval from the IACUC at Vanderbilt University School of Medicine and at Albany Medical College. Anesthesia was performed using 1–5% isoflurane or 100 mg/kg and 10 mg/kg ketamine and xylazine, respectively. Euthanasia was performed by CO2 overdose followed by cervical dislocation.
C57BL/6J mice were purchased from the Jackson Laboratory (Bar Harbor). CD1d-/- mice were a gift from L Van Kaer (Vanderbilt University School of Medicine). Jα18-/- mice were generously provided by M Taniguchi (RIKEN Institute, Yokohama, Japan). Mice were bred and maintained in the School of Medicine vivarium and provided with food and water ad libitum. LVS infection was performed in an ABSL-2 facility. Vanderbilt’s IACUC approved the experiments described here.
F. tularensis LVS was provided by S. Khader (Washington University, St. Louis, MO). Preparation of working stocks and CFU determination from infected tissue were performed as described [31]. LD50 was determined by the method of Reed and Muench [32]. Male, 8–12-week old mice were anesthetized by i.p. administration of a ketamine/xylazine mixture and ~8–10x103 CFU LVS were administered i.n. in 50μL sterile PBS. Mice were monitored daily for weight loss and signs of morbidity. Criteria for clinical score were developed based upon observation of mice from at least three separate experiments: 1) no outward signs of illness; 2) consistently ruffled fur; 3) hunched back and altered gait; 4) reduced mobility/reaction to stimulus, labored breathing, lethargy. Mice were humanely euthanized when weight loss exceeded 30 percent.
Experiments with Schu S4 were conducted with approval of the Albany Medical College IACUC (protocols 15–04001 and 15–04002). Six-to-eight week old BALB/c and C.129-CD1d-/- mice were inoculated i.n. with the indicated dose of Schu S4 as described above for LVS.
Cytokines were measured in serum or lung homogenates using Cytokine Bead Array assay (BD Biosciences). Right lobes of lung were homogenized using a Tissue Tearor (Biospec Products, Inc.) in 1 mL sterile PBS containing a cocktail of protease inhibitors (Roche). Homogenates were centrifuged and supernatants frozen at -80°C until analyzed.
SpO2 was measured using the PhysioSuite with MouseSTAT Pulse Oximeter (Kent Scientific Corp.). Mice were anesthetized using 1.5% isoflurane for induction and maintenance. Sensor was attached to the right hind foot according to the manufacturer’s instructions. Measurements were recorded at 5-second intervals for at least 4 minutes and averaged.
Spleen, liver, and lungs were processed as previously described [1,79]. Left lung lobes were used for flow cytometry and microscopy and right lobes for CFU determination.
NKT cells were analyzed as described [79]. All data were acquired on an LSR II (BD Biosciences) and analyzed with FlowJo software (FlowJo, LLC). Cell populations were identified as follows: B cells (CD19+); T cells (CD3ε+); NK cells (CD3ε-NK1.1+); NKT cells (CD3ε+CD1d/αGC tetramer+); DCs (CD45+CD64-CD24+MHCII+CD11c+). DC subsets were identified as described [80]. CD1d monomers were provided by the NIH Tetramer Core Facility. Cell counts were determined using AccuCheck counting beads (Invitrogen).
To analyze the formation of tertiary lymphoid follicles in the lungs following LVS infection, tissues were fixed in PLP buffer (2% paraformaldehyde, 0.05 M phosphate buffer containing 0.1 M l-lysine, and 2 mg/ml NaIO4) overnight. Tissues were then dehydrated in 30% sucrose and subsequently embedded in OCT media. Twenty micron frozen sections were preincubated with Fc-block (anti-mouse CD16/32 Ab; Biolegend) diluted in PBS containing 2% goat serum and fetal bovine serum (FBS). After incubation for 1 hour at room temperature, sections were washed with PBS and stained with the following antibodies diluted in PBS containing 2% goat serum and FBS for 1 hour at room temperature: anti-B220-Alexa-488 (Biolegend), anti-CD3-allophycocyanin, anti-CD11c-phycoerythrin (Biolegend) and Fc-block. Sections were washed with PBS and mounted using Immu-Mount (Thermo Fisher Scientific). Images were acquired using a Zeiss 780 confocal microscope (Carl Zeiss). The imaging data were processed and analyzed using the Imaris software (Bitplane; Oxford Instruments).
Statistical analyses of representative data were performed using GraphPad Prism version 5.02 for Windows, GraphPad Software, San Diego California USA (www.graphpad.com). Where indicated in the figure legends, data were aggregated across several (from two to four) replicate experiments. To address the clustered nature of the final datasets, we used a linear regression analysis with cluster robust standard error estimation (an extension of Huber-White Sandwich Estimator [81–83]), a method that accounts for intraclass correlation when determining statistical significance of regression coefficients. In all reported regression analyses individual data points were entered as the dependent variable, and replicate experiments were identified as the clustering variable. Experimental group type (KO and WT) was entered as the independent variable and was dummy coded for the analyses with “KO” group as the reference category. In such case, the intercept reflects the mean of the reference group (= KO), and the slope reflects the difference between the KO and the WT groups. We have reported herein the significance of coefficients for slopes (differences between group means). Clinical score data was analyzed using Generalized Estimating Equation (GEE) model based on Poisson regression. In the model, clinical score was predicted by group type and day of assessment (dummy coded), and their interaction. We report comparisons between groups at each time of assessment, which were obtained by changing the reference category between days of assessment. All analyses were carried out using R Project for Statistical Computing (http://www.r-project.org/).
|
10.1371/journal.pcbi.1006820 | Modeling and MEG evidence of early consonance processing in auditory cortex | Pitch is a fundamental attribute of auditory perception. The interaction of concurrent pitches gives rise to a sensation that can be characterized by its degree of consonance or dissonance. In this work, we propose that human auditory cortex (AC) processes pitch and consonance through a common neural network mechanism operating at early cortical levels. First, we developed a new model of neural ensembles incorporating realistic neuronal and synaptic parameters to assess pitch processing mechanisms at early stages of AC. Next, we designed a magnetoencephalography (MEG) experiment to measure the neuromagnetic activity evoked by dyads with varying degrees of consonance or dissonance. MEG results show that dissonant dyads evoke a pitch onset response (POR) with a latency up to 36 ms longer than consonant dyads. Additionally, we used the model to predict the processing time of concurrent pitches; here, consonant pitch combinations were decoded faster than dissonant combinations, in line with the experimental observations. Specifically, we found a striking match between the predicted and the observed latency of the POR as elicited by the dyads. These novel results suggest that consonance processing starts early in human auditory cortex and may share the network mechanisms that are responsible for (single) pitch processing.
| In this work, we argue that human auditory cortex processes pitch and consonance by means of a common neural network mechanism operating at early cortical stages. We introduce a neural population model of cortical pitch processing that contains biophysically realistic synaptic and neural parameters. The model quantitatively explains, for the first time, the neuromagnetic responses observed in human auditory cortex during pitch perception. The model is subsequently used to elucidate the cortical processing of musical dyads, in which concurrent pitches lead to the perception of consonance or dissonance. Interestingly, the model predicts that sounds perceived as dissonant need more time for cortical processing than consonant sounds. This prediction is experimentally validated by recording cortical neuromagnetic fields in response to consonant and dissonant dyads. Taken together, our results suggest a novel mechanistic explanation for early cortical processing of musical harmony, in the sense that the differential response to consonance and dissonance starts early, simultaneously to (single) pitch processing, in auditory cortex.
| Pitch is the perceptual correlate of the periodicity in a sound’s waveform, and thus a fundamental attribute of auditory sensation. It forms the basis of both music and speech perception. However, understanding pitch processing as elicited by concurrent sounds in human auditory cortex is still a major challenge in auditory neuroscience [1–5].
A combination of two sounds that simultaneously elicits two different pitches is called a dyad, and the pitch interactions within the dyad give rise to a sensation that can be characterized by its consonance or dissonance. Loudness, timbre, and the fundamental periodicities of the two sounds can have subtle effects on whether a dyad is perceived as consonant or dissonant. However, the dominant factor in determining the degree of a dyad’s consonance is the relationship between the fundamental periods of the sounds that make up the dyad: simple periodicity ratios result in more consonant sensations. In contrast, the sensation becomes more and more dissonant as the complexity of the periodicity ratio increases [6, 7]. It has been previously proposed that dissonance correlates with the beating or roughness sensation that is elicited by the interfering regularities of the dyad components [6, 7]. However, listeners who showed impaired pitch perception but were sensitive to beating and roughness were unable to differentiate between consonant and dissonant dyads [1, 8]. This suggests that pitch- rather than roughness-related auditory processing is responsible for consonance perception.
Neurophysiological evidence for a close link between consonance and pitch has recently been provided by Bidelman and colleagues [2]. Their study showed, using electroencephalography (EEG), that the amplitude of the cortical pitch onset response (POR) is strongly modulated by a dyad’s perceived consonance. The POR is a pitch-selective component of the transient auditory evoked potential/field (AEP/AEF) that occurs within the time range of the well-known N100 deflection, around 100 ms after pitch onset [9]. The morphology of the POR is strongly correlated with the perceived pitch in single tones: its latency scales linearly with the period of the sound and its amplitude increases with the strength of the pitch percept [9, 10]. The neural sources of the POR are located in the anterolateral section of Heschl’s gyrus (alHG) in auditory cortex [2, 9], in agreement with the anatomical location of pitch-selective neurons in non-human primates (e.g., [11–13]), and with pitch-selective regions that were reported for human listeners [14–18].
Further experiments in human subjects demonstrated that the dyad-evoked frequency-following response in the brainstem is predictive for the perceived consonance of a dyad (for a review, see [19]). However, functional magnetic resonance imaging (fMRI) studies showing selective activation to consonance/dissonance contrasts in the superior temporal gyrus [20] and in frontal cortex [21] led the auditory community to link neural representations of consonance and dissonance with higher cognitive processes [22].
In this study, we used a combined experimental and theoretical approach to assess whether consonance and pitch share similar processing mechanisms in human auditory cortex. Towards this goal, we first developed an ensemble model of cortical pitch responses, specifically designed to understand the mesoscopic representation of pitch in alHG. The model can account, mechanistically, for the POR latency effects that have been reliably reported in numerous experimental settings [9] but remained poorly understood. Second, we recorded the AEF elicited by consonant and dissonant dyads using magnetoencephalography (MEG). Our experimental results revealed a strong correlation between the POR latency and the degree of consonance, extending previous EEG findings [2]. Finally, we aimed to replicate the results from the MEG experiment using our model. If the hypothesis that consonance and pitch are processed by similar mechanisms in cortex is correct, we would expect the model to explain the dependence of POR latency on the degree of consonance without the inclusion of higher processing stages within the auditory hierarchy [20, 21]. In line with this hypothesis, the model provides a quantitative explanation for the relationship between the POR dynamics and consonance, suggesting that consonance and dissonance perception might be linked to pitch processing regions in auditory cortex, prior to higher-order processing.
Next, we recorded neuromagnetic fields evoked by six different dyads from 37 normal hearing subjects. Data were preprocessed using standard MEG procedures and equivalent current dipoles were fitted for the POR, independently in each subject and hemisphere, and pooled over conditions (see Methods). Dipole locations in Talairach space are plotted in Fig 4B.
Dyads consisted of two IRN sounds. The lower note pitch was 160 Hz; the pitch of the upper note was adjusted accordingly to form either a consonant dyad (unison, P1; perfect fifth, P5; major third, M3) or a dissonant dyad (tritone, TT; minor seventh, m7; minor second, m2). To dissociate the energy onset response in planum temporale from the POR in alHG, the dyads were preceded by an energy-balanced noise segment, cross-faded with the dyad to avoid discontinuous waveforms (like for the single IRN sounds analyzed in the previous section; see Methods).
Fig 4A presents the MEG grand-mean source waveforms, for both hemispheres, in response to the six stimulus conditions. The noise onset from silence (depicted in grey below the source waveforms) was followed by a transient P1m-N1m-P2m AEF complex. Since the first stimulus segment did not vary between conditions, we did not expect to find any significant differences in the corresponding neuromagnetic activity at this point.
In contrast, the transition to the second stimulus segment (IRN dyads, black signal below the source waveforms) elicited prominent POR waves and the morphology of the POR varied considerably between conditions. Fig 4C shows close-up views of the POR. Consonant dyads (pooled conditions [P1+P5+M3]) elicited a much earlier (p <.0001) and larger (p <.0001) POR than dissonant dyads (pooled conditions [m7+TT+m2]). Fig 4D depicts 99% bootstrap confidence intervals for the POR amplitudes and latencies pooled over hemispheres in response to the experimental conditions; the activity pattern observed here also points to a close relationship between the degree of a dyad’s consonance and the morphology of the respective POR.
When pooling across conditions, we found a difference between the left and the right hemisphere in the POR amplitude (p = .01), but not in the POR latency (p = .36); also, the difference between the neuromagnetic responses to consonant or dissonant dyads did not significantly vary between hemispheres (latency: p = .58, amplitude: p = .48).
The POR latency difference in response to consonant and dissonant dyads in alHG suggests that consonance and dissonance are computed at relatively early stages of the cortical auditory hierarchy. We used our model of cortical pitch processing, designed to reproduce the neuromagnetic responses elicited by iterated rippled noises, to test this interpretation. If the differential responses to consonance and dissonance in alHG were intrinsic to pitch processing, we would expect our model to be able to reproduce this behavior.
First, we verified that the model was able to provide a joint representation of the two pitches comprised in the dyads; the results are shown in Fig 5A–5C. As in Fig 3, the plots show the average activation of the different ensembles (x-axis) for the 13 dyads in the chromatic scale (y-axis). The vertical line in Fig 5C indicates the first note common to all dyads; the diagonal the neural representations of the second notes. It should be emphasized that even phenomenological (biologically inspired but with low realism) models of pitch perception are generally unable to decode sounds with concurrent pitches (e.g., [45, 46]; see [47] for a review).
Fig 5J shows the latency predictions of the model and the experimental data for the respective dyads (see also Fig 5D–5I). Note that this is a genuine out-of-sample test, since model parameters previously fitted (see Methods) were held fixed to account for this new data.
Although the model predicted a slightly shorter POR latency for the semitone (m2) dyad than observed (see Discussion), latency predictions match the experimental trend; moreover, the differential response to consonant (P1, M3, P5) and dissonant (m2, TT, m7) dyads found in the MEG data was accurately replicated by the model (latency of P1 and P5 < latency of dissonant dyads: p < 10−7, W > 5050; latency of M3 < latency of m2: p = .00002, W = 4414; latency of M3 < latency of TT: p = .38, W = 3688; latency of M3 < latency of m7: p = .096, W = 3878; one-tailed Wilcoxon rank-sum tests performed over the results of N = 60 runs of the model). The full temporal dynamics of the dipole moment predicted by the model is shown in Fig 5D–5I.
Last, we extended the POR latency predictions to all 13 dyads comprising the entire chromatic scale (see Fig 5K), and tested if the differential responses to consonance and dissonance were generalizable to additional dyads. Following Helmholtz [6], we considered an extended set of consonant dyads, including the octave (P8) and the perfect fourth (P4); and an extended set of dissonant dyads, including the major seventh (M7) and the major second (M2). Once again, consonant dyads produced shorter latencies than dissonant dyads (latencies of P1, P4, P5 and P8 < latencies of the extended set of dissonant dyads: p <.0003, W > 4445; latency of M3 < latency of M2: p = .00002, W = 4420; latency of M3 < latency of M7: p = .75, W = 3501; one-tailed Wilcoxon rank-sum tests, N = 60). These results, fully in line with our previous findings, suggest that the differential response of our model to consonance and dissonance is a consequence of the harmonic relationships between the periodicities of the two dyad components. These analyses are extended to further families of dyads in S9 Fig, yielding similar results.
The model can also be used to explain how interactions between the components of the dyads influence processing time: consonant dyads consist of tones that share a larger number of lower harmonics than the ones in dissonant dyads. For instance, in the just intonation, the perfect fifth of a given fundamental shares one in every two harmonics with that fundamental, whilst only one in 16 harmonics are shared by a minor second and its fundamental. The proposed mechanism is based on the idea that cortical pitch processing is triggered by the joint activation of, at least, three periodicity detectors characterizing a specific harmonic series. Consonant dyads elicit a dramatically larger signal-to-noise ratio in the periodicity detectors tuned to their common harmonics, resulting in a collaborative effort towards pitch extraction that effectively speeds up processing dynamics (see video S2 Video for an animation depicting the full process).
This work combines new theoretical and experimental findings to elucidate how human auditory cortex processes pitch and consonance/dissonance through similar network mechanisms.
First, we introduced a novel ensemble model designed to reproduce the neuromagnetic fields elicited in alHG during pitch processing. The model was used to understand the POR morphology and the dependence of its peak latency on the perceived pitch, a phenomenon that although robustly observed for over two decades [9], has remained poorly understood. Second, we designed an MEG protocol to investigate whether the POR properties are influenced by the degree of consonance or dissonance, as elicited by different dyads common in Western music. Our results revealed a strong correlation between the POR peak latency and the degree of consonance elicited by each dyad, extending previous EEG results that also reported a modulation of the POR amplitude by consonance and dissonance [2].
Third, we showed that our model (originally designed to explain pitch processing in IRN stimuli with a single pitch) quantitatively replicates the correlation between POR latency and the degree of consonance and dissonance. We provide a mechanistic explanation for the shorter POR latencies in response to consonant dyads as an effect of the harmonic facilitation during pitch processing. Combined, our results indicate that the neural mechanisms accounting for pitch processing show differential responses to consonant and dissonant dyads, showing that the sensation of consonance may be initially elicited as a result of pitch coding in alHG before subsequent cognitive processing.
A new systematic interpretation of the POR latency can be deduced from the dynamics of the decoder network: the POR might reflect the amount of time that is necessary for the network to robustly stabilize in a state representing a unequivocal pitch (see Fig 2). Although an association between POR latency and processing time has been previously hypothesized in experiments (such as in [9]) and in a phenomenological model [45], a detailed understanding of this mechanism was still lacking. In our model, the latency of the POR coincides with the instant in which the net inhibition at the decoder network overtakes the excitatory activity from the periodicity detectors. From a dynamic systems perspective, this is equivalent to the instant in which the trajectory in the phase-space is unequivocally directed towards the attractor state dominated by the neural ensemble that is characterized by the perceived pitch (see the phase space portrait in S1 Video).
The model only performs a robust perceptual decision concerning stimulus pitch after the cortical system identifies three peaks from the harmonic series of the stimulus period in the representation of the periodicity detectors. This accounts for the relation between the POR latency and the stimulus period [9]. In addition, this also explains why pitch identification is only robust when the stimulus duration exceeds four times the pitch periodicity [9]. Although previous studies had postulated that cortical pitch processing mechanisms must integrate along several period cycles in order to make a perceptual decision [9], a specific mechanism for such an integration has not been proposed to the date.
Moreover, since phase-locked activity is not robustly present above 50–200 Hz in the cortex [15], integration along several repetition cycles is only possible in subcortical areas. The decoder network in our model takes advantage of the input harmonic representations provided by an autocorrelation model that does not require phase-locking to transmit information concerning several repetition cycles [28], and thus provides a parsimonious solution to this problem.
Combined, our results suggest that cortical processing of dissonant dyads is slower than the processing of consonant dyads; i.e., it requires a longer processing time. The model constitutes a physical rationale for this phenomenon: cortical extraction of consonance is based on the accumulation of activity in the columns with preferred periods characterizing the lower harmonics of the target sound; thus, concurrent pitch frequencies sharing common lower harmonics contribute to the build-up of each other’s representation, thereby speeding up the stabilization of the network. Since consonant dyads are characterized by simpler frequency ratios, their components share a larger number of lower harmonics than the components of dissonant dyads, and hence this stabilization is promoted.
Early phenomenological models based on Helmholtz’s roughness theory described dissonance as the beating sensation produced by tones with fundamental periodicities that were not harmonically related [6, 7]. More recent explanations of consonance, based on pitch processing, have linked the regularity of the autocorrelation harmonic patterns elicited by dyads to their evoked consonance and dissonance percepts [1, 4, 19]. Thus, previous phenomenological consonance models have consistently described the degree of consonance as the perceptual correlate of the degree of overlap between the dyad components’ lower harmonics. Our model introduces a potential explanation for the biophysical rationale underlying this description.
Although our modeling results generally show a good fit with the data from the MEG experiment, the model prediction falls around 5 ms short when explaining the POR latency evoked by the minor second dyad. This underestimation might result from the limited number of harmonics considered during the integration step in the decoder network: dissonant dyads, whose components do not share any common harmonic within the first three peaks of their harmonic series, present comparable processing times. More accurate results would occur if an adaptive mechanism adjusted the number of harmonics required to trigger the decoding process according to the degree of peak overlap in the input. This adaptive mechanism would be necessary to explain how humans can differentiate dyads that differ in a quarter of a semitone.
Our study did not assess whether the general (yet not universal [3, 5]) association between consonance and pleasantness might be a consequence of the differential responses of the decoder network to consonant and dissonant dyads. Future work should investigate whether this link could be better explained by processes at higher levels of the auditory hierarchy that might be more sensitive to cultural and background modulations.
Our neuromagnetic findings concerning the POR morphology in response to consonant and dissonant dyads resemble and extend recent EEG data reported by Bidelman and Grall [2], and by Proverbio et al. [48]. Specifically, Bidelman and Grall [2] applied EEG in a smaller sample (N = 9) of musically trained listeners and revealed a close relation between their subject’s consonance/dissonance ratings and the morphology of the POR that was elicited by the respective dyads in alHG. In their study, the POR latency difference between consonant and dissonant dyads (cf. their Fig 4b) appears to have a non-significant (p = .22) effect size, smaller than the results that were obtained in our study by means of MEG.
One reason for this might be that Bidelman and Grall [2] applied shorter IRN stimuli with a higher number of iterations, resulting in an increased saliency of the pitch percept; moreover, they employed a dichotic stimulation paradigm in which each ear was presented with only one dyad component, whereas in our experiment sounds were delivered diotically to the listeners. Since our model does not predict an effect of the number of iterations of the IRNs on latency (predictions for dyads of 32 iterations IRNs are shown in S9D Fig), we speculate that the diotic presentation of the dyads is responsible for the stronger effect shown in our data. This hypothesis cannot be explored by our current model because it does not consider binaural integration. The modeling of this process could be informed by the divergence between Bidelman and Grall’s and the present results in future work.
In line with results from previous experiments (e.g., [9, 11, 30, 32, 49]), our findings are consistent with the notion that lateral HG acts as a cortical pitch center. Many of these earlier studies employed IRN stimuli; however, a number of fMRI experiments (e.g., [50, 51]) have argued that the activity observed in HG might be confounded by slow fluctuations in the IRN spectrum. Indeed, pitch-sensitive cell ensembles seem to overlap, within HG, with other neural populations that are more sensitive to other (e.g., spectral) sound features [52, 53]. However, this does not speak against the existence of a pitch-specialized subregion in HG since overlapping neuron ensembles are difficult to disentangle by means of fMRI [54]. Thus, based on the relatively homogeneous pattern of results in the current study and in experiments using different stimuli and neuroimaging methods [2, 18, 31, 35], the existence of a pitch center in HG might be viewed as highly probable.
Numerous phenomenological models have been designed to predict pitch for a wide range of complex sounds (e.g., [24, 27, 28, 45, 46, 55], see [47] for a review). These models can account for a variety of perceptual phenomena [24]. The weaker pitch of frequency-transposed harmonic complex sounds [56], for example, was explained using nonlinear filters to simulate the compression taking place in the basilar membrane [24]. Although the present work constitutes a first effort towards a mechanistic model of early pitch and consonance processing, future efforts should focus on broadening the model scope to pitch phenomena not addressed in the current model implementation.
The correlation between pitch and cortical AEFs has been qualitatively studied in the Auditory Image Model’s buffer [46] and its derivative [57], and quantitatively in the derivative of the model output in [45] and [10]. However, these models did not provide a mechanistic explanation of the processes underlying the generation of the POR or its latency dependence with pitch.
Other models, designed to explain the biophysical mechanisms of pitch perception, focused primarily on subcortical processing. Two of these models describe how neurons, mainly in subcortical nuclei, might process periodicities from the auditory nerve activity: Meddis and O’Mard’s model [23] proposes a biophysical implementation of the summary autocorrelation function [27, 28], based on the joint action of chopper neurons in the cochlear nucleus and coincidence detectors in the inferior colliculus. More recently, Huang and Rinzel [58] described a neural implementation of an array of coincidence detectors able to detect periodicities by comparing neural activity across different cochlear channels. Despite their mechanistic differences, both models present an output comparable to that of the autocorrelation function [58]. The model presented here is downstream with respect to Meddis’ and Huang’s models because it focuses on explaining how pitch decisions based on the later subcortical representation are made in alHG.
In our model, pitch processing is mediated by a connectivity pattern among interacting columns specialized in characteristic periods. Similar connectivity patterns were found in mice AC, stemming from L4 and targeting L6 neurons [36], and in the cat AC in earlier studies [37]. Neurons that respond selectively to harmonically related input have been recently identified in the core region of the marmoset’s auditory cortex [13].
Inhibitory and facilitatory interactions between neurons encoding harmonically related frequencies are often reported in the mammal auditory cortex (see [59] for a review). Specifically, intracranial recordings in marmoset AC revealed that activation elicited by a given tone resulted in the facilitation of neurons encoding higher harmonics, and in the suppression of neurons encoding lower harmonics [60], in line with the decoder network mechanisms of our model. Harmonic co-activation has also been shown in human AC [61].
In a more speculative vein, we suggest that this connectivity pattern might result from spike-timing-dependent plasticity (STDP) operating over cortical neurons receiving the upstream outputs of periodicity detectors. To illustrate this, let us consider the processing of a sound of fundamental period T. After tone onset, the first periodicity detector responding to the sound provides input to the upstream excitatory ensemble encoding T, which subsequently activates its inhibitory counterpart. Assuming an initial all-to-all connectivity, this inhibitory drive propagates in the network and provides concurrent input to neurons receiving excitatory drive from periodicity detectors 2T, 3T, and so forth. Input synchrony would result in a stronger connectivity change though STDP in these harmonically related ensembles, whilst the uncorrelated asynchronous inputs to the remaining ensembles would result in a net decreased connectivity weight. A similar STDP mechanism for spectral pitch integration was proposed earlier in [62].
The decoding strategy or our model is based on the well-known winner-takes-it-all architecture [29, 63, 64]: excitatory populations in the decoding network compete with each other, while the inhibitory ensemble arbitrating this competition is the one in the column that is sensitive to the fundamental period (Fig 1). In this way, multiple fundamentals can be simultaneously decoded (Fig 5). Moreover, akin to recent models of sensory integration [29], once a fundamental period is represented in the decoder network, the activity of the winner column is reinforced by the sustainer network (rather than the pitch being repeatedly decoded). This ensures stability until a significant change in the subcortical input triggers a new decoding process (see S2 Fig).
This sustaining strategy is also related to predictive coding-related strategies [45, 65, 66], where top-down efferent convey expectations about the input, whereas bottom-up afferents convey prediction errors [65]. Additional top-down expectations could coexist at higher cognitive levels based on, for example, prior knowledge, experience, or focused attention. Such biases could modulate the sustainer network by increasing the baseline activity of the inhibitory ensembles that characterize the target pitch values, thereby facilitating pitch processing in the decoder network.
To summarize, in this study we proposed a model specifically designed to understand the neural mechanisms of cortical pitch processing at a mesoscopic scale. We introduce a possible mechanistic link between the latency of the POR component in the N100 deflection and the processing time required for the system to achieve convergence, explaining the classical result that tones with a lower pitch elicit PORs with longer latencies. More intriguingly, our modeling and experimental results indicate that processing time varies with the degree of consonance in dyads, suggesting that the sensation of consonance and dissonance might start early in auditory cortex, prior to higher-order processing.
|
10.1371/journal.pgen.1000511 | The Level of the Transcription Factor Pax6 Is Essential for Controlling the Balance between Neural Stem Cell Self-Renewal and Neurogenesis | Neural stem cell self-renewal, neurogenesis, and cell fate determination are processes that control the generation of specific classes of neurons at the correct place and time. The transcription factor Pax6 is essential for neural stem cell proliferation, multipotency, and neurogenesis in many regions of the central nervous system, including the cerebral cortex. We used Pax6 as an entry point to define the cellular networks controlling neural stem cell self-renewal and neurogenesis in stem cells of the developing mouse cerebral cortex. We identified the genomic binding locations of Pax6 in neocortical stem cells during normal development and ascertained the functional significance of genes that we found to be regulated by Pax6, finding that Pax6 positively and directly regulates cohorts of genes that promote neural stem cell self-renewal, basal progenitor cell genesis, and neurogenesis. Notably, we defined a core network regulating neocortical stem cell decision-making in which Pax6 interacts with three other regulators of neurogenesis, Neurog2, Ascl1, and Hes1. Analyses of the biological function of Pax6 in neural stem cells through phenotypic analyses of Pax6 gain- and loss-of-function mutant cortices demonstrated that the Pax6-regulated networks operating in neural stem cells are highly dosage sensitive. Increasing Pax6 levels drives the system towards neurogenesis and basal progenitor cell genesis by increasing expression of a cohort of basal progenitor cell determinants, including the key transcription factor Eomes/Tbr2, and thus towards neurogenesis at the expense of self-renewal. Removing Pax6 reduces cortical stem cell self-renewal by decreasing expression of key cell cycle regulators, resulting in excess early neurogenesis. We find that the relative levels of Pax6, Hes1, and Neurog2 are key determinants of a dynamic network that controls whether neural stem cells self-renew, generate cortical neurons, or generate basal progenitor cells, a mechanism that has marked parallels with the transcriptional control of embryonic stem cell self-renewal.
| Neural stem cells make all of the neurons in the brain. A key feature of these cells is the ability to regulate the balance between making more neural stem cells, the process of self-renewal, and making nerve cells, the process of neurogenesis. Too much self-renewal would result in a brain with too few neurons and abnormal circuitry; too much neurogenesis would deplete all of the neural stem cells too quickly, resulting in a small brain and neurological abnormalities. Little is currently known of the how neural stem cells control this fundamental choice. We used one transcription factor, Pax6, which is important for this decision, as an entry point to define the cellular networks controlling neural stem cell self-renewal and neurogenesis in the developing mouse brain. We found that the relative amount of Pax6 controls the balance between self-renewal and neurogenesis in neural stem cells. Increasing Pax6 levels drives the system towards neurogenesis, at the expense of self-renewal, by turning on a genetic programme for making neurons, whereas decreasing Pax6 turns off the genetic programme for neural stem cell self-renewal. In both cases, altering the levels of Pax6 ultimately leads to a small brain, but through very different mechanisms.
| A fundamental feature of neural development is the production of defined types of neurons in a temporal order from multipotent, regionally-specified neural stem and progenitor cells [1]. During nervous system development, maintaining the balance between stem cell self-renewal and neurogenesis is essential for the generation of the correct proportions of different classes of neurons and subsequent circuit assembly. Little is known of the molecular control of the key neural stem (NS) cell properties of multipotency and self-renewal. This is in contrast to other classes of stem cells, most notably embryonic stem (ES) cells, in which a group of three transcription factors, Sox2 and the two ES-specific factors Oct4 and Nanog, co-operate to control pluripotency and self-renewal in a non-redundant manner [2],[3].
The paired-domain, homeodomain-containing transcription factor Pax6 is highly conserved among vertebrate and invertebrate species and is essential for the development of much of the central nervous system, including the eye, spinal cord and cerebral cortex, as well as pancreatic islet cells [4]–[7]. Detailed analyses of neocortical and retinal development in mice mutant for Pax6 have identified defects in neural stem and progenitor cell proliferation, multipotency, neurogenesis, the generation of specific types of neurons, and marked changes in spatial pattern [8]–[19]. In the neocortex, loss of Pax6 function results in microcephaly, abnormal development of the secondary progenitor population of the subventricular zone (SVZ, also known as basal progenitor cells, BP cells) and a disproportionate reduction in the production of later-born, upper layer neurons [12], [14], [15], [17], [20]–[23].
Given the functions of Pax6 in stem cell self-renewal/proliferation and neurogenesis, a potentially fruitful approach to uncovering cellular pathways controlling these processes is to identify the downstream targets of Pax6 in neocortical stem cells. Therefore, in this study we used Pax6 as an entry point to define the cellular networks regulating neural stem cell self-renewal and neurogenesis by combining chromatin immunoprecipitation (ChIP) to identify Pax6-bound promoters with the anatomical phenotypes and gene expression changes observed in Pax6 mutant cortices.
To do so, we identified where in the genome Pax6 binds in mouse neocortical stem cells during normal development, defining the potential cellular networks regulated by Pax6. As binding does not necessarily imply regulation, we also studied the transcriptional consequences of altering Pax6 levels in the neocortex in vivo. Developing tissues are sensitive to Pax6 dosage: heterozygous human mutations in Pax6 result in aniridia and forebrain abnormalities [24],[25], as do a number of mouse mutations (for reviews, see [7],[26]), humans and mice homozygous for Pax6 mutations typically lack eyes and have marked microcephaly and absent olfactory bulbs [5],[6],[19], whereas increasing Pax6 levels in transgenic mice results in microphthalmia and forebrain abnormalities [27]–[29]. Therefore, we examined the transcriptional consequences of both increasing and decreasing Pax6 levels in the developing cortex to identify those promoters actively regulated by Pax6 in neocortical stem and progenitor cells.
We find that Pax6 controls the balance between self-renewal and differentiation in neural stem cells in a dose dependent manner, positively and directly regulating cohorts of genes that promote self-renewal, basal progenitor cell genesis and neurogenesis. In addition, we found that Pax6 interacts with three other regulators of cortical stem cell neurogenesis, Neurog2, Ascl1 and Hes1. The four transcription factors regulate one another and many of the same target genes in an antagonistic manner, defining a core self-renewal/neurogenesis network that is dependent on a critical level of Pax6. In support of this, we found in phenotypic analyses of Pax6 gain and loss of function cortices that an increase in Pax6 levels drives the system towards neurogenesis and basal progenitor cell genesis at the expense of self-renewal, whereas removing Pax6 reduces cortical stem cell self-renewal. In both cases, altering the levels of Pax6 ultimately leads to microcephaly, but through different cellular and biological pathways.
The goal of this research was to define the genetic networks directly and indirectly regulated by Pax6 in neocortical stem and progenitor cells and to characterise the biological networks of which Pax6 is a component. To do so, we combined location analysis of Pax6 bound genomic binding sites in neocortical stem and progenitor cells in vivo with transcriptome analysis of the changes in gene expression resulting from altered levels of Pax6, together with phenotypic data from these and previous studies of Pax6 gain- and loss-of-function in the developing mouse cerebral cortex (Figure 1). Finally, we integrated these data with expression data from previous studies of other transcription factors that have roles in controlling neocortical neurogenesis, most notably Neurogenin2 (Neurog2), Hes1 and Ascl1/Mash1 [15],[30],[31], in order to extend our knowledge of the biological networks in which Pax6 operates.
To identify promoters targeted by Pax6, we applied in vivo location analysis to identify where in the genome Pax6 is bound in neocortical stem and progenitor cells. We carried out chromatin immunoprecipitation (ChIP) for Pax6-bound DNA in the developing mouse neocortex at embryonic day 12.5 (E12.5), at which stage Pax6 is expressed by all neocortical ventricular zone stem and progenitor cells, but not by post-mitotic neurons [32]. A set of three biologically independent Pax6 ChIP experiments was carried out using oligonucleotide microarrays spanning ∼8.5 kb around the transcriptional start sites of over 17,000 mouse genes (see Materials and Methods for details). As the arrays used sample the proximal promoters of these genes, this analysis will not include intronic or distally located enhancers bound by Pax6, such as the Pax6-bound cortical enhancer of Neurog2 expression, found ∼9 kb upstream of the Neurog2 start site [33],[34].
Data from the three experiments were analysed to identify regions of promoters bound by Pax6 in each ChIP and the individual analyses combined to define a set of 1560 genes with Pax6-bound promoters (Figure 2; Table S1), of which 1172 are genes with known or predicted functions (Figure 2H). A set of five genes defined as Pax6-bound by the array study were confirmed as bound by qPCR analysis of additional Pax6 ChIP from the E12.5 cortex using two different Pax6 polyclonal antibodies (Figure 2G).
Functional annotation of the set of Pax6-bound genes by Gene Ontology analysis found a significant enrichment for developmental processes among Pax6 target genes, including nervous system and eye development (Figure 3). Broad functional categories enriched in the Pax6-bound set included cell cycle regulation, regulation of proliferation, cell-cell signalling, neurogenesis and neuron differentiation. At the functional level, Pax6-bound genes are enriched for genes with transcription factor activity and chromatin binding (Figure 3A and 3D). Pax6 binds transcription factors that are expressed in stem and progenitor cells (Pax6 itself, Hmga2, Cutl1, Nr2f2, Emx2, Sox9, Neurog3, Tle1) and basal progenitor cells in the cortex (Eomes/Tbr2, Neurod1), as well as to transcription factors expressed in newly born, differentiating neurons (Sox4, Sox11) or in terminally differentiating cortical neurons (Rorb, Etv1). Together with the binding of Pax6 to sets of regulators of cell cycle progression and proliferation (for example, Ccna1, Pten, Cdkn1b, Cdk4, Fzr1 and Ctnnb1), regulation of these transcription factors confers the potential for Pax6 to control stem cell self-renewal/proliferation, neurogenesis and cell fate determination within the cortex.
Pax6 binding data defines the potential networks controlled by Pax6 in this cell type. However, binding does not necessarily equate with activity. Therefore, to characterize the transcriptional networks within which Pax6 acts, we studied gene expression in the E12.5 mouse cortex, using microarrays, following Pax6 gain and loss of function (Figure 4). Intersection of genes that are positively or negatively regulated by Pax6 defined the transcriptional networks dependent on Pax6, without discriminating between direct and indirect regulation.
For Pax6 gain of function, we used the D6 enhancer that drives cortex-specific expression in the ventricular zone and cortical plate [35]–[37] to generate transgenic mice expressing the canonical Pax6 isoform, giving rise to animals with a two-fold increase in Pax6 protein in the ventricular zone, as assessed by immunohistochemistry (Figure 4A–4C). Expression profiling of cortices from single wild-type and D6-Pax6 transgenic littermates at E12.5 found 3784 genes showing significantly altered expression (false discovery rate, FDR, of 0.1), of which 2238 (59%) were upregulated (Figure 4D; Table S2). In comparison, expression profiling of the Pax6 homozygous mutant cortex (Sey/Sey) at E12.5 identified 938 genes with significantly altered expression, of which 600 (64%) were up-regulated (Figure 4E; Table S3). Of the 938 genes showing altered expression in the Pax6 null mutant cortex, 298 (32%) were also altered in expression in the D6-Pax6 cortex (Figure 4F). From the transcriptome analyses, we defined two sets of genes showing Pax6-dependency: the set of genes showing positive dependency on Pax6 was the set formed by the union of genes up-regulated in the D6-Pax6 cortex and the genes down-regulated in the Pax6 mutant cortex (Figure 4F and 4G; 2344 genes); the set of genes showing negative dependency on Pax6 was the set formed by the union of genes down-regulated in the D6-Pax6 cortex and the genes up-regulated in the Pax6 mutant cortex (Figure 4F and 4G; 1920 genes).
Genes demonstrating positive dependency on Pax6 were notable for including many genes expressed in basal progenitor cells, including Eomes/Tbr2, Gadd45g, Neurod1, Tcfap2c and Hes6 (Figure 4H). Genes similarly dependent on Pax6 for expression include genes expressed in cortical stem and progenitor cells associated with self-renewal and proliferation, such as the transcriptional regulators Hmga2 and Hes5, and the cell cycle regulators Cdk2, Cdk4 and Ccne1. However, transcription factors associated with neurogenesis, such as Neurog2, Sox4 and Sox21, also show positive dependency on Pax6 for their expression. In addition, increasing Pax6 expression leads to an increase in expression of transcription factors that are preferentially expressed in cortical neurons of layers 5 and 6 (Tbr1, Zfp312/Fezf2, Nfia, Nfib).
In contrast, genes demonstrating negative dependency on Pax6 included genes with periodic expression in the cell cycle (FoxM1, Mcm3, Mcm5). These genes were reduced in expression in the Pax6-overexpressing cortex, indicating that increasing Pax6 may alter cell cycle length. Together, the transcriptional analyses demonstrate that Pax6 regulates neural stem cell maintenance, basal progenitor cell genesis and neurogenesis.
To identify those genes that are both bound and regulated by Pax6, we analysed the intersection between the sets of Pax6 bound genes and those genes showing Pax6 dependency from the gain and loss of function transcriptional analyses (Figure 5). Of the set of 1560 Pax6-bound genes identified by the in vivo location analysis, 343 (22%) show significantly changed expression in either the Pax6 gain or loss of function cortex at E12.5 (Figure 5A; Table S4). Of the 343 bound and regulated genes, 180 were positively regulated and 143 negatively regulated by Pax6 (Figure 5B and 5C), with 20 genes showing conflicting regulation (either up- or down-regulated upon both gain and loss of Pax6 function).
Using publicly available in situ hybridization data (GenePaint.org; [38]), we assigned cellular expression to 279 of the 343 Pax6 bound and regulated genes in the E14.5 mouse cortex, the mRNAs of 223 of which were detectable in the cortex at that age (Figure 5D–5F; Table S5). Ventricular zone expression was found for 188 of those genes (84%), 81 of which were solely expressed in the ventricular and subventricular zone. A noteworthy minority of Pax6 bound and regulated genes (21; 9% of detectable mRNAs) were exclusively expressed in differentiating neurons in the cortical plate.
The combination of Pax6 binding data with the transcriptional profiling of Pax6 gain- and loss-of-function cortices enabled the delineation of the direct and indirect networks regulated by Pax6 (Figure 6) and prediction of Pax6 functions in cortical stem and progenitor cells. As shown, Pax6 positively controls sets of genes promoting neural stem cell self-renewal and/or maintenance: Pax6 positively regulates transcription of Hmga2, Tle1 and Cdk4 directly, promoting cell cycle progression and neural stem cell maintenance [39]–[41]. Furthermore, Pax6 controls expression of D-class cyclins, Hes5 and Notch ligands indirectly [39],[42],[43], underlining the importance of Pax6 in promoting neural stem cell self-renewal.
However, the self-renewal functions of Pax6 are offset by its binding to, and positive regulation of, genes and transcription factors that promote neurogenesis, including the tumour suppressor genes Pten and Fzr1/Cdh1 and the transcription factors Neurog2 and Sox4 [31], [44]–[46]. In addition, Pax6 directly and positively controls genes that promote basal progenitor cell genesis, including one of the major determinants of the basal progenitor fate, Eomes/Tbr2, as well as Neurod1 and Gadd45g [47]–[49]. Together, these functions of Pax6 indicate that one biological role for Pax6 is to promote neurogenesis by increasing the number of neocortical stem and progenitor cells either exiting the cell cycle or becoming basal progenitor cells, and thus neurons at their next division.
In addition to roles in neurogenesis and self-renewal, Pax6 controls genes that regulate the neocortical identity of the neurons produced in the dorsal pallium. It does so by primarily negatively regulating genes associated with neuronal cell fates, and subpallially-derived inhibitory neuron and interneuron identity in particular, through repression of key transcription factors that confer interneuron fates (Isl1, Foxd1, Ascl1, Dlx1, Lhx8; Figure 6; [50]).
Many of the findings reported here on the functions of Pax6 in neurogenesis and self-renewal are in contrast with the recently described functions of the bHLH transcription factor Hes1, sustained overexpression of which represses many of the genes upregulated by Pax6 [51]. Comparison of the expression data from Hes1 overexpression with the Pax6 data presented here shows that Pax6 and Hes1 have opposing functions on neurogenesis and basal progenitor genesis through the same sets of effector genes (Figure 6): Pax6 positively regulates Neurog2 and Eomes/Tbr2, for example, whereas both of these key transcription factors are repressed by Hes1. As discussed in more detail below, this overlap suggests that Pax6 and Hes1 are both regulating a core mechanism for controlling neural stem cell self-renewal and neurogenesis.
To test the prediction of the functions of the Pax6-regulated network described above, we studied neurogenesis and cell fate determination in the Pax6 gain-of-function cortex (D6-Pax6 transgenic mouse; Figure 7). As predicted, increased Pax6 resulted in a marked increase in the expression of Eomes/Tbr2 at E12.5, demonstrating the increase in the basal progenitor cell population. Furthermore, there was with an increase in early born, layer 6 neurons (Tbr1-expressing cells), without an increase in the total number of neurons produced by this stage of development, compared to controls (as evidenced by the number of cells expressing neuron-specific tubulin, Tuj1, Figure 7).
However, by E14.5 the D6-Pax6 transgenic cortex was significantly smaller than that of controls, with a reduction in total neuron number (Figure 7). The reduction in cortical neuron number at this stage is consistent with the early increase in Pax6 expression driving cortical stem and progenitor cells towards an inappropriately early basal progenitor and neuronal fate. The reduction in cortical size is not due to an increase in cell death in the Pax6 overexpressing cortex, as no significant increase in apoptosis could be detected by TUNEL staining (Figure S1). The early depletion of the stem cell pool reduces the number of stem cells available for neurogenesis at subsequent stages, resulting in an overall reduction in cortical size and total cortical neurogenesis by E14.5 (Figure 7).
Neurogenin2 (Neurog2) and Ascl1/Mash1 are two proneural bHLH transcription factors with well-established roles in neocortical neurogenesis and cell fate [15],[31]. Both interact with Pax6: Pax6 positively regulates Neurog2 via an enhancer [34], whereas Ascl1 expression is upregulated in Pax6-null cortices [52]. Gene expression profiling of the Neurog2 mutant cortex and the Ascl1-null ventral telencephalon has been analysed to define genes downstream of both factors [53]. As with Hes1, there are striking overlaps in the sets of genes downstream of Ascl1, Neurog2 and Pax6 in the developing telencephalon.
Placing Pax6 in the context of these three transcription factors controlling cortical neurogenesis, Hes1, Ascl1 and Neurog2, enables the definition of a core transcriptional circuit controlling cortical neural stem cell self-renewal and neurogenesis (Figure 8A). Pax6 both positively regulates the expression of Neurog2 and also synergises with Neurog2 to promote basal progenitor cell genesis and thus production of cortical excitatory projection neurons, whereas Hes1 opposes this process by repressing many of the same genes. Ascl1 has complex functions in cortical neurogenesis: while it also promotes basal progenitor cell genesis, it also drives expression of transcription factors to promote inhibitory interneuron genesis (Lhx8 and Isl1, for example). Pax6 and Hes1 both repress Ascl1, and Pax6 represses Lhx8 and Isl1, to inhibit the interneuron-producing functions of Ascl1.
This network also leads to clear predictions of the normal functions of Pax6 in regulating neurogenesis and the consequences of altered Pax6 expression in the early cortex (Figure 8). The network indicates that Pax6 is essential for the cortical identity of the basal progenitor cells produced in the cortex, as it is an essential driver of Eomes/Tbr2 expression, a key determinant of cortical basal progenitor cell identity [47],[48]. Given the repression of Ascl1 by Pax6 in cortical stem cells and the interneuron-promoting function of Ascl1, loss of Pax6 function should result in a basal progenitor cell population lacking cortical identity (due to loss of Eomes/Tbr2 expression), a decrease in the genesis of cortical pyramidal neurons and an increase in the production of inhibitory interneurons (Figure 8B). These are all phenotypes observed in the Pax6 null cortex [8],[9],[14],[52].
In contrast, increasing Pax6 expression would be predicted to increase basal progenitor genesis early in cortical development and increase pyramidal cell genesis both directly from the ventricular zone and indirectly from basal progenitor cells (Figure 8C). In principle, Pax6 could increase cortical stem cell self-renewal/proliferation, but this would be in competition with all of the functions of Pax6 that promote neurogenesis: reduced proliferation/cell cycle exit (via Pten and Fzr1/Cdh1 expression), basal progenitor cell genesis (via Eomes/Tbr2 and Neurod1) and neurogenesis (via Neurog2 and Sox4). In the D6-Pax6 cortex, as described above, an increase in basal progenitor genesis is observed early in development, accompanied by an increase in the production of early-born cortical pyramidal cells.
Our in vivo analysis supports the finding that Pax6 operates in and regulates a core transcriptional network that controls the balance between neurogenesis and stem cell self-renewal in a highly dosage-sensitive manner. Altering the amount of Pax6 in neural stem cells has profound effects on the output of neural stem cells, ultimately compromising the ability of neural stem cells to generate all of the required neurons for normal assembly of the cerebral cortex. This raises the question as to whether Pax6 levels do vary or oscillate in vivo, as has been reported for several other genes in developing systems [54].
The levels of Hes1 and Neurog2 proteins vary with cell cycle stage and have also been shown to oscillate in neocortical stem and progenitor cells with a 2–3 hour frequency [51], in a Notch-dependent fashion in the case of Hes1. Immunohistochemistry for Pax6, Neurog2 and Hes1 shows that Pax6 does not show obvious cell cycle-dependent changes in levels, unlike Hes1 and Neurog2, as assessed by nuclear location in the ventricular zone, as it is expressed in all neocortical stem and progenitor cells at relatively high levels (Figure 9A; Figure S2). This approach cannot resolve oscillations with a short periodicity, as observed for Hes1 and Neurog2 by live imaging [51]. However, Hes1-expressing cortical stem and progenitor cells also express high levels of Pax6, so it is unlikely that Hes1 represses Pax6 expression as it does Neurog2 [51]. In the presence of Notch signalling, increased Hes1 and Hes5 activity suppresses neurogenesis and promotes self-renewal, in part by repression of Neurog2, assisted by the activation of Hes5 expression by Pax6 (Figure 9B). In the absence of Notch signalling and Hes1 activity, Pax6-driven increase in Neurog2 and the resulting drive to neurogenesis are unopposed, allowing Pax6 to promote neurogenesis (Figure 9C).
Therefore, the identification of the direct transcriptional targets of Pax6 in neocortical stem and progenitor cells, combined with phenotypic analyses of the Pax6 gain and loss of function cortex, have enabled the elucidation of a regulatory network controlling the balance between neurogenesis and self-renewal. The development of this model provides a framework in which to explore questions arising from the mechanism reported here. For example, only a subset of Pax6+/Hes1− stem cells in G1 phase of the cell cycle go on to generate neurons in the next round of cell division. Therefore, there are likely to be additional components of this network that promote self-renewal and suppress neurogenesis.
Neural stem cell self-renewal, neurogenesis and cell fate determination are three forces that control the generation of specific classes of neurons at the correct place and time during development. Much remains to be discovered of the cellular networks operating to control these processes. We report here that the transcription factor Pax6 directly regulates genes controlling the balance between neocortical stem cell maintenance, neurogenesis and the production of basal progenitor cells in a dosage-dependent fashion. While consistent with genetic loss-of-function studies [10], [14], [20], [52], [55]–[58], the direct nature of the control of these processes by Pax6 and their dosage sensitivity are unexpected. We propose that this dosage sensitivity reflects the need for a critical level of Pax6 within neocortical stem and progenitor cells.
Using Pax6 ChIP, we have identified a set of the promoters bound by Pax6 in neocortical stem cells in vivo at E12.5. This set of Pax6-bound genes defines the components of potential networks regulated by Pax6 in this tissue, and is noteworthy for the enrichment for genes involved in controlling cell cycle progression, transcription factors expressed in the three main cell types found in the early cortex (ventricular zone stem cells, basal progenitor cells and differentiating neurons) and also transcription factors expressed specifically in non-cortical neurons, including the ventral forebrain, spinal cord and retina, as well as in pancreatic islet cells.
These binding data are compatible with a number of models for Pax6 action. For example, Pax6 could repress expression of genes that are not normally expressed in the neocortex, or alternatively could simply bind to target sites in the promoters of genes normally expressed in other cell types in a Pax6-dependent manner, without driving their expression in the cortex. Similarly, for those genes expressed in cortex, Pax6 could either positively or negatively regulate cell cycle progression, and positively or negatively regulate basal progenitor cell genesis. Evidence for gene regulation by Pax6 is essential to resolve these questions. Therefore, we combined the binding data with transcriptome data from Pax6 gain- and loss-of-function experiments in the early developing cortex in order to identify those genes whose expression is dependent on Pax6 and the nature of that dependency, finding that 22% of Pax6-bound genes show evidence for regulation in vivo.
Transcriptome analyses of Pax6 gain and loss of function cortices identified a subset of genes that show reciprocal regulation, but also clearly demonstrate that the gain and loss of function gene expression changes are not simply complementary, in agreement with the reported anatomical phenotypes of Pax6 null and Pax6 over-expressing cortices [14],[21],[27],[29]. However, there is a striking overlap and complementarity in the changes of expression in basal progenitor genes observed in each mutant. We found that Pax6 directly promotes expression of a large set of genes specifically expressed in basal progenitor cells, including the key determinant of that cell type, the transcription factor Eomes/Tbr2: mutations in Eomes/Tbr2 lead to a loss of the cortical intermediate progenitor cell population, accompanied by a reduction in neurons in all cortical layers [47],[48]. Increasing Pax6 levels drives basal progenitor cell genesis from cortical stem cells, primarily by increasing Eomes/Tbr2 expression, along with the other basal progenitor cell genes such as Gadd45g, Neurod1, Sstr2 and Hes6 [49]. Basal progenitor cells undergo a limited number of mitotic divisions to generate neurons [59], thus the overall effect of shifting the stem cell population towards basal progenitor cells is to increase neurogenesis at the expense of neural stem cell maintenance in the early stages of cortical development, ultimately resulting in microcephaly, as observed here.
However, there are also changes in expression specific to either gain or loss of Pax6 function, with many more genes showing altered expression upon increased Pax6 levels. For example, from the combined Pax6 binding and regulation data we have found that Pax6 expression positively regulates stem cell self-renewal by promoting expression of the transcription factor Hmga2 and the G1 cyclin dependent kinase, Cdk4. Hmga2 promotes neural stem cell self-renewal by reducing expression of two negative regulators of the cell cycle, p16Ink4a and p19Arf [39]. Hmga2 reduces expression of p16Ink4a and p19Arf indirectly via repression of JunB, a positive regulator of their expression, and we also found Pax6 binding to the promoter of JunB in cortical stem cells. p16Ink4a slows cell cycle progression by inhibiting the G1 cyclin-dependent kinase Cdk4 [60], and Cdk4 is bound and positively-regulated by Pax6 in cortical stem cells. In contrast, Pax6 also directly promotes expression of Pten and Fzr1/Cdh1, both of which reduce neural stem cell proliferation and self-renewal [44],[61]. Thus, under normal conditions in vivo Pax6 has the potential to both promote and limit stem cell self-renewal. However, when Pax6 levels are increased, as in the D6-Pax6 transgenic cortex, the neurogenic functions of Pax6 are dominant over its ability to promote self-renewal.
We have also placed Pax6 in the context of other transcriptional regulators of self-renewal and neurogenesis, Hes1, Neurog2 and Ascl1/Mash1, in order to extend our coverage of the cellular networks controlling these processes. The marked overlap between those genes directly and indirectly regulated by Pax6 with the genes regulated by all three of the other factors provides strong evidence for the operation and architecture of the network regulating cortical neurogenesis, and the central importance of the basal progenitor population as a major output of that network. Pax6 and Neurog2 cooperate to promote neurogenesis, both directly and via the basal progenitor population, and this is opposed by the oscillating expression of Hes1 [51]. At the same time, Pax6 also promotes stem cell self-renewal in a manner that counterbalances its neurogenesis-promoting activity. However, when over-expressed, the promotion of neurogenesis and basal progenitor cell genesis by Pax6 is dominant over the promotion of self-renewal.
Therefore, we propose that there is an optimal level of Pax6 that determines the balance between neocortical stem cell self-renewal and neurogenesis: increasing that level drives stem cells to a neuronal or basal progenitor fate, whereas reducing the level leads to early cell cycle exit, manifest as increased early neurogenesis [14]. In both cases, altering Pax6 levels leads to a depletion of the stem cell population by exiting to neurogenesis, but by different pathways and with different neuronal fates: cortical pyramidal cells when Pax6 is increased, inhibitory interneurons when Pax6 is absent. The sensitivity of cortical development to Pax6 levels underlines the importance of assessing subtle structural and functional anomalies in humans heterozygous for Pax6 mutations, as has been done for aniridia patients [22],[25].
Finally, the findings of Pax6 function in neocortical stem and progenitor cells presented here have similarities with the functions of the ES cell pluripotency regulator Oct4 [3]. Oct4 shows marked dosage effects in ES cells in vitro such that a reduction in Oct4 levels leads to trophectoderm differentiation and a two-fold increase in Oct4 levels leads to differentiation to primitive endoderm and mesoderm [3]. Loss of Pax6 leads to a depletion of the cortical stem cell pool, via increased early neurogenesis secondary to a failure to self-renew, and also a switch in the fates of the neurons produced from glutamatergic cortical neurons to an inhibitory interneuron identity [52]. Similarly, increased Pax6 expression also leads to depletion of the stem cell pool, but in this case by driving stem cells to a basal progenitor fate, leading to an overproduction of early-born, deep-layer cortical neurons. Thus the level of Pax6 controls whether neural stem cells will self-renew, generate cortical neurons or produce basal progenitor cells.
Chromatin immunoprecipitation (ChIP) was performed as described [62], with minor modifications. An average of 36 neocortices from E12.5 MF1 mouse embryos were used for each ChIP. For these and subsequent studies, all animal work was approved by local ethics review committees and, where relevant, carried out according to UK Home Office national guidelines. ChIP was carried out with specific rabbit polyclonal Pax6 antibodies generated against C-terminal peptides that recognise both splice variants of Pax6: C-term 1 -Chemicon, Cat no. AB5409 and C-term 2 - Covance PRB-278P. For array analysis, Pax6-bound genomic DNA was purified by ChIP using the Chemicon antibody, and ChIP material and whole cell extract DNA were globally amplified by ligation-mediated PCR [62]. To validate ChIP-on-chip data, unamplified ChIP material prepared with both Pax6 antibodies and from control samples lacking the primary antibody were used for gene-specific, quantitative PCR. Primers to amplify 100–250 bp target regions surrounding the predicted genomic binding locations for Pax6 were designed using Primer3 (http://frodo.wi.mit.edu/) [63] or Primer Express (Applied Biosystems), and checked for specificity in the genome using the BLAT algorithm (http://genome.ucsc.edu/). Quantitative PCR was carried out using the Roche Lightcycler system or the Applied Biosystems 7300 system. The amount of target regions was quantified in Pax6 and control chromatin immunoprecipitations lacking the primary antibody (no-antibody controls, NoAb). The enrichment for each gene was calculated by normalising the Pax6/NoAb ratio against the Pax6/NoAb ratio for a promoter region that was not found to be bound by Pax6 (non-bound control region from the Syt8 gene).
Amplified Pax6 ChIP and whole cell extract DNA were labelled by indirect incorporation of Cy3 or Cy5-labelled nucleotides and hybridised to Agilent mouse promoter 244 K arrays according to the manufacturer's instructions. Slides were scanned in an Agilent scanner and data extracted and normalised using Agilent Feature Extractor. Data were analysed by neighbour analysis [62] implemented in ChipAnalytics (Agilent) to calculate probability scores of Pax6 binding for each array oligonucleotide (p[Xbar] scores). Binding events were identified by sliding a 1000 bp window across the genomic space covered by the promoter array. Within a window, the best p[Xbar] score from each of the three ChIP-chip experiments was identified. A binding event was called if: (1) 2 or more scores were < = 0.015, and (2) the product of three scores was < = 0.000025. Gene ontology analysis of Pax6-bound genes was performed using GOToolBox (http://crfb.univ-mrs.fr/GOToolBox/home.php).
To generate D6-Pax6 transgenic mice, the 5.7 Kb D6 promoter fragment was cloned upstream of the canonical Pax6 open reading frame [35]–[37] and injected into the pronucleus of fertilized (C57BL/6×BALB/c) F1 mouse oocytes to generate founder mice. In situ RNA hybridization using digoxigenin (DIG)-labeled RNA probes was performed according to methods described at the Rubenstein lab website (http://physio.ucsf.edu/rubenstein/protocols/index.asp). Sections from the different genotypes (WT and D6-Pax6) were processed in parallel. Basal ganglia expression was used an internal control to compare results between different experiments and between experimental and WT samples. Probes used: Pax6 (P. Gruss), Neurog2 (F. Guillemot), Tbr1 and Eomes/Tbr2 (Rubenstein lab). Immunohistochemistry was performed on frozen sections (10 or 20 µm). Antibodies used: monoclonal anti-βIII-tubulin antibody (clone TUJ1; Covance), 1∶1000; anti-phospho-Histone H3 (Ser10) (Upstate) 1∶200; anti Pax6 (Developmental Studies Hybridoma Bank) 1∶1000; anti-Eomes/Tbr2 (gift from Robert Hevner, University of Washington School of Medicine, Seattle). Pax6 levels were quantified in 9 wildtype and 14 D6-Pax6 tissue sections by the intensity of the immunofluorescent signal: the signal along the dorsal-ventral axis was quantified in the ventricular zone by histogram (ImageJ, NIH), using the mean intensity. For this quantification the Developmental Studies Hybridoma Bank anti-Pax6 antibody was used at 1∶1000. TUNEL analysis was performed on 20 µm cryostat sections using the Apoptag Kit following the manufacturer's recommendations (Millipore, CA, USA). Similar results were observed with activated Caspase 3 staining (data not shown).
Triple immunohistochemistry for Pax6, Hes1 and Neurog2 was carried out on sections of E12.5 cortex by standard techniques (rabbit anti-Pax6 antibody, Covance PRB-278P; guinea pig anti-Hes1 antibody, gift from Ryoichiro Kageyama; goat anti-Neurog2, Santa Cruz Biotechnology). Quantification of the fluorescent intensity of nuclear staining for each antibody was carried out on confocal microscrope images (Radiance, Biorad) of a minimum of three sections using Volocity software (Improvision).
Cortices were dissected from individual embryos in two litters of E12.5 Pax6Sey (Edinburgh Small-eye, [6]) mutant embryos and from one litter of E12.5 D6-Pax6 transgenic mice. Total RNA was extracted and cDNA synthesised using the SMART system, as described [64]. Gene expression in neocortices from three single Pax6Sey/Sey embryos was compared to that in four single wildtype littermates on six dye-swapped oligonucleotide microarrays; while gene expression in neocortices from three single D6-Pax6 embryos was compared to that in three single wildtype littermates at E12.5 in a set of 6 paired dye-swapped hybridizations. Arrays of the MEEBO oligonucleotide set (Invitrogen), produced by the Pathology Department, University of Cambridge were used for all studies. cDNA labeling, array hybridization, slide scanning (Axon GenePix microarray scanner, Molecular Devices) and data extraction were performed as described [64]. Expression data were archived and lowess normalized using the Acuity system (Molecular Devices). Log ratios of all expression measurements for each array were median-centered and expression ratios variance normalized across all of the arrays (without adjusting the average fold-change). Genes were filtered on the basis of the number of arrays on which they were detected, being required to be present in 4 of the 6 microarrays. For the D6-Pax6 analysis, dye-swapped replicate pairs of arrays were averaged, and in the case of one missing value, the single value used. Significant differences in gene expression were identified using the significance analysis of microarrays (SAM) algorithm (Version 2; one class, at least 200 permutations), using a false discovery rate (FDR) of 0.1 [65].
Dataset intersections between array datasets were performed using gene symbols. Gene symbols were retrieved using the SOURCE (http://smd-www.stanford.edu/cgi-bin/source/sourceSearch) database and the gene accession numbers provided by the array/oligonucleotide manufacturers. Network modeling was carried out using BioTapestry (www.biotapestry.org) [66].
|
10.1371/journal.pcbi.1006741 | Spatial synchronization codes from coupled rate-phase neurons | During spatial navigation, the frequency and timing of spikes from spatial neurons including place cells in hippocampus and grid cells in medial entorhinal cortex are temporally organized by continuous theta oscillations (6–11 Hz). The theta rhythm is regulated by subcortical structures including the medial septum, but it is unclear how spatial information from place cells may reciprocally organize subcortical theta-rhythmic activity. Here we recorded single-unit spiking from a constellation of subcortical and hippocampal sites to study spatial modulation of rhythmic spike timing in rats freely exploring an open environment. Our analysis revealed a novel class of neurons that we termed ‘phaser cells,’ characterized by a symmetric coupling between firing rate and spike theta-phase. Phaser cells encoded space by assigning distinct phases to allocentric isocontour levels of each cell’s spatial firing pattern. In our dataset, phaser cells were predominantly located in the lateral septum, but also the hippocampus, anteroventral thalamus, lateral hypothalamus, and nucleus accumbens. Unlike the unidirectional late-to-early phase precession of place cells, bidirectional phase modulation acted to return phaser cells to the same theta-phase along a given spatial isocontour, including cells that characteristically shifted to later phases at higher firing rates. Our dynamical models of intrinsic theta-bursting neurons demonstrated that experience-independent temporal coding mechanisms can qualitatively explain (1) the spatial rate-phase relationships of phaser cells and (2) the observed temporal segregation of phaser cells according to phase-shift direction. In open-field phaser cell simulations, competitive learning embedded phase-code entrainment maps into the weights of downstream targets, including path integration networks. Bayesian phase decoding revealed error correction capable of resetting path integration at subsecond timescales. Our findings suggest that phaser cells may instantiate a subcortical theta-rhythmic loop of spatial feedback. We outline a framework in which location-dependent synchrony reconciles internal idiothetic processes with the allothetic reference points of sensory experience.
| Spatial cognition in mammals depends on position-related activity in the hippocampus and entorhinal cortex. Hippocampal place cells and entorhinal grid cells carry distinct maps as rodents move around. The grid cell map is thought to measure angles and distances from previous locations using path integration, a strategy of internally tracking self motion. However, path integration accumulates errors and must be ‘reset’ by external sensory cues. Allowing rats to explore an open arena, we recorded spiking neurons from areas interconnected with the entorhinal cortex, including subcortical structures and the hippocampus. Many of these subcortical regions help coordinate the hippocampal theta rhythm. Thus, we looked for spatial information in theta-rhythmic spiking and discovered ‘phaser cells’ in the lateral septum, which receives dense hippocampal input. Phaser cells encoded the rat’s position by shifting spike timing in symmetry with spatial changes in firing rate. We theorized that symmetric rate-phase coupling allows downstream networks to flexibly learn spatial patterns of synchrony. Using dynamical models and simulations, we showed that phaser cells may collectively transmit a fast, oscillatory reset signal. Our findings develop a new perspective on the temporal coding of space that may help disentangle competing models of path integration and cross-species differences in navigation.
| A prominent temporal code of neural activity [1–3] is the phase precession of rodent place cell and grid cell activity relative to the septal-hippocampal theta rhythm (6–11 Hz) [4, 5], in which firing begins late in the theta cycle and advances to earlier phases as the animal moves across a spatial firing field. Theta-phase precession is strictly unidirectional, which ensures that phase unambiguously encodes the distance traveled through a place field [6]. This unidirectionality may follow from mechanisms such as neuronal adaptation that halts firing before the peak of dendritic excitation [7], place-cell network plasticity that learns an asymmetric ramp of depolarizing input through experience [8], or temporal interference between a somatic theta oscillation and a speed-tuned [4, 9] or spatial [7, 10–12] dendritic oscillation. In open-field foraging, these mechanisms may lock the phase-distance code of phase precession to trajectory details (that is, the speed, running direction, and path) of individual passes through a spatial firing field [13, 14], thus preventing a direct mapping of phase to spatial locations. It is unclear whether phase codes with different properties (for example, bidirectionality, spatial symmetry, or trajectory independence) operate in other brain areas to process spatial information.
Temporal interference models theorized that multiple velocity-controlled oscillators (VCOs) [15, 16] perform path integration to collectively synthesize the hexagonally periodic spatial firing of grid cells [17]. Electrotonic soma-dendrite coupling ruled out dendritic implementations of VCOs [18], leading to models of neuronal oscillators that project path-integrating phase codes to the grid cell network [19–21]. Experimental evidence for neuronal VCOs includes our previous report of thalamic theta-bursting neurons with the theoretically required burst-frequency tuning of direction [22] and observations of full phase precession at the periphery of grid cell fields as predicted by temporal interference but not continuous attractor networks or ramp depolarization models [14, 23–25]. Organizing VCOs into ring attractor networks provides some internal stability [26, 27], but biological variance in spike timing and local theta cycle periods limits the temporal precision of VCO phase computations [28, 29]. Likewise, continuous attractor models of grid-cell path integration accumulate position errors, even before considering sources of biological variance. In open environments that allow rotations, and particularly at low speeds, bounded network topologies cause error-inducing ‘ripples’ that perturb an otherwise flat energy landscape [30].
To counter the accumulation of position errors, path integrators must reset to the current position based on environmental cues [31, 32]. Models combining the continuous attractor and VCO frameworks have proposed resetting VCOs via descending grid cell feedback [27, 33, 34]. However, for mice in complete darkness, grid cell patterns are rapidly disrupted [35] while path integration is sufficiently preserved to maintain a global heading angle [36]. Thus, grid cell networks in different species may not have the spatial stability to support a feedback role (as in the combined attractor/oscillator models) and may not directly compute the spatial vector maintained by path integration (as in continuous attractor models).
Subcortical targets of the hippocampal formation, typically studied as regulators of the theta rhythm (cf. [37, 38]), may additionally contribute to neural computations of space. In rats, the lateral septum (LS), but not the medial septum, has revealed spatial modulation of firing rates in open environments [39, 40] that diverged with respect to hippocampal remapping over time [41]. However, LS neurons have also been reported to carry a phase code for one-dimensional (1D) tracks that precisely reflected hippocampal phase precession [42]. The degree to which LS or other spatially-modulated subcortical neurons are computationally dependent on hippocampal activity is unclear, especially in open two-dimensional (2D) environments.
In this study, we asked two questions: (1) Can spatial theta-phase codes be found in subcortical theta-rhythmic structures? (2) What computational function might such phase codes serve in downstream circuits related to spatial cognition? Our approach integrated, respectively, single-unit recordings in rats during open-field foraging, and computational modeling of spatial phase-coding networks and their downstream targets. We found a class of LS and hippocampal neurons with 2D spatial phase codes for which we analyzed the relationship between rate and phase, stability of rate and phase coding, temporal organization by theta, spatial firing patterns, and spatial vs. trajectory-related selectivity. Our analysis was consistent with an absolute, allocentric representation of space, thus we studied models of temporal coding mechanisms distinct from those hypothesized for the relative, field-centered representation of hippocampal phase precession. We suggest the theory that intrinsic neuronal and network processing of convergent hippocampal inputs form an independent and collective encoding of the animal’s current (not prospective) position. This spatial transformation may enable rapid and flexible phase-resetting of path integration.
We will first describe recordings of subcortical and hippocampal theta-modulated neurons in freely behaving rats. By setting criteria for spatial phase coding, we analyzed a subset of these neurons that we termed ‘phaser cells’ to reveal how spatial information was carried in the phase alignment of firing with the hippocampal theta oscillation observed in local field potentials (LFPs). We posit a theoretical account of the relationship between firing rate, shifts in spike phase, and ongoing theta oscillations that is supported by generalized linear models (GLMs) trained across a spatial partition of the recording arena. Lastly, we demonstrate models of intrinsic theta-bursting and spike synchronization in both artificial 1D and realistic 2D simulations of phaser cells that collectively corrected phase-position errors in downstream path-integration networks.
We obtained tetrode recordings from 8 rats as they foraged in an 80-cm cylindrical arena during sessions lasting an average of 2.1 hours. Long sessions helped to ensure sufficient sampling of phase differences across the environment. Hippocampal LFP signals were recorded from an electrode located in the hippocampal stratum oriens, referenced to animal ground. Across 110 sessions, LFPs were collected concurrently with 1,073 single-unit recordings (we use ‘recording’ to refer to a unit’s data from one session) of 671 uniquely identified neurons (some of which were observed in multiple recordings) from sites including the LS and medial septum, hippocampus, thalamus, midbrain, and other subcortical areas (Table 1; Methods).
In some recordings, units exhibited spatial tuning of firing rate as well as spatial tuning of spike phase with respect to the LFP theta oscillation. Fig 1 shows one such cell from LS that fired preferentially in the west/southwest of the arena (Fig 1A) and was moderately theta-rhythmic (index: 0.392; Fig 1A, inset, top; Methods) and theta-modulated (index: 0.288; Fig 1A, inset, bottom; Methods). Across space, the cell’s mean firing rate (‘ratemap’; Fig 1B; Methods) revealed a single-peaked firing field that broadly covered much of the arena. Surprisingly, the spatial distribution of the mean theta-phase of spikes (‘mean-phase map’; Fig 1C, left; Methods) varied in a pattern of spatial modulation that qualitatively matched the ratemap in Fig 1B. The cell fired at LFP theta peaks (0 radians) in locations corresponding to low firing rates (Fig 1C, left, green regions) and during example low-firing-rate time intervals (Fig 1D, top). Conversely, the cell fired near LFP theta troughs (−π or π radians) in locations corresponding to high firing rates (Fig 1C, left, pink regions) and example high-firing-rate intervals (Fig 1D, bottom). To quantify phase reliability during a recording, we computed at every location the mean resultant vector length (MVL) of spike phase, which varies from 0 (uniformly random) to 1 (perfectly reliable). Thus, we display the full effect of spatial modulation on spike phase with a ‘phase-vector map’ (or simply ‘phase map’) where mean phase is indicated by color hue (as in Fig 1C, left) and maximum-normalized MVL by color saturation (Fig 1C, right; Methods). The example cell had typical phase MVL around 0.2 except for a high-variance region along the westward wall (Fig 1C, right, dark pixels) and a high-reliability region >0.3 near the center of the arena (Fig 1C, right, bright pixels).
To study the characteristic phase relationships in our data, we examined spiking activity over individual traversals of the arena and whole-session spatial maps. A 15-s trajectory segment illustrates a series of bursts emitted by the example LS neuron (Fig 1E, left). The cell initially burst around theta peak in a low-rate region in the northeast of the arena, precessed to earlier phases in the high-rate region as the animal moved to the southwest, and then shifted back to later phases when the animal returned to a low-rate region (Fig 1E). Burst phase during this short trajectory was noisy, but the activity symmetrically followed the rate-phase regression line in both directions (Fig 1E, right), corresponding first to phase advance and then to phase delay. To measure this phase modulation over the 2.2-h session, we regressed the mean-phase map (Fig 1C, left) onto the ratemap (Fig 1B), revealing a negatively sloped rate-phase relationship (circular-linear correlation: n = 3,190 map pixels, estimated r ^ = - 0.836, p ^ ≈ 0; Methods) around which the cell’s spatial data was narrowly distributed (Fig 1F). For this cell, spike phase was symmetrically and bidirectionally coupled to firing rate over multiple timescales.
By inspecting our dataset for this phenomenon, we defined ‘phaser cells’ as neurons whose spike phase coded for position and was strongly coupled to firing rate. To classify phaser cell recordings, we imposed criteria on three measures of phase, rate, and space (Methods): (1) Spatial phase information Iphase quantified the spatial content of spike alignment to LFP theta oscillations as the Shannon mutual information between spike phase and position; (2) Total phase shift captured the depth of phase modulation as the regressed phase difference from the minimum to maximum rate; (3) The rate-phase correlation indicated the strength of rate-phase coupling based on a recording’s ratemap and mean-phase map.
To determine the criteria, we asked how recordings that carried spatial information in spike theta-phase differed from others. Significant phase-coding recordings (Iphase shuffled phase test, p < 0.02; n = 156 cells; S1 Fig, panel D) exhibited less variable theta-burst frequency (variance ratio, 0.624; Iphase-significance bootstrap test, p = 0.001; Methods) than non-significant recordings (n = 570 cells; S1 Fig, panel B), suggesting that phase-coding cells were more reliably periodic. Furthermore, significant phase-coding recordings exhibited more variable rate-phase correlation coefficients (variance ratio, 3.87; p = 0.001) and more broadly distributed total phase shifts (interquartile range ratio, 1.96; p = 0.001) than non-significant recordings (S1 Fig, panel E). Thus, we classified phaser cell recordings as unit-session data that met each of several criteria:
The fourth criterion ensured sufficient levels of spatial activation, at least one spike every other theta cycle, to convey rate and phase relationships. A total of 101 recordings from 5 rats satisfied the phaser cell criteria. Phaser cell recordings revealed moderate firing rates, corresponding to 1 or 2 spikes per theta cycle in preferred regions, and similar theta rhythmicity to other significant phase-coding recordings (S2 Fig, panel A). By analyzing which recordings followed the same neuron across multiple sessions (Methods), we determined that 69 unique phaser cells were observed by the 101 recordings: 50 phaser cells were located in the lateral septum, 15 in the hippocampal formation, and 4 in other subcortical structures (Table 1).
The validity of the above criteria for phaser cells depended on whether they selected a meaningful subset of our data. Fig 2A visualizes the measures tested by the first two criteria (Iphase and total phase shift) with respect to their thresholds; the third measure (rate-phase coupling strength) is indicated by the size of the plot markers. In Fig 2A, significant phase-coding recordings (n = 233) are shown with individual data points, the distribution of non-significant recordings (n = 840) is represented by contours in the background, and phaser cell criteria (1) and (2) above are overlaid as red lines that cross out the region excluded by the criteria. Non-significant recordings (Fig 2A, contours) displayed a wide range of Iphase values that failed to achieve statistical significance (S1 Fig, panel D) and no relationship with total phase shifts that were narrowly distributed around zero (S1 Fig, panel E, right). However, significant phase-coding recordings (Fig 2A, circles) fell into roughly three clusters: (1) low Iphase, total phase shift near zero, and minimal rate-phase coupling; (2) moderate Iphase, large positive phase shifts, and moderate coupling; (3) high Iphase, large negative phase shifts, and strong coupling. The first cluster was excluded, and the latter two clusters were selected as phaser cell recordings. Due to the striking division of the direction of phase shifts between the selected clusters, we labeled them as ‘positive’ and ‘negative’ subtypes. That is, negative phaser cells advanced to earlier phases, like hippocampal phase precession, and positive phaser cells delayed to later phases, unlike previously described spatial phase codes.
To verify that differences in the direction of phase shifts were not artifacts of the recording configuration, we inspected our dataset for colocation, stability, and simultaneous observation of the two subtypes. Phaser cells were predominantly recorded from LS (Table 1; S3 Fig, panel A). Two-thirds of phaser cells (48/69) were negative and one-third (24/69) were positive. For 19 phaser cells with multiple recordings, all but 3 preserved the sign of phase shift across their phaser-classified recordings (S2 Fig, panel B, right). In some cases, negative and positive phaser cells were recorded simultaneously against the same LFP reference electrode and/or observed on the same tetrode. These observations, together with the fact that the LFP signal was always recorded from the hippocampal stratum oriens, indicate that the direction of rate-phase coupling was a stable property of individual phaser cells and not an artifact of variations in LFP signal polarity.
To quantify phaser cell accuracy and reliability, we examined, respectively, a measure of spatial uncertainty and the spatial distribution of spike-phase MVL. We computed spatial uncertainty as R / 2 I phase for arena radius R = 40 cm. Increasing magnitude of total phase shift was associated with lower spatial uncertainty for negative (n = 65 recordings; mean ± s.e.m., 33.5 ± 0.378 cm; linear regression, r = 0.363, p = 0.00292) and positive (n = 36; 35.4 ± 0.349 cm; r = −0.441, p = 0.00707) phaser cells (Fig 2B). Across spatial locations, MVL was distributed from nearly zero up to a typical maximum value of 0.414 (median, n = 101 recordings; Fig 2C). In order to statistically test for differences between subtypes, we averaged values across recordings for unique cells with multiple recordings. Negative phaser cells demonstrated both lower spatial uncertainty (n = 48/24 negative/positive cells; post hoc Welch’s t = −2.32, p = 0.0236) and higher phase-code reliability (mean MVL; t = 2.68, p = 0.010) than positive phaser cells. Thus, phaser cells exhibited spatial accuracy on the order of body length based on a reliable mapping of spike phase to position in certain locations.
If phaser cells contribute to navigation or other spatial functions, then they must stably reflect a given context or environment. Cell-specific spatial modulation and rate-phase coupling should be preserved over both long experiences and multiple days. To analyze spatial stability of phase coding in phaser cells, we compared early vs. late portions (<1 h) of each recording to a baseline of pair-wise measurements between different cells (Methods). For spatial stability, the distributions of spatial correlations between ratemaps revealed significant similarity above baseline across the multiple-hour recording sessions (median, 0.502; within-cell (n = 101) vs. between-cell (n = 9, 986) early-late pairs; Kolmogorov-Smirnov D = 0.694, p = 2.07e−43; Fig 2D, left). For phase-coding stability, changes in total phase shift were distributed narrowly around zero, significantly lower than baseline (1.07 radians; D = 0.371, p = 1.00e−12; Fig 2D, right). Likewise, for the 19 phaser cells with multiple recordings, spatial correlations between different recording days were significantly higher than baseline (0.345; within-cell (n = 57) vs. across-cell (n = 4, 986) day pairs; D = 0.431, p = 7.52e−10; Fig 2E, left) and changes in total phase shift were distributed close to zero, significantly lower than baseline (1.30 radians; D = 0.399, p = 1.66e−8; Fig 2E, right). Further, all but 3 of these phaser cells maintained similar Iphase values and total phase shifts across days (S2 Fig, panel B), suggesting a global stability of the phase code beyond the pair-wise stability implied by Fig 2E.
The stability of Iphase and total phase shift is necessary for phase-code stability, but those are spatially averaged measurements and relative phase shifts remain constant even if phase-code angles systematically drifted. Thus, we addressed the relationship between specific locations and the magnitude of changes in mean-phase angles. We calculated absolute phase differences between the early and late mean-phase maps from the analyses in Fig 2D. To relate these phase differences to spatial variation of phase reliability (Fig 2C), we display them according to spike-phase MVL. Low/high MVL locations would be expected to show larger/smaller phase differences over time. Fig 2F shows MVL and absolute early-late phase differences for the LS cell from Fig 1; the wedge shape reflects the expected relationship, but the placement of the bulk of the data distribution revealed that typical MVL values coincided with phase differences of <π/4 radians (that is, 1/8th of a theta cycle or ∼17 ms). Averaging across phaser cell recordings revealed a similar pattern in which the region of highest spatial density corresponded to absolute phase-code changes of <1/8th of a theta cycle (Fig 2G). As in Fig 2B+2C, positive phaser cells demonstrated weaker phase-coding than negative phaser cells, as shown by the relatively higher density of the ‘tail’ leading up to maximal phase difference (|Δ| = π) at low MVL (Fig 2G, right). Thus, phase reliability (Fig 2C) implied location-dependent phase-code stability over multiple hours (Fig 2G). The spatial and phase-coding stability of phaser cells across hours and days was consistent with functional contributions to the spatial computations of the hippocampal formation.
We asked what theoretical mechanism could support our observations of the spatial phase code carried by phaser cells. We considered the crucial feature that spatial data points, such as the conditional spike-phase distributions in Fig 1F, were tightly coupled to the rate-phase regression. Strong rate-phase coupling suggested that the rate-phase relationship was maintained across spatial locations and that rate and phase did not systematically diverge over short or long timescales. We surmised that, on average, rate and phase deflected together on approaches to a preferred location (that is, a high mean firing-rate region), and then symmetrically retraced those deflections on leaving the preferred location (Fig 3A). Thus, we theorized that the phaser cell code was a spatially homogeneous coupling of rate and phase that was symmetric and, because they deflect and retrace, bidirectional.
In contrast, Souza & Tort (2017) [43] examined hippocampal place-cell theta-phase at low firing rates and revealed a distinct angle-shaped rate-phase relationship across place fields. The resulting curve (adapted in Fig 3B) reflects the combination of two effects that progress from entry to exit of hippocampal place fields: (1) the strict unidirectionality of spike theta-phase precession [4], and (2) the single-peaked rise and fall of firing rate, which may be symmetric or skewed with respect to the field center [12, 44]. To reconcile these differences, we suggest that symmetric, bidirectional phaser cell coding (Fig 3A) and asymmetric, unidirectional hippocampal phase precession (Fig 3B) reflect experience-independent vs. experience-dependent models of temporal coding, respectively. Mehta et al. (2002) [8] proposed that theta-rhythmic inhibition combines with spatially asymmetric input learned from the place-cell network to monotonically shift spike phase across place fields. However, absent learning, that mechanism generates a symmetric rate-phase relationship mediated by the rise and fall of external input (Fig 3C). Thus, theta-rhythmic inhibition combined with depolarization by external inputs may explain the rate-phase relationship of negative phaser cells (Figs 1F and 3A). As noted in Mehta et al. (2002) [8], coupling phase to rate precludes a precise mapping between phase and specific locations within a place field. Instead, a rate-coupled phase signal in a 2D environment is restricted to encoding isocontours of the depolarizing spatial input (Fig 3D; Discussion).
Our observations of positive phaser cells, which modulated timing in the opposite direction to negative phaser cells, presented a conundrum. In models described below, we suggest a network mechanism to account for this difference, but the key prediction is that positive modulation requires theta-rhythmic excitation instead of inhibition. A consequence of theta excitation is that positive cells would fire at theta peak (0 radians) at low firing rates, and then delay to later phases at higher rates. Negative phaser cells based on a symmetric ramp mechanism (Fig 3C) would fire following the minimal inhibition of the theta trough (−π or π radians) at low firing rates, and then advance to earlier phases at higher rates. This distinction implies a temporal segregation of phaser cell activity. To assess this temporal organization, we show rate-phase regressions for every phaser cell recording according to subtype (Fig 3E). Negative and positive phaser cells fired during the rising phase [−π, 0] at low firing rates, and, with increasing firing rate, followed opposing paths to the falling phase [0, π], thus complementarily spanning the theta cycle (Fig 3E). Positive phaser cell activity clustered before theta peak at low rates (Fig 3E) as predicted by theta excitation and a high threshold. Distributions of typical spike phases, computed as the spatial average of mean-phase maps to avoid the sampling biases of time averages, show that the subtypes were segregated by theta phase: negative/positive phaser cells typically fired at theta trough/peak (Fig 3F). Thus, temporal segregation by subtype may reflect underlying differences in theta drive.
Negative phaser cell ratemaps revealed diverse spatial representations including place-like fields, broad gradient-like fields, and boundary (including on/off) responses along the arena wall (Fig 4A; recordings #444 and #768 produced remarkably similar rate and phase maps from different rats). Maximal firing rates (Fig 4A, top) corresponded to pre-theta-trough timing (Fig 4A, middle, blue/pink). Conditional spike-phase distributions (Fig 4A, bottom) revealed a tendency for phase modulation to halt after approximately one-half theta cycle, perhaps indicating a minimum latency to spike following theta-peak inhibition; this nonlinearity means that some rate-phase regression lines (Fig 3E) overestimated the total phase shifts. Positive phaser cells likewise showed diverse spatial modulation, but the responses were more subtle, involving higher baseline firing rates and heterogeneous compositions of boundary-like and place-like selectivity (Fig 4B, top). Maximal firing rates typically mapped to post-theta-peak timing (Fig 4B, middle, green/blue) and the rate-phase relationships were weaker (n = 24 cells; median, rate-phase correlation r ^ = 0 . 42; Fig 4B, bottom) than those of negative phaser cells (n = 48; r ^ = - 0 . 54; Fig 4A, bottom; absolute values, post hoc Welch’s t = 2.053, p = 0.0442). Thus, subtype differences in patterns of spatial modulation reinforced our analysis showing higher spatial uncertainty and weaker phase stability in positive phaser cell recordings (Fig 2B+2C and 2G).
To quantify spatial modulation, we calculated spatial rate information Irate using a standard measure of position coding in place cells [45] and determined its statistical significance in phaser cell recordings with a spike-train shift test (criterion p < 0.02; Methods); 47/48 negative and 24/24 positive phaser cells attained significance. As expected from prior analyses, negative phaser cell spikes carried significantly higher Irate (n = 47 significant cells, p < 0.02; 0.381 ± 0.06 bits/spike, mean ± s.e.m.) than positive phaser cell spikes (n = 24, 0.111 ± 0.048; log values, post hoc Welch’s t = −3.92, p = 0.0002). The least-squares optimized slope between Irate and Iphase was 0.640 (n = 101 recordings; S3 Fig, panel B, left), indicating that spike phase contributed substantial spatial information (∼56.3%) in excess of firing rate alone. Most of the phaser cell recordings (10/16) with the highest Irate values (>0.6 bits/spike) were from hippocampal sites (S3 Fig, panel B, left) and most of those (9/10) were negative phaser cells, consistent with place cells that may have reflected phaser cell activity (Discussion). However, our hippocampal sample was too small to draw clear conclusions. Thus, negative and positive phaser cells may represent diverse spatiotemporal relationships resulting from circuits combining theta-rhythmic inhibition or excitation with varied patterns of spatial drive.
Our thesis that phaser cells map spike phase to spatial isocontours (Fig 3D) requires that spiking is predominantly driven by allocentric spatial factors (that is, external cues in a world-centered reference frame). To compare allocentric spatial modulation with other factors, we calculated the spike information content of speed (an idiothetic self-motion signal) and movement direction (an allocentric, but not spatial, signal; Methods). In contrast to the Irate comparison, the least-squares optimized slopes between Iphase and directional (0.086; n = 101 recordings) or speed information (0.023; S3 Fig, panel B) indicated minimal coding overlap between Iphase and other trajectory-based factors. However, it is possible that the spatial modulation apparent in ratemaps (Fig 4) was a spurious by-product of trajectory-based factors and biased spatial sampling of the arena. Firing-rate modulation indices (Methods) for direction (median, 0.379; n = 101 recordings) and speed (0.318; S3 Fig, panel C) were suggestive of possible trajectory dependence. Such a confound can result from directionally biased visits to particular locations for which a recorded cell happened to have a similar directional preference. For example, a cell responding to clockwise movement around the arena may produce a spatial ‘wall’ representation if the rat only moved clockwise when contacting the wall.
To isolate spatial-behavioral confounds, we studied a Poisson-distributed generalized linear model (GLM) of spatial (allocentric) and trajectory-based (idiothetic speed, and allocentric non-spatial direction) variables. GLMs have been shown to learn independent spatial and directional contributions to firing that avoid trajectory-driven biases [46, 47]. To capture inhomogeneous changes in spatial or trajectory-dependent selectivity, we fitted GLMs independently to every phaser cell recording for data restricted to sections of a 3 × 3 spatial grid spanning the arena (Methods). The model was trained to predict the spike count for any 300-ms interval i Y ^ i = β ^ 0 + β ^ L L i + β ^ Q Q i + β ^ W W i + β ^ S S i + β ^ D D i (1)
where L and Q are linear and quadratic spatial variables, W is a sigmoidal wall-proximity signal, S is linear speed, and D is movement direction. L, Q, and W are purely spatial whereas S and D capture the rat’s trajectory as a velocity vector. Thus, we termed this spatial family of GLMs the ‘LQW-SD’ model. To train LQW-SD, we standardized the position and trajectory data from our recordings, but several properties of the data needed to be addressed: (1) statistical dependence among the predictors contributed to an ill-posed problem; (2) spatial predictors had more reliable short-timescale correlations than the trajectory-based predictors; and (3) variable data density across spatial grid segments reduced the validity of model comparisons across the arena. To mitigate these issues, we imposed constraints on model coefficients by training LQW-SD as a ridge regression with ℓ2- regularization [48]. Further, to maximally expose the spatially inhomogeneous directionality that could have produced behavioral confounds, we chose the regularization penalty that optimized the trade-off between maximizing model directionality and minimizing spike-prediction errors (S4 Fig, panel B+C; Eq (14); Methods). While we did not cross-validate spike-count predictions from the model, our analysis goal was not prediction but to statistically isolate consistent drivers of phaser cell spiking versus spurious factors that may have arisen due to behavioral biases. However, training the model independently within the 3 × 3 grid sections effectively performed a 9-fold cross-validation in space.
We asked whether phaser cell recordings demonstrated directional selectivity that could produce spurious spatial modulation. To quantify directionality, we computed a directional homogeneity index (DHI) on [0, 1] measuring alignment of the 9 βD vectors (Eq (1)) across the 3 × 3 grid; additionally, we computed a directional strength index (DSI) on [0, 1] measuring the magnitude of βD relative to the other predictors (Methods). The DHI of phaser cells (median, 0.265; n = 69 unique cells with at least one phaser-classified recording) revealed higher homogeneity than nonphaser cells (0.213; n = 602; post hoc Mann-Whitney U = 15, 423, p = 0.0005). The DSI of phaser cells (median, 0.0248) and nonphaser cells (0.0127) indicated low overall directionality (U = 15, 268, p = 0.0003), but it was more widely distributed for nonphaser cells (range, [0, 0.199]) than phaser cells ([0.003, 0.105]). Thus, phaser cells excluded both homogeneous (high DHI, high DSI) and inhomogeneous (low DHI, high DSI) directionality.
Our analysis was predicated on the ability of the model to explain firing patterns. To verify that LQW-SD could reproduce patterns of spatial modulation, we generated spike-count predictions across the 3 × 3 grid to reconstruct firing ratemaps (Methods). Quantifying accuracy as the vector cosine similarity between ratemaps, we found phaser cells (median, 0.986; n = 69 unique cells with at least one phaser-classified recording) and nonphaser cells (0.908; n = 602) to have highly accurate reconstructions (post hoc Mann-Whitney U = 16, 960, p = 0.012). Actual and LQW-SD-predicted ratemaps are shown in Fig 5A for the negative phaser cells in Fig 4A with overlaid arrows representing the modeled directionality (βD) of each grid section. To verify that LQW-SD also captured strong directional (high DSI) cells accurately, examples of homogeneous (high DHI) and inhomogeneous (low DHI) directionality are shown in S5 Fig. Thus, LQW-SD provided a high-fidelity account of single-unit firing in our dataset, including spatial and directional cells.
What does the LQW-SD model reveal about spatial vs. trajectory-based predictors? Like DSI for directionality, we computed the relative strength of each model variable (Eq (15); Methods). Box plots (Fig 5B) show the distribution of variable weights for phaser cells (n = 69 unique cells with at least one phaser-classified recording) and nonphaser cells (n = 602). Both cell types had similar central tendencies with nonphaser cells exhibiting wider ranges of variable strengths. The second-order spatial variables (L and Q) overwhelmed the wall and trajectory variables, constituting approximately 30% and 60% of the model weight, respectively. Wall/boundary cells were (by inspection) a small number within the dataset, but we considered that the trajectory-based factors (S and D) might be non-normally distributed, leading to artificially low coefficients. Thus, we computed the importance of model variables by their maximal contribution to predictions over the length of the recording. For variable X, we computed its maximal contribution
Contribution ( X ) = max i | β ^ X X i | (2)
across time intervals i and sum-normalized the variables (Methods). The contribution profile (Fig 5C) was also dominated by L and Q, but the W, S, and D contributions were enhanced relative to the strength profile in Fig 5B. Wall and direction variables each constituted ∼8% of the total contribution and nonphaser cells revealed a wide range of speed contributions (Fig 5C, S, gray) consistent with the availability of speed signals throughout space-related brain areas [49, 50]. Sorted recording data confirmed this pattern by showing an inverse relationship between spatial and speed-based contributions for phaser cells (S6 Fig); this relationship held for both negative and positive phaser cells (S6 Fig, panel E). Thus, LQW-SD revealed a trade-off between allocentric spatial coding and idiothetic speed modulation, and that phaser cells were overwhelmingly spatial, not directional.
To gain insight into the possible mechanisms and functions of phaser cell populations, we developed computational models based on minimal dynamics for intrinsic processing of spatial and theta-rhythmic inputs. Crucially, our models assumed that postsynaptic averaging of convergent hippocampal-LS projections produces input to phaser cells that is independent of hippocampus-specific coding (Discussion). Our modeling approach balanced two goals: (1) qualitatively capture salient neurocomputational features of the data, and (2) minimize degrees-of-freedom to avoid model complexity and parameter fine-tuning. Our neuron and network models were broadly tuned to recapitulate several phenomena: (1) theta-bursting rhythmicity (Fig 1A+1D; S2 Fig, panel A, right), (2) symmetric and bidirectional rate-phase coupling (Figs 1E+1F and 3A; S1 Fig, panel E, left), (3) negative/positive phase-shift subtypes (Figs 2A–2C, 3E and 4), (4) temporal segregation of subtypes (Figs 3E+3F and 4), and (5) allocentric phase coding of spatial isocontours (Figs 1B+1C and 2E; S2 Fig, panel B, right; Figs 3D and 5B+5C). Thus, to ensure rhythmicity and realistic spike timing, we based our neuron models on two-variable dynamical systems (Eq (5); Methods) featuring intrinsic bursting dynamics and spike initiation tuned to the activity of hippocampal low-threshold bursters [51, p. 310].
To outline the computational role of phaser cells, our simulations focused on feedforward models in which phasers project to targets that ‘read out’ the phaser cell code. (We will refer to model phaser units as ‘phasers,’ ‘negative phasers,’ or ‘positive phasers’ to distinguish them from our observed ‘phaser cells.’) In the following sections, we present model simulations in several stages: (1) single-neuron phaser models with 1D external inputs, (2) a demonstration model of a small phaser network with artificial 1D spatial inputs and a downstream target cell, and (3) a realistic model of a large phaser network with 2D spatial inputs and several downstream target networks.
Model negative phasers combined inhibitory theta input and excitatory external input (Eq (6)) with parameters (Tables 2 and 3) that enabled theta-bursting (Methods). Fig 6 shows phaser simulations in which the external input varied up and down over its full range (Eq (8)). For low levels of excitatory input, the negative phaser (Fig 6A+6B, Low1 and Low2) emitted single spikes near theta peak every few theta cycles. For high excitatory input (Fig 6A+6B, High), the negative phaser burst with spike triplets near the theta trough on alternating theta cycles. This cycle-skipping rhythmicity is reminiscent of observations in medial entorhinal cortex and the head direction system [52, 53], but this model has no relationship to those phenomena: cycle skipping was a side-effect of the particular theta-bursting parameters (Table 2) that we chose to qualitatively match phaser cell characteristics, which do not include skipping. (The skipped cycles entailed that the resultant spike phase signal was perhaps weaker than if the units had fired every cycle.) Expanded time intervals (Fig 6B) clearly show that the negative phaser shifted to earlier phases of the reference theta wave at high input levels. The model’s rate-phase correlation (n = 399/512 nonzero input-level bins; r ^ = - 0 . 809, p ^ ≈ 0; Fig 6D) revealed strong, consistent phase modulation from peak (0 radians) to trough (−π). That is, spike-phase advanced during rising inputs (Fig 6A; 0–10 s) and then delayed to later timing during falling inputs (10–20 s). The simulated rate-phase coupling is symmetric and bidirectional as predicted (Fig 3C) and it advances to the theta trough as observed for negative phaser cells (Fig 3E).
To model positive phaser cells, we proposed a circuit mechanism whereby a bursting unit driven by excitatory theta input is suppressed by a negative phaser and does not directly receive spatial inputs. We modeled the feedforward inhibition as incrementing a slow 100-ms inhibitory conductance in the positive phaser for each presynaptic spike from the negative phaser (Table 3; Eq (9); Methods). The positive phaser burst at the peak of every theta cycle when disinhibited by low external input to its presynaptic negative phaser (Fig 6A+6B, Low1 and Low2). As the external input rose and fell (Fig 6A), the negative and positive phasers fired in complementary patterns: low/high input silenced the negative/positive phasers (Fig 6C). The model’s rate-phase correlation was indeed positive (n = 351/512 nonzero input-level bins; r ^ = 0 . 705, p ^ ≈ 0; Fig 6E), but weaker and with a shallower phase modulation than both the negative phaser (total phase shift, 0.654 vs. −2.44 radians; Fig 6D) and the positive phaser cell data (∼83% of the low end of the observed range). Positive phaser weakness in the model was commensurate with the higher spatial uncertainty (Fig 2B), lower phase reliability (Fig 2C), and lower phase-code stability (Fig 2G) of positive phaser cells in our dataset. Crucially, negative and positive phasers were temporally segregated according to rate-phase coupling direction (Fig 6D+6E) as in the phaser cell recordings (Fig 3E). Thus, a simple connectivity pattern between theta-bursting models qualitatively recapitulated phaser cell temporal organization.
To demonstrate how a downstream target may learn to decode phaser cells, we constructed an artificial 1D spatial paradigm with which to study a model network of 128 negative and 128 positive phasers. The top half of Fig 7 (panels A+B) presents the phaser network and its outputs, and the bottom half of Fig 7 (panels C-G) presents the inputs and outputs of a target neuron model. To emulate the spatial diversity of phaser cells (Fig 4), we created two sets of spatial inputs that each drive one-half of the phaser network: (1) 64 place-like tuning functions (Fig 7A, spatial information flows from the middle to the top of the network diagram), and (2) 64 inverted place-like tuning functions that we termed ‘notch’ functions (Fig 7A, spatial information flows from the middle to the bottom of the network diagram; Methods). A notch function is equivalent to a corresponding place function that has been vertically flipped about its middle so that it is active everywhere except for one location; it is a spatial function and not a frequency filter as the term is used in other domains. Example joint space-phase distributions show the spatiotemporal firing patterns (Fig 7B) that were expressed by the phasers at the spatial mid-point of the network (position 0.5; Fig 7A, highlighted phasers). The four network layers represent the possible combinations of spatial input type (place vs. notch) and phaser subtype (negative vs. positive). These space-phase patterns (Fig 7B), replicated at each of 64 positions across the 1D space, were available as presynaptic inputs to a downstream theta-bursting neuron (‘target burster’; Fig 7B, right). We next demonstrate how this downstream target can utilize phaser activity to learn a spatial phase code.
To demonstrate how phaser inputs can entrain a downstream target, we devised an artificial 1D phase code consisting of two modes: theta-trough timing to the left (position 0) and theta-peak timing to the right (position 1) (Fig 7C). This code associated opposite ends of the 1D space with opposing theta phases. We tuned the target burster model (Table 4; Eq (11); Methods) to emit spike doublets without cycle skipping (Fig 7C, inset; S7 Fig, panel A). Its intrinsic burst rate approximately matched the reference theta frequency (7.5 Hz) of our simulations, but a small deviation caused the burst phase to slowly precess over time (S7 Fig, panel B). That is, the target burster was an intrinsic theta generator independent of other model elements. To amplify its independence, we injected a noisy current (Table 4; Eq (11); S7 Fig, panel C) that caused its burst phase to randomly drift (0.924 angular s.d. over 30 s, n = 36 trials; S7 Fig, panel D). To determine feedforward weights from phaser network inputs, we computed the vector cosine similarity between the space-phase distributions of each phaser (as in Fig 7B) and the supervised phase code (Fig 7C). Inputs with the highest similarity were selected by k-winners-take-all (kWTA; k = 25 negative + 25 positive phasers; Table 4; Methods). The resulting weights showed that the theta-trough mode to the left was supported by place/negative phasers, the middle part of the space was not strongly represented, and the theta-peak mode to the right was supported by notch/positive phasers (Fig 7D). The total weighted phaser-network input revealed a qualitative match to the supervised phase code (Fig 7C).
In a 1-h simulation without injected noise, the target burster’s phase revealed distinct stereotyped phase trajectories for movement to the right or the left (Fig 7F, arrows). Importantly, phaser network activity was not directional (Fig 7B); however, the target burster was directional because its phaser input was effectively released in the middle part of the space (Fig 7D). Thus, in the middle, the target preserved its most recently entrained phase until the simulated spatial trajectory approached the other phase mode. This entrainment dynamic was visibly preserved in a simulation with injected noise (Fig 7G): moving left caused a smooth phase advance to the theta-trough mode, while moving right slowly delayed toward the theta-peak mode until discontinuously jumping ahead of it. The vertical extent of the burst-timing channels at either side (∼π/2; Fig 7F+7G) indicated the degree of phase misalignment allowed by this competitive phaser-target burst-synchronization mechanism. While the entrainment did not act perfectly, it prevented the target burster from substantially drifting from the phase code across a range of parameters (S8 Fig). Thus, a phaser network robustly entrained a noisy target cell to a phase code in an artificial 1D space.
To model realistic phaser cell activity, we drove our model phasers (Eqs (5)–(10); Tables 2 and 3; Fig 6) with spatial input functions sampled from a generative model of the open-field spatial modulation of phaser cells (S10 Fig, panel A). The generative sampling model was based on the ‘LQW’ model (Eq (3)), a reduced LQW-SD model that was trained on full recording data (that is, a 1 × 1 grid instead of the 3 × 3 grid) without the trajectory-based variables S (speed) or D (direction). The result is a seamless model of allocentric spatial selectivity
F LQW ( x ( t ) ) = β ^ 0 + β ^ L L ( x ( t ) ) + β ^ Q Q ( x ( t ) ) + β ^ W W ( x ( t ) ) (3)
for any trajectory x(t) inside the 80-cm recording arena. In the same way that LQW-SD was optimized to expose directionality (Eq (14); Methods), LQW was optimized to expose wall signals (S4 Fig, panel A) to ensure that the less prevalent boundary/wall responses were captured. The generative model processed and randomized LQW representations to synthesize novel patterns of spatial modulation (S10 Fig, panel A) for negative phasers (as only negative phasers received direct spatial inputs). Given a sampled input function F LQW *, the external input current followed
I ext ( t ) = g e F LQW * ( x ( t ) ) (4)
with excitatory input gain ge (Eq (8)) and other parameters unchanged (Table 3). We simulated 1,000 pairs of negative (S9 Fig, panel A) and positive (S9 Fig, panel B) phasers, in which the negative phaser inhibited the positive (Eq (9); Fig 6). Simulated phasers expressed place-like, gradient-like, and boundary/wall-like responses (S9 Fig) similar to our phaser cell recordings (Fig 4). We next demonstrate how this realistic phaser network can entrain a downstream target cell.
To demonstrate realistic phaser entrainment of a single cell, we simulated a target burster neuron using an actual behavioral trajectory (1 h from Fig 1A). Without phaser input, the target’s bursting phase map illustrated the baseline spatial modulation (Fig 8A; maximum MVL, 0.486) to be expected from a randomly drifting oscillator (S7 Fig, panel D). We devised spatial phase codes representing oscillatory path integration (Discussion) that spanned the arena and the theta cycle. Two such codes with different phase offsets represented path integration of movement in the 45° direction at the scale of the arena (Fig 8B). As in Fig 7D, we calculated the 2D kWTA weights (k = 35 negative + 35 positive phasers; Table 4) based on spatial phase-tuning similarity between phasers and the supervised phase code. As in Fig 7E, the total weighted phaser-network inputs to the target burster revealed a spatial phase pattern that approximated the desired phase code (Fig 8C). This input pattern comprised a post-theta-peak band (π/2; Fig 8C, top, blue), due to positive phasers, alternating with a theta-trough band (π; Fig 8C, bottom, pink), due to negative phasers; the location of these bands (Fig 8C) tracked corresponding phase stripes in the phase codes (Fig 8B). With phaser input, the target’s phase maps revealed two broad modes of high burst-phase reliability (Fig 8D; bright colors; maximum MVL, 0.994, top; 0.973, bottom) reflecting location-dependent phaser entrainment. The division between the post-theta-peak and theta-trough modes was visibly sharper (Fig 8D, dark stripe) than in the phaser input itself (Fig 8C), suggesting an attractor-like nonlinearity in the input-output phase transformation of phaser-target burst-synchronization. Further, the two entrainment modes were expanded and shifted in the 45° direction relative to phaser input (Fig 8C+8D), analogous to the directionality and delayed onset of entrained bursting observed in the 1D phase trajectories (Fig 7F+7G). Thus, for a single target cell, realistic phasers controlled the spatial distribution of burst timing, but the limited spatial frequency and phase-modulation depth of phaser activity (especially positive phasers, Fig 6E) dynamically constrained the phase-code output.
To overcome the constrained output of single target cells, we asked whether a downstream network of multiple cells with phaser inputs would provide a stronger position signal. We considered target networks to be simple collections of target burster units (Eq (11); Table 4); each unit had its own set of competitive synapses carrying input from the 2D phaser network. We constructed three target collections of 64 units (Fig 8E; S10 Fig, panel B). By analogy with oscillatory ring-attractor models of path integration [26, 27], we created the ‘Ring’ collection with identical preferred directions but a full range [0, 2π] of phase offsets (Fig 8E, top). Because a single ring network is directionally biased, we expected that it would not support a clear open-field position signal on its own. The remaining two collections were constructed with a full range [0, 2π] of preferred directions but identical phase offsets across units (Fig 8E, bottom). These collections, ‘Phase 1’ and ‘Phase 2,’ were equivalent to taking a single-phase slice across a population of ring attractor networks (S10 Fig, panel B). For each collection, every unit’s phase code (Fig 8E) was learned via kWTA competition and simulated with a 600-s behavioral trajectory. Due to the feedforward phaser-target connectivity, all units were simultaneously entrained by the same open-field phaser network (as in Fig 8C+8D). The phaser input and unit output maps are shown as movies for the Ring (S1 Movie), Phase 1 (S2 Movie), and Phase 2 (S3 Movie) collections. Thus, realistic 2D phasers enabled functionally flexible phase-code entrainment of many downstream targets.
To uncover the collective position signal in these collections, we applied the method of Bayesian spike-count decoding of position [54] to the phase domain (Eq (13)) to infer estimated trajectories from simulated burst timing (Methods). If this position signal were to support the resetting of path integration, then it should be quantified in terms of position-error correction. Example 6-s trajectories with maximum a posteriori (MAP) estimates of position revealed that, as expected, the Ring network poorly tracked the trajectory (Fig 8F, top left), but the Phase 1 and Phase 2 collections more closely approximated the trajectory’s position and shape (Fig 8F, top right and bottom). To quantify error correction, we decoded a benchmark trajectory across collections and bootstrap unit samples (Methods). The mean squared error (MSE), based on the distance between actual positions and MAP estimates (Methods), showed that the Ring network consistently performed poorly, but the Phase 1 and Phase 2 collections’ performance substantially improved by collectively decoding larger numbers of units up to the total of 64 (Fig 8G). Phase 1, Phase 2, and the combination of all collections exhibited average decoding errors of 8.25, 11.6, and 8.70 cm, respectively.
To be useful, phase resets should occur quickly. To measure the timescale of error-correction in phaser-entrained targets, we computed temporal auto-correlations of decoding errors for the benchmark trajectory (S10 Fig, panel C). We quantified the typical timescale of error-correction as the correlation’s half-width at half-maximum (HWHM; Methods). Across target collections, the HWHM timescale (Fig 8H) revealed subsecond correction in the Phase 1 (0.667 s) and Phase 2 (0.267 s) collections and 1-second correction in the combined collection (1.067 s). In our framework, correcting path integration errors depended on populations of ring networks (as represented by the Phase collections) or other structures with diverse preferred directions. As expected, a single ring network (or other directionally homogeneous integrator) would be insufficient to support a 2D position signal. Further, our target units were not performing path integration: they were noisy, intrinsic theta-bursters. Thus, error-correction performance in our models provided a lower bound: presumably, a path-integrating target would have fewer errors to correct than randomly drifting oscillators.
We recorded single-units from freely exploring rats in septal, hippocampal, thalamic, midbrain, and other brain areas and found neurons in LS and the hippocampus whose spiking theta-phase was symmetrically and bidirectionally coupled to spatial modulations of firing rate. Tight rate-phase coupling entailed that spike phase mapped to isocontour levels of spatial inputs. We theorized that phaser cells serve to transform spatial information into the temporal-phase domain for downstream spatial computations. Phaser cells exhibited negative (phase advance) or positive (phase delay) modulation for increasing firing rates. Temporal segregation of negative and positive phaser cell activity was consistent with experience-independent phase-coding mechanisms and our models’ assumptions of inhibitory/excitatory theta input to negative/positive phaser cells. We trained space–trajectory GLMs to verify that phaser cell spiking was overwhelmingly driven by allocentric spatial factors and not spatially inhomogeneous modulation by speed or movement direction. We asked what mechanisms could explain the spatiotemporal organization of phaser cells and what functions they could serve in LS output targets. We demonstrated minimal circuit models of bursting neurons that qualitatively accounted for our main observations. In artificial 1D and realistic 2D open-field spatial simulations, we showed that phaser networks collectively entrained target neurons and networks to spatial phase codes using a competitive learning rule. Moreover, Bayesian position decoding of simulated burst phase in phaser-entrained targets revealed a strong, error-correcting spatial signal organized by location-dependent synchrony. Our results suggest a framework in which LS spatial phase representations enable flexible computations of spatial synchrony in subcortical networks interconnected with the hippocampal formation.
Hippocampal place fields [55] were studied extensively as a spatial firing-rate code prior to the characterization of spike theta-phase precession [4, 6, 56]. Theoretical models and in vivo manipulations have explored how interacting oscillations, ramp currents, or intrinsic dynamics may account for the link between phase precession and firing rate [7–12]. An analysis of pooled hippocampal activity highlighted the asymmetry of phase precession (Fig 3B) by finding clear theta coupling before the animal entered the classical rate-based place field [43]. This extended oscillatory coupling may reflect a critical role for phase precession in compressing place cell activity [57] into the timescale of synaptic plasticity [58, 59]. If phase precession is primarily involved in the internal temporal organization of place cell activity, then spatial and theta-rhythmic input from the hippocampus may be transformed for other functions by other brain areas.
Our analysis characterized the rate-coupled phase code of phaser cells as distinct from hippocampal phase precession. Most phaser cells in our dataset were located in LS (Table 1), a primary subcortical target of dense, convergent hippocampal efferents [42, 60] that had previously been shown to carry a degraded spatial rate code [39–41]. Tingley & Buzsáki (2018) [42] reported that many LS neurons recorded during track running carried spatial phase codes that were similar to phase precession except for rate independence and larger spatial extents than typical place fields. Their analysis [42] indicated that the LS phase code depended specifically on hippocampal phase precession coordinating theta sequences in CA3 and CA1 inputs. However, this leaves open the questions of what LS phase codes in the open field look like and whether previously described LS rate-coding neurons also carry a phase code. Examining a single open-field behavioral condition, we found that 15.6% (50/321) of LS neurons yielded phaser-classified recordings according to our criteria (16 medial septal cells were not phaser cells; Table 1). Unlike the Tingley & Buzsáki [42] phase code on tracks, LS phaser cells had strongly rate-coupled phase modulation and a wide range of spatial patterns including wall/boundary responses [61–63] that may be available to the LS via subicular afferents [60]. LS phaser cells demonstrated a symmetric and bidirectional code for allocentric space (Fig 3A), whereas hippocampal phase precession is an asymmetric and unidirectional code for distance relative to the boundaries of a place field (Fig 3B). Thus, rate-coupled phaser cells and rate-independent precession may represent distinct neuronal populations or distinct operating modes within LS and/or other structures, possibly mediated by heterogeneous connectivity patterns.
Delay-based phase codes as in our positive phaser cells have not, to our knowledge, been previously demonstrated. Three of our positive phaser cells were located in the dentate gyrus, which receives input from a LS-supramammillary pathway [60], suggesting possible hippocampal entrainment by LS phaser cell activity. Hippocampal negative phaser cells with strong spatial rate codes (and place-like selectivity) additionally demonstrated stronger directional and speed coding (S3 Fig, panel B), thus contributing to the trajectory component of the space–trajectory trade-off observed in our GLM analysis (S6 Fig). Our sample of hippocampal cells was too small to draw conclusions, but that relationship suggests that some hippocampal phaser cells may have been place cells reflecting phaser-entrainment signals from subcortical pathways. Our positive phaser model was based on theta excitation and negative-phaser inhibition (Fig 7), consistent with the prevalence of GABAergic neurons and recurrent collaterals in LS [60]. Our bursting models showed that, given convergent spatial and theta-rhythmic input, phaser cells could operate intrinsically without inheriting phase relationships from CA3 or CA1. Convergent inputs allow the possibility that the longitudinal-to-vertical-band topography of the hippocampus-LS projection [60] averages over the spatial and theta-rhythmic activity of many place cells, effectively displacing hippocampal tuning specificity so phaser cells can exploit hippocampal input while computing distinct codes. Thus, both extrinsic and intrinsic phase transformations of hippocampal spatial information may arise in the LS and/or other structures depending on contextual and behavioral requirements.
Early theoretical models suggested that hippocampal sequences, learned via phase precession and/or temporally asymmetric synaptic plasticity, enabled context-dependent predictions of future positions [64–68]. Experimental studies revealed theta-rhythmic forward-sweeping sequences during active locomotion [69, 70] that mentally probed paths ahead of the animal’s current position to guide navigational decisions [71, 72]. This research suggests a major function of theta-rhythmic information processing along the trisynaptic circuit of the hippocampal formation is to generate memory-guided predictions of future states given the current state. The current state may be reflected in CA3 or CA1 activity at the trough of local theta waves [56], but it could also be directly encoded by other theta-rhythmic structures. Specifically, if recurrent network plasticity and phase precession enable future-oriented sequences, then phase codes in extrahippocampal circuits without those elements may be more likely to encode the current state by default. Such phase codes would be symmetric and bidirectional, similar to phaser cells as well as hippocampal place fields during initial exposure to a novel environment [8, 44, 73]. Thus, phaser cells may provide an experience-independent temporal code for the current state.
The phaser cell spatial transformation is inherently less precise than phase precession. Its bidirectionality assigns the same phase to different locations: for example, a single phase would map to opposite edges of a 1D place field on a track (Fig 3C) or a concentric ring (isocontour) of a 2D place field (Fig 3D). In contrast, the unidirectionality of phase precession enhances the rate-coded position signal of a place cell by contributing unambiguous information about distance traveled through its place field [4, 6]. Phase precession constructively adds to coding precision, but the phaser cell code may serve to directly transform spatial information. We showed that the phaser cell code was stable across hours and days, suggesting that it may contribute to the context-dependent spatial computations of hippocampal/entorhinal circuits. LS spatial modulation has been previously shown to exhibit distinct responses to context changes compared to hippocampal place cell remapping [41]. Our study did not address context-dependence, but it did reveal spatial heterogeneity across phaser cells (for example, Fig 4), thus supporting our theoretical notion that phaser cell responses provide a basis for flexible spatial learning across contexts.
One benefit of a bidirectional phase code is that positive phase modulation can coexist with negative phase modulation in the same network. To illustrate the spatiotemporal activation of symmetric rate-coupled phase codes, we could imagine layers of negative and positive phaser cells with 2D bell-shaped spatial tuning and uniformly distributed fields. At the trough of a theta wave, negative phaser cells representing the current location fire first and strongest, followed by their neighbors in all directions. Activation continues in a radial wave extending outward and dissipating by theta peak. Positive phaser cells, conversely, follow a reverse radial wave that begins with a wide concentric circle of weak firing at theta peak and collapses onto the current location with strong firing before theta trough. This expansion and contraction of radial waves would collectively span the theta cycle as a consequence of the theta-segregation of negative and positive phaser cells (Fig 3E). Thus, phaser cells may form a spatiotemporal cursor marking the present.
The neural mechanisms of path integration are not well understood. In rats, experimental inactivation of the medial septum has been shown to reduce the theta rhythm and disrupt grid cell firing [74, 75], but preserve the spatial firing of hippocampal place cells [76] except in conditions such as large environments (or wheel running) in which performance would be expected to rely more on path (or time) integration than external cues [77]. Similarly, septal inactivation of theta using gabazine (but not muscimol or tetracaine) was demonstrated to preserve hippocampal spatial activity while impairing navigation to a hidden goal [78]. In mice, path integrating behavior is preserved in the dark (cf. the control animals tested by [36]) even though spatial grid cell activity has been shown to require visual input [35]. These findings suggest the theta rhythm is critical to path integration independent of place field maps or grid cell periodicity, raising the question whether it plays a direct computational role or a supporting role (such as phase reset), or contributes to both.
The temporal interference models based on VCO units [19–21] posited a direct role in which relative phases between oscillators constitute a spatial vector anchored to a previous reference point. We previously showed that a generalized VCO model could be effectively calibrated by extended sensory cue interactions that mediated phase-code feedback [31], although that study was agnostic to the feedback mechanism. Here, we demonstrated burst-synchronized entrainment of target neurons that learned VCO-like activity patterns (Fig 8). However, detecting collective synchrony among a population of phaser cells is a general decoding mechanism that could theoretically support a continuous attractor network of grid cell activity [30]. In that case, temporal coordination within the theta cycle might act as a signal boost for spatial feedback to reset the location of the activity bump (cf. [32]). Additionally, a main criticism of VCO theories followed from the finding of grid cells in bats without continuous theta oscillations [79]. However, like VCO-based path integration (see discussion in [25]), a phaser-based reset does not necessarily require rhythmic periodicity: synchrony could arise from structured latencies due to shared arrythmic inputs. Indeed, phase locking and phase coding by hippocampal and medial entorhinal neurons in crawling bats has been reported to be organized by nonoscillatory LFP fluctuations [80]. While our phaser models required theta rhythmicity, the mechanism of spatial synchrony that they demonstrated could be generalized to nonoscillatory systems. Despite widely varying navigational and perceptual requirements across species, synchronous (but not necessarily oscillatory) neural activation may be organized by allocentric features. The main requirement is that path integration reset must be linked to the current state of the world. Thus, LS phaser cells in rats may operate a present-focused reset mechanism parallel to future-focused hippocampal dynamics.
The phaser models assumed that temporal contiguity, as measured by spatial phase-tuning similarity, promotes associative synaptic weights [58, 59] between phaser cells and their targets. The supervised competitive mechanism was not realistic, but our modeling goal was to demonstrate the functional implications of having competitively weighted phaser inputs. The simplified learning mechanism represented the end result of an animal’s familiarization with a given environment. During exploration, we supposed that path integration produces a ‘teacher’ signal that associates internal states with external cues represented in phaser cell inputs. This would be a noisy signal in novel environments or disoriented animals, but investigatory behaviors in those situations emphasize incremental exploration and active management of path integration [81]: shorter excursions, direct returns to home base, and more visual fixations and/or head scanning [82]. These behaviors may stabilize the teacher signal to allow the path integrator to learn new weights from phaser cells (or other inputs). For example, in a VCO-based path integrator, relative phases between ring networks would coherently advance and delay relative to idiothetic motion signals [26, 27]. As long as those phase modulations were relatively continuous between sensory fixations, then any resulting spatial structure in the relative phase pattern would serve to reinforce itself by enhancing co-active inputs from phaser cells with similar spatial phase tuning. Our supervised phase codes (Figs 7C and 8B+8E) temporally collapsed the process of learning a teacher signal into a single pattern.
An additional complication for VCO-based path integration is that learning requires theta-rhythmic coupling between the target and its phaser inputs. However, the burst frequency of VCOs increases with movement in the preferred direction [19, 22]. Thus, phase-coupled synaptic modification would be restricted to the subset of VCOs with preferred directions orthogonal to the animal’s current direction. This limitation would be mitigated by ring attractor organization of VCO cells [26, 27], in which learning would be continuous because every orthogonal direction would be represented by a cell in the network. For continuous attractor-based path integration in grid cells, phaser cells and grid cells would be phase coupled via the shared hippocampal-entorhinal theta rhythm [83], but phase locking of layer III grid cells to the local theta trough [5] could restrict learning to negative phaser cell inputs. Future studies are needed to determine biologically plausible learning mechanisms.
The continuous activity of phaser cells further raises the question of how a path integrator would switch from internally integrating self-motion to receiving phase-code feedback to reset errors. Presumably, both processes could not occur concurrently. Our models (including [31]) suggest that resetting to stabilize the spatial representation of a familiar environment requires theta-phase coupling (similarly to learning) but it only needs to punctuate path integration briefly enough to achieve burst synchronization (Fig 8H; S7 Fig). Punctuated resets could be adaptively driven by investigatory behaviors like head scanning [82] or boundary visits [84], or by error signals mediated by grid cells [27, 85]. Ring attractor organization of VCOs could enhance the robustness of phase-code resets by propagating updated phase offsets via intrinsic connectivity. Furthermore, our examination of LS phase codes may be biased by our sample of recording sites. Tingley & Buzsáki (2018) [42] found a dorsal-ventral dissociation in LS phase coding properties, including evidence that local theta is a traveling wave in the dorsal-ventral and medial-lateral directions. Thus, the theta-phase diversity of phaser cells is potentially much broader than our sample, enabling additional entrainment or switching mechanisms in downstream targets.
Theories of the neural circuits of spatial cognition should go beyond representations to describe how target brain areas read, decode, and translate signals along the path to decisions and behavior. We presented exploratory single-unit data revealing a rate-coupled spatial phase code in neurons found in the LS, hippocampus, and other subcortical areas. Dynamical bursting models helped to explain observations in the data, but they also demonstrated how collective synchronization codes among phaser cells could be learned and decoded by target cells and networks. Our data and models suggest a subcortical phase-code feedback loop for allocentric space may be mediated by phaser cells in LS and/or other regions. Future studies of the role of theta oscillations in spatial navigation may consider the phaser cell mechanism or our theorized feedback pathway to provide a useful perspective. Further research is needed to determine which pathways might support this feedback, but the LS is ideally positioned to translate hippocampal spatial and theta-rhythmic output to downstream subcortical areas [60, 86] that regulate the theta rhythm [37, 38] and theta-bursting thalamic nuclei [22, 87, 88] including the nucleus reuniens with hippocampal and entorhinal projections [60, 89, 90]. Spatial synchronization codes may resonate through limbic loops to reconcile internal maps with external sensory experience.
Rats were chronically implanted with recording devices under deep isoflurane anesthesia. All experiments were conducted in accordance with the U.S. National Institute of Health Guide for the Care and Use of Laboratory Animals (NIH Publications No. 90-23), and were approved in advance by the animal subjects review committee at the University of California, Los Angeles.
We define a quadratic integrate-and-fire model [51] of intrinsic bursting with a fast variable for the spiking limit cycle (V) and a slow adaptive variable for terminating bursts (u). The dynamics follow
τ V ˙= Φ ( V ) - u + I ( t ) τ u ˙= a ( b V - u ) (5)
where I(t) is a cell-specific time-varying input, Φ(V) = 0.04V2 + 5V + 140 is a quadratic nonlinearity for spike initiation, a and b control adaptive feedback, and τ sets a shared time-scale for spiking and bursting (in addition to the time constants implicit in Φ(V) and a). Whenever V > Vt, a spike is recorded, V is reset to c, and u is incremented by d. Bursting parameters are listed in Table 2. While V is approximately millivolt scale, we treat this system as a qualitative, not biophysical, model for which the parameters are in arbitrary units. For theta-rhythmic inputs and recording theta phase, simulations tracked a reference theta wave at frequency fθ = 7.5 Hz, matching the typical burst rate in our single-unit recordings.
For negative phasers, we set the time-varying input (Eq (5)) to the combination
I ( t ) = I θ ( t ) + I ext ( t ) (6)
of sinusoidal theta inhibition (for inhibitory gain gθ < 0)
I θ ( t ) = g θ [ 0 . 5 ( cos ( 2 π f θ t ) + 1 ) ] (7)
and external excitatory input (for excitatory gain ge)
I ext ( t ) = g e F ext ( t ) (8)
where the external input function Fext(t) had range [0, 1].
The positive phasers had theta gain gθ > 0 and followed Eq (5) with negative-phaser input
I ( t ) = I neg = - g inh ( V - E inh ) (9)
where ginh was a slow inhibitory conductance
τ inh g ˙ inh = - g inh (10)
that was incremented by dinh with every pre-synaptic spike (Table 3).
The target bursters had a shorter time-constant (↓τ) and lower burst excitability (↑d; Table 2). In place of Eq (5), the fast variable followed
τ d V d t = Φ ( V ) - u + I syn ( t ) + I const + σ ξ τ d t (11)
where normalized white noise ξ was controlled by gain σ, and Isyn(t) was the total synaptic drive from the phaser network
I syn ( t ) = ∑ k ∈ { neg , pos } [ g k ∑ j = 1 n p W k j δ ( t - t k j ) ] (12)
where np was the number of phasers in each subtype layer, gneg and gpos were subtype-specific feedback gains (Table 4), Wneg and Wpos were the phaser weight vectors (for example, Fig 7B), and tneg and tpos were most-recent-spike vectors. Constant input current was tuned (Iconst, Table 4) so that the intrinsic burst rate, without noise or synaptic input, was close to reference theta frequency (7.519 s−1 compared to fθ = 7.5 Hz).
Spiking neuron and network models were implemented in the equation-based Brian simulator [91]. Simulations were integrated in 1-ms timesteps. Phaser layers and the target burster without noise were evolved with Runge-Kutta 4th-order integration; the target burster with noise used the Euler-Maruyama method. Burst timing in simulations was determined as spike times following interspike intervals ≥ 25 ms.
For 1D spatial simulations, place tuning functions were Gaussian functions with bandwidth 1/64 normalized to the range [0, 1] and centered at 64 evenly-spaced positions from 0 to 1. Each notch tuning function was 1 minus a place tuning function. The gain of phaser input onto the target burster (Table 4) was manually tuned for visually matched ‘middle of the road’ synchronization at both fixed points.
For 2D spatial simulations, phase code gratings had 80-cm spatial periods so that one cycle covered the environment. Phaser gain onto the target burster (Table 4) was manually tuned to roughly equalize the size of negative and positive synchronization modes across different reference phases.
Based on 1-hr training simulations, we generated joint space-phase distributions from phaser spikes: 15 × 36 (x × ϕ) bins for 1D simulations; 15 × 15 × 36 (x × y × ϕ) bins for 2D simulations. The supervised phase code was either directly specified as a binary array for 1D simulations or binned from a spatial grating function for 2D simulations. We computed the vector cosine similarity between the space-phase distributions of the phasers and the supervised phase code as the basis for feedforward synaptic weights from the phaser layers to the target burster. To determine competitive weights, we chose the kWTA negative and kWTA positive phasers (Table 4) with the highest similarities and normalized those similarities to the range [0, 1] via [(similarity − min)/(max − min)]. Inactive weights were set to 0. Total phaser input (Figs 7E and 8C) was computed as the product-sum of the weight vector and an array of all space-phase distributions.
We simulated target networks with 64 bursting units that each learned different ranges of phase offsets and preferred directions (Fig 8E). Burst timing was decoded in 267-ms sliding windows (2 theta cycles) that were incremented in 133-ms steps (1 theta cycle). For each unit, the average burst phase was computed in each window; the previous average was used if no bursts occurred in the window. Analogous to methods for decoding spike counts [54], we calculated the posterior probability distribution of spatial position P(x|ϕ) for an array of phase values ϕ as
P ( x | ϕ ) = P ( x t | ϕ , x ^ t - 1 ) = C ( τ , ϕ ) exp ( - | | x ^ t - 1 - x t | | 2 σ c 2 ) ∏ i = 1 n exp ( cos ( ϕ i - Φ x , i ) ) (13)
where xt was the position for the current window, x ^ t - 1 was the MAP position estimate for the previous window, C was a normalization factor based on ϕ and window-size τ that ensured ∑x P(x|ϕ) = 1, σc = 15 cm was the Gaussian width of a spatial contiguity prior, n was the number of units, and Φx,i was the phase value at position x of the 2D spatial phase code that was used to train unit i.
Decoding MSE was computed as the mean squared Euclidean distance between the MAP position and the average of recorded trajectory samples within each window across a 60-s trajectory segment used as a performance benchmark. We decoded the activity from three target-burster networks with 64 units (Fig 8E; S10 Fig, panel B) and the combination of all three networks with 192 units. Each network condition was bootstrapped by sampling (or subsampling to smaller network sizes as in Fig 8G) with replacement the units in the network and then decoding the sample’s activity and computing the MSE as described. Temporal autocorrelations (S10 Fig, panel C) were computed using full-size networks (64 or 192 units) by correlating each bootstrap MSE time-series with itself and normalizing the minimum and maximum of the mean bootstrap correlations to [0, 1]. HWHMs were calculated as the time lag of the earliest window with normalized correlation <0.5 for each bootstrap; data are shown (Fig 8H) as means and empirical 95% confidence intervals of bootstrap HWHMs.
Male Long-Evans rats (350–400 g) were individually housed and kept at 85% of ad libitum weight. They were trained over 5 d to forage for food pellets in an enclosed environment. Under deep isoflurane anesthesia, rats were chronically implanted with tetrode arrays targeting (across rats) the medial and lateral septum, dorsal hippocampus, anterior thalamus, midbrain, and/or other subcortical areas. Each rat was implanted with 16 tetrodes (64 electrode channels) that were grouped into four independently drivable bundles of four tetrodes each.
Data collection methods including conduct of recording sessions, video tracking analysis, and single-unit acquisition have been described previously [22]. Spike trains recorded during different sessions were considered to be from the same cell if (1) they were obtained from the same tetrode, (2) the tetrode had been advanced <80 μm between recordings, and (3) cluster boundaries and waveform shapes were visually similar on all tetrode channels for both sessions. The phase of the septal-hippocampal theta oscillation was quantified from the LFP signal on a reference electrode in the hippocampal stratum oriens. In one subject (rat 11), a strong theta-rhythmic cell was used as phase reference instead of the LFP signal and was not included in data analysis. All analysis data was filtered for linear movement speeds >5 cm/s.
To handle large variance in spatial data density from long recordings, we computed spatial maps with adaptive scaling kernels. We used a KD-tree algorithm to generate a nearest-neighbor model of the data points for the map. For every pixel to evaluate, we found the enclosing radius of the nearest 4% of data points. If the radius was <8% or >30% of the arena diameter, then it was fixed at 8% or 30%, respectively. A Gaussian kernel set weights for each data point in this evaluation radius. For ratemaps, we computed weighted averages of trajectory data and spike data to create occupancy and spike density maps; dividing the spike density by the occupancy map produced the ratemap. For phase maps, we computed weighted mean resultant phase vectors from which we retrieved the mean phase and MVL. The mean phase across pixels produced the mean-phase maps; otherwise, the MVL was maximum-normalized and composited as a color saturation overlay onto the mean-phase map to produce the phase-vector map. Phase maps used colors drawn from the CIELUV color space to maintain perceptual uniformity of intensity across hues.
The rhythmicity index and burst-frequency estimates were derived from spike-timing autocorrelations. We adaptively smoothed 128-bin 0.5-s correlograms to find stable estimates of the first trough and first (non-central) peak of the correlograms. Rhythmicity was calculated as the ratio [(peak − trough)/peak]. Burst-frequency was calculated as the average of the first-peak mode estimate and an estimate based on a weighted-average of the first-to-second-trough correlations.
The theta modulation index was computed from a 10° binned phase histogram on [−π, π]. We circularly convolved the histogram with a 10° bandwidth Gaussian kernel for smoothing. Theta modulation was calculated as the ratio [(max − min)/max] of the smoothed histogram.
We implemented the method of Kempter et al. (2012) [92] for computing circular-linear regressions with stable estimates of the correlation coefficient and p-value. This method was used for all rate-phase regression lines and rate-phase correlation values. For a given unit recording, the input data consisted of the common trajectory-sampled pixels from the 64 × 64-pixel ratemap and mean-phase map computed (as described above) from the unit’s spike data, LFP theta signal, and spatial trajectory. To compute the total phase shift, we multiplied the estimated rate-phase regression slope by the range of firing rates [max − min] in the ratemap.
We calculated spatial correlations as the mean-adjusted cosine vector similarity between the common trajectory-sampled pixels in 64 × 64-pixel ratemaps computed with the adaptive kernel (as described above). We calculated changes in total phase shift as the absolute difference between total phase shifts computed from rate-phase regressions on 64 × 64-pixel ratemaps and mean-phase maps. For the early-late within-session comparisons, the early portion consisted of up to 1-h after the start or the first half of the recording session data (whichever was shorter); the late portion consisted of up to 1-h before the end or the last half of the recording session data (whichever was shorter). The across-cell baseline consisted of each recording’s early portion paired with the late portion from every recording of all other identified cells. For the multiple-day comparisons, spatial correlations and changes in total phase shift were computed using the ratemaps and mean-phase maps based on the full recording session data (as in every analysis apart from the early-late comparisons). The within-cell comparison consisted of all unique pairs of a given cell’s recordings for all cells with multiple recordings. The across-cell baseline consisted of each recording from a cell with multiple recordings paired with every recording of all other identified cells.
We computed spatial phase information Iphase as the mutual information between phase (ϕ) and position (x)
I ( ϕ ; x ) = ∑ x ∑ ϕ p ( ϕ , x ) log 2 ( p ( ϕ , x ) p ( ϕ ) p ( x ) )
based on joint space-phase distributions of spikes binned into 15 × 15 × 36 (x × y × ϕ) arrays. This measure yielded information in units of bits. We permuted spike phases 1,000 times to calculate p-values.
We computed spike information content based on Skaggs’ formulation [45]
I K = 1 F ∑ k ∈ K p ( k ) f ( k ) log 2 ( f ( k ) F )
where K was position, direction, or speed of the trajectory; p was the occupancy density; f was a firing-rate function; and F was the mean firing rate. Position was binned into 15 × 15 arrays on [0, 80] cm along the x and y axes; direction into 36 bins on [0, 2π]; and speed into 18 bins on [5, 50] cm/s excluding bins with <3 s occupancy. These measures yielded information rates in units of bits/spike. We randomly shift-wrapped spike trains with 20-s minimum offsets and re-interpolated trajectory data 1,000 times to calculate p-values.
The direction modulation index was computed as the ratio [(max − min)/max] of a smoothed firing-rate function of movement direction. Average firing rates in 36 direction bins on [0, 2π] were circularly convolved with a 10° bandwidth Gaussian kernel. The speed modulation index was computed as the ratio [(max − min)/max] of a firing-rate function of speed. Average firing rates were calculated for 14 bins on [5, 40] cm/s excluding bins with <8 s occupancy.
Ridge regression models were trained on 9 scalar predictors representing the vector components of the 5 model variables: L = (x, y), Q = (x2, y2, xy), W (scalar), S (scalar), and D = (ux, uy). The wall predictor W was a sigmoid proximity signal [1/(1 + exp(−k(r − w0)))] for radius r from arena center, k = 0.5, and w0 = 30 cm. S was linear trajectory speed. D was the unit vector along the movement direction. Training samples were 300-ms bins and predictors were interpolated at the midpoint of each bin. Each predictor was standardized by subtracting its sample mean and dividing by its sample standard deviation. The response variable was the log spike-count Y for each bin, as in a Poisson-distributed GLM. The trajectory was divided into equal-sized 2 × 2 or 3 × 3 grids based on data limits. For each grid section, the GLM was trained on all data samples inside the section according to interpolated (x, y) position. Estimated model intercepts and coefficients for each recording and grid section were stored for analysis (or for the reduced LQW generative model). To regularize the model, tuning parameter α determined the ℓ2- norm penalty for least-squares optimization
β ^ = arg min β [ ∑ i = 1 n t ( Y i - Y ^ i )2 + α ∥ β ∥ 2 2 ]
where nt was the number of training samples. We maximized model directionality (or, similarly, the wall response W in the LQW generative model) by choosing
α ^ = arg max α [ 1 n r ∑ k = 1 n r e ∥ β D , k ∥ 2 · n t , k ∑ j ∈ { L Q W S D } e ∥ β j , k ∥ 2 ∑ i ( K i , k - K ^ i , k )2 ] (14)
which maximizes (over nr = 1, 073 single-unit recordings) the softmax directional coefficients while minimizing spike-count (K = exp(Y)) prediction errors (MSE; S4 Fig). The value α = 1.2496 from the 2 × 2 model was used for analysis because of higher likelihood, lower MSE, lower penalty, and complete wall contact across grid sections compared to the 3 × 3 model.
The relative strengths of GLM variables were computed as normalized vector norms
Strength ( X ) = ∑ i = 1 g ∥ β X i ∥ 2 2 ∑ j ∈ { L Q W S D } ∑ i = 1 g ∥ β j i ∥ 2 2 (15)
for variable X ∈ {L, Q, W, S, D} across g grid sections. Thus DSI was computed as Strength(D) and DHI was computed as 1 minus the angular s.d. of the βD vectors across the grid. The maximal contributions of GLM variables were computed similarly to Eq (15) but with maximum linear predictors (Eq (2)) instead of coefficient vector norms. The sum across variables for both relative strength and maximal contribution was normalized within recordings and then averaged by unique cell (Fig 5). Grid matrix plots (S6 Fig, panel A+C) show these values prior to the grid summations (Eq (15)).
To reconstruct ratemaps, we used the midpoints of grid-specific training samples to predict spike counts from the model for each grid section. We collated the counts and sample positions across grid sections to reconstitute a complete dataset for generating the ratemap.
To create the LQW generative model, we used a COBYLA search to find the arena-bounded minimum and maximum of the linear predictor for each recording. We normalized the LQW parameters to [0, 1] and applied a clipping sigmoid [1/(1 + exp(−10(f − 0.5)))] to smoothly enforce the range of the resulting spatial function. To sample the generative model, we randomly selected a negative phaser’s spatial function, added 20% Gaussian noise to its LQW parameters, and rotated the function about the center by a random angle.
Data analysis and modeling were conducted using custom python packages that depend on libraries from the open-source ecosystem: numpy, scipy, matplotlib, seaborn, pandas, scikit-learn, pytables, Brian2, and others. The source code, including a complete specification of the python environment, is available at doi.org/10.6084/m9.figshare.6072317.
|
10.1371/journal.pcbi.1003575 | Constraint-Based Modeling of Carbon Fixation and the Energetics of Electron Transfer in Geobacter metallireducens | Geobacter species are of great interest for environmental and biotechnology applications as they can carry out direct electron transfer to insoluble metals or other microorganisms and have the ability to assimilate inorganic carbon. Here, we report on the capability and key enabling metabolic machinery of Geobacter metallireducens GS-15 to carry out CO2 fixation and direct electron transfer to iron. An updated metabolic reconstruction was generated, growth screens on targeted conditions of interest were performed, and constraint-based analysis was utilized to characterize and evaluate critical pathways and reactions in G. metallireducens. The novel capability of G. metallireducens to grow autotrophically with formate and Fe(III) was predicted and subsequently validated in vivo. Additionally, the energetic cost of transferring electrons to an external electron acceptor was determined through analysis of growth experiments carried out using three different electron acceptors (Fe(III), nitrate, and fumarate) by systematically isolating and examining different parts of the electron transport chain. The updated reconstruction will serve as a knowledgebase for understanding and engineering Geobacter and similar species.
| The ability of microorganisms to exchange electrons directly with their environment has large implications for our knowledge of industrial and environmental processes. For decades, it has been known that microbes can use electrodes as electron acceptors in microbial fuel cell settings. Geobacter metallireducens has been one of the model organisms for characterizing microbe-electrode interactions as well as environmental processes such as bioremediation. Here, we significantly expand the knowledge of metabolism and energetics of this model organism by employing constraint-based metabolic modeling. Through this analysis, we build the metabolic pathways necessary for carbon fixation, a desirable property for industrial chemical production. We further discover a novel growth condition which enables the characterization of autotrophic (i.e., carbon-fixing) metabolism in Geobacter. Importantly, our systems-level modeling approach helped elucidate the key metabolic pathways and the energetic cost associated with extracellular electron transfer. This model can be applied to characterize and engineer the metabolism and electron transfer capabilities of Geobacter for biotechnological applications.
| Microorganisms play a major role in the global carbon cycle. Insights into the various mechanisms and energetic constrains which govern their behavior will advance our understanding of carbon fluxes and might ultimately allow for a rational perturbation of the carbon cycle. Key features of the carbon cycle are the conversion of organic and inorganic carbon and the energy flow through the system. Qualitative and quantitative knowledge of carbon assimilation and electron flow to and from key microorganisms is critical when evaluating certain aspects of the carbon cycle. Members of the genus Geobacter are ubiquitous in the soil environment, and have been described to utilize various organic substrates while transferring electrons to insoluble metals externally [1]. Furthermore, certain Geobacter species, such as G. metallireducens have been reported to transfer electrons directly to poised electrodes [2] and even to other microbes [3], a process coined direct interspecies electron transfer or DIET. Quantitative assessment of carbon and energy flow in G. metallireducens by computational modeling approaches therefore provides valuable insight into the role of this bacterium in the carbon cycle.
Geobacter metallireducens, the first Geobacter species that was isolated [4], serves as a pure culture model for the study of many of the important reactions that Geobacter species catalyze in the biogeochemistry of anaerobic soils and sediments, groundwater bioremediation, and several bioenergy applications [5]. For example, Geobacter species play a major role in the biogeochemical cycling due to their ability to couple the oxidation of organic compounds to the reduction of Fe(III) and Mn(IV) oxides [5]. G. metallireducens was the first microorganism shown to be capable of the anaerobic degradation of aromatic hydrocarbons [6] and Fe(III) is an important electron acceptor for the removal of aromatic hydrocarbons in many contaminated subsurface environments [5]. Direct electron transfer from electrodes to microorganisms to drive anaerobic respiration has potential applications in bioenergy and bioremediation [7].
Constraint-based reconstruction and analysis (COBRA) is a powerful method for characterizing the content of an organism, or systems of organisms, and understanding the limits of its collective functionality [8]. Metabolic network reconstruction (the most widely utilized form of COBRA) enables the enumeration of the genome-wide machinery (i.e., enzymes, uptake systems, etc.) in an organized fashion for use in modeling [9]. With a reconstructed network for an organism, predictions can be made about its functionality when combined with physiological data in a modeling framework. Further, a validated and accurate network can be utilized for prospective design and engineering of cellular networks [10].
There is a history of modeling Geobacter sp. using COBRA [11]. One of the first studies utilizing COBRA and Geobacter was for G. sulfurreducens [12]. The key findings of this study were an initial reconstruction and examination of the extracellular electron transport, the examination of the efficiency of internal biomass biosynthetic pathways, and predictions of gene deletion phenotypes. A subsequent study branched off to build a reconstruction of G. metallireducens based on the original content of the G. sulfurreducens reconstruction [13]. The initial G. metallireducens reconstruction was used to examine the efficiency of pathway usage in the network along with yield on a variety of substrates.
In this work, an updated reconstruction was built and analyzed to better understand key capabilities of G. metallireducens. The updated reconstruction effort was fueled by the appearance of a more complete genome annotation [14] and new data available for the electron transport chain and key metabolic content [15], [16].
An updated reconstruction of G. metallireducens GS-15, iAF987, was generated by reconciling an existing genome-scale reconstruction [13] and an updated genome annotation, performing a bottom-up reconstruction of additional metabolic pathways. This new reconstruction was functionally tested for performance under known growth conditions (Figure 1A). The final reconstruction contained 987 genes, 1284 reactions, and 1109 metabolites. In the first phase, the existing reconstruction was compared to the updated genome annotation [14] to identify a list of agreements, discrepancies, and scope for expansion. A distinct periplasm compartment was determined to be important as G. metallireducens has the unique ability to transfer electrons extracellularly [5]. Thus, characterizing the electron transfer pathways from the cytosol through the periplasm to the extracellular space was crucial for understanding this unique capability. Furthermore, the addition of the periplasm compartment allows for a more accurate representation of metabolism, such as p-cresol and 4-hydroxybenzyl alcohol degradation, which partially occurs in the periplasm [17]. A wild-type and a reduced ‘core’ biomass objective function [18] were formulated to validate whether the reconstruction could generate the appropriate biomass components necessary to replicate, and for use in simulation to predict the growth rates of the organism on the different substrates. Gaps were filled in the network using data characterizing growth of G. metallireducens GS-15 on 19 different carbon sources/electron donors with Fe(III) as the electron acceptor, and the SMILEY algorithm [19] (see Text S1).
The iAF987 reconstruction was compared to the previous version [13] and an automatically generated reconstruction from the ModelSEED framework [20] (Figure 1B) was used to identify and evaluate newly reconstructed and unique content. The ModelSEED reconstruction was found to have 114 unique genes that were not present in the iAF987 reconstruction. Of these, 86 genes were involved in macromolecular synthesis, DNA replication, and protein modifications that are beyond the scope of a metabolic network, and 8 of them did not have a specific reaction association in the ModelSEED (i.e., generic terms such as aminopeptidase, amidohydrolase). Of the remaining 20 genes, only two (Gmet_0988 and Gmet_2683) were added to the reconstruction as isozymes for existing reactions; the other 18 assignments conflicted with our functional annotation of the genome and thus were not included. The iAF987 reconstruction contains 227 genes not in either reconstruction, thus representing a significant advancement of coverage. These newly included genes encode several unique pathways, encoding 325 unique reactions, which have not previously appeared in a collection of 14 representative reconstructions (Table 1) from the UCSD database from which iAF987 was constructed and is internally consistent (see Text S1 for a detailed comparison of iAF987 to previous work).
Transcriptomic data profiling a growth shift from acetate to the aromatic compound benzoate was integrated with the metabolic model to validate its content. Specifically, the computational analysis was performed using the MADE algorithm [21] which uses the statistical significance of changes in gene expression to create a functional metabolic model that most accurately recapitulates the expression dynamics. Of the 987 genes in the metabolic model, transcriptomic data indicated that the expressions of 857 genes do not change significantly and 130 were differentially expressed (>2-fold and p-value<0.05). Specifically, 77 genes were up-regulated and 53 were down-regulated during this shift. The MADE algorithm predicted that during this metabolic shift, the expression of 885 genes in iAF987 do not change significantly and 102 genes are differentially expressed. Of the 102 differentially expressed genes, the MADE algorithm predicted the up-regulation of 70 genes and down-regulation of 32 genes. Specifically, the model predicted upregulation of 70 genes, while data indicated 77 genes to be upregulated during this shift. Similarly while the model predicted downregulation of 32 genes, the data actually indicated that 53 genes were downregulated during this shift. The model-based prediction of change in expression disagreed with the in vitro transcriptomic data for only 28 of the 987 genes leading to 97% overall agreement (for a more detailed breakdown, see Text S1 and Table S1). Among the genes differentially expressed during the shift, the genes encoding for benzoyl-CoA reductase were up-regulated over 100-fold during benzoate growth. It was determined that this key enzyme that links the degradation of aromatic substrates to central metabolism is not ATP driven as previously thought [13], but is likely membrane bound and proton translocating [16]. Thus, a proton translocating reaction was added to the reconstruction for this step in metabolism. A translocation stoichiometry of 3 protons per electron was determined to be the likely extent of coupling through a thermodynamic analysis (see Text S1). Similar transcriptomic analyses for growth shifts on two other aromatic electron donors (i.e., toluene and phenol) yielded 86% and 84% agreement, respectively (Table S1). These findings will likely broaden our knowledge of how G. metallireducens can be utilized for bioremediation.
The genome of G. metallireducens GS-15 encodes two out of the six known carbon fixation pathways [22]. The pathways which were reconstructed in iAF987 are the reductive citric acid cycle (rTCA) and the dicarboxylate–hydroxybutyrate cycle [22] (Figure 2A). Key enzymes for the rTCA include the 2-oxoglutarate synthase (abbreviated OOR2r in the reconstruction) and ATP-citrate lyase (ACITL), both of which enable the citric acid cycle to run in reverse. For the dicarboxylate–hydroxybutyrate cycle, the key enzyme is 4-hydroxybutyryl-CoA dehydratase (4HBCOAH). The rTCA and the dicarboxylate–hydroxybutyrate cycles share four reactions. The reconstruction of these carbon fixation pathways led to the prediction of a new growth condition for G. metallireducens (Table 2). With the expanded content including the carbon fixation pathways, it was computationally predicted that G. metallireducens can grow with formate as the electron donor and Fe(III) as the electron acceptor when a computational screen of all possible media combinations was performed with the model (Table 2). Investigating the resulting flux distributions revealed that the CO2 derived from formate oxidation is reduced via the rTCA to form acetyl-CoA which is subsequently assimilated into biomass. The electrons derived from formate oxidation are split between running the rTCA and for Fe(III) reduction. The energy gained by formation of a proton gradient during Fe(III) reduction was instrumental for providing the required ATP for carbon fixation. This prediction of growth on formate and Fe(III) was experimentally validated, Figure 2B. The requirement of CO2 fixation for growth of G. metallireducens solely on formate and CO2 as carbon sources is further highlighted by a study which examined G. sulfurreducens reducing Fe(III) with formate as the electron donor. While G. sulfurreducens was able to reduce Fe(III) with formate as electron donor, it required the addition of 0.1 mM acetate to assimilate cell carbon (i.e., grow) [23]. This was attributed to the lack of an rTCA in G. sulfurreducens, specifically due to the absence of the ATP-dependent citrate lyase. Overall, this example represents a unique power of the model to rapidly generate hypotheses in silico that can then be verified experimentally. This particular result proves to be of great interest for examining carbon fixation.
To reconstruct the electron transport system of G. metallireducens, key steps involving electron transfer to the terminal electron acceptor were subject to a thermodynamic analysis (Text S1). Specifically, the feasibility of proton translocation and the theoretical maximum proton translocation stoichiometry was determined [24], [25]. Subsequently, the included content was analyzed with physiological data from growth screens to predict stoichiometries for the key reactions of the electron transport system (ETS). This was performed in a model-driven iterative process with an approach that delineates the energetics of extracellular electron transfer by examining three distinct modules (Figure 3A–B). These modules were characterized by representative electron acceptors; fumarate, nitrate, and Fe(III), respectively. Reduction of fumarate in G. metallireducens requires the strain to harbor the dcuB gene [26]; nitrate is reduced via the ETS and the nitrate reductase [14], [27]; extracellular Fe(III) requires several extracellular c-type cytochromes and pili [28], which have been shown in the closely related G. sulfurreducens to have metal-like conductivity [29]. Acetate consumption has previously been described as a preferred optimal growth condition for G. metallireducens [27], therefore it was used as the primary electron donor when analyzing the ETS in this study. The pathways and key ETS reactions for this conversion are shown in Figure 3C and evidence for the inclusion of the reactions in iAF987 is given in the Text S1.
The first step in the modeling process was to examine the fumarate reductase reaction and maintenance energies necessary for predicting phenotypes using constraint-based analysis (Step 1, Figure 3C). To examine this content, data was generated using the G. metallireducens dcuB strain and utilized in simulations. An electrogenic bifunctional fumarate reductase/succinate dehydrogenase (FRD2rpp) was included based on recent biochemical evidence in a similar species [15]. This inclusion was validated with the growth data of G. metallireducens on acetate and fumarate using the dcuB strain. The earlier version of the G. metallireducens model [13] and the current model (iAF987) without an electrogenic fumarate reductase were unable to produce a feasible solution when constrained with experimentally measured acetate and fumarate uptake rates (see Table 3). However, the electrogenic fumarate reductase enabled the model to reproduce the experimental observations and was assigned to translocate two protons per two electrons transferred to fumarate. It was determined that two protons were necessary to drive the endergonic oxidation of succinate with menaquinol as part of the TCA cycle during Fe(III) respiration and nitrate reduction (see Text S1). Maintenance energies were also evaluated in this step by choosing a growth condition that eliminated any use of the external electron transferring reactions (so these reactions could be examined in isolation later). G. metallireducens dcuB strain was grown in batch using acetate/fumarate medium and also in a chemostat at a set growth rate (Table 3). The growth rate determined in this study was similar to that previously reported (0.114 hr−1 vs 0.105 hr−1, respectively) [26]. From these two conditions, it was possible to estimate the maintenance costs (GAM and NGAM) for the model using an established procedure [9], [18]. The calculated costs were 79.20 mmol ATP gDW−1 for the GAM and 0.81 mmol ATP gDW−1 hr−1 for the NGAM. These values are similar to those found previously for G. sulfurreducens [12], thus it provided confidence in the use of the GAM and NGAM throughout the study. The additional content included in the ETS was further analyzed using experimental data.
The second step in the modeling process to examine extracellular transfer was to examine the energetics of the menaquinone cytochrome oxidoreductase reaction (CYTMQOR3) in the ETS (Step 2, Figure 3B). This was performed by switching the electron acceptor from fumarate to nitrate, while keeping acetate as the electron donor. Assuming the nitrate reductase translocates two protons per two electrons (a value that has been verified [30]), the translocation stoichiometry of the menaquinone cytochrome oxidoreductase can be isolated. Analyzing the energetics of the CYTMQOR3 reaction, it was determined that a likely number of three protons are translocated per two electrons transferred to the cytochrome pool (see Text S1). Using this stoichiometry and simulating growth with acetate and limiting nitrate (see Table 3), the predicted optimal growth rate was calculated to be 0.054 hr−1 as compared to the experimentally determined value of 0.050 hr−1 (in this experiment, the growth rate is equal to the set dilution rate in a steady state chemostat). Thus, it was concluded that the proton translocating assignment of the CYTMQOR3 reaction was consistent with the experimental results of the growth screen.
The final step taken to predict the energetic cost of transferring electrons to an external substrate was to examine growth of G. metallireducens on acetate and Fe(III) after reconciling the other components of the ETS consistent with observed phenotypic data. Figure 3C, step 3 shows the reconstructed path from the inner membrane cytochromes to the outer membrane cytochromes and eventually to reduce Fe(III). An energetic cost was estimated in terms of an ATP cost proportional to the flux of electrons to Fe(III). Growth screens of wild type G. metallireducens (acetate/Fe(III)) were performed in triplicate in both batch and chemostat cultures. When culturing with Fe(III) as an electron acceptor, a measurement of biomass is challenging given that the optical density is used to calculate the amount of reduced iron (see Methods). Therefore, the ratio of acceptor produced to donor consumed was used to compare to simulations, as it is more precise than a biomass-normalized uptake and production rate that was calculated by measuring protein content in the chemostat. Furthermore, the ratio of acceptor to donor calculated for both the batch and chemostat conditions was very similar, thus providing a consistent experimental comparison (see Table 3). A Phenotypic Phase Plane (PhPP) analysis [31] was used to compare model-predicted performance to the experimentally measured acceptor to donor ratio. It was calculated that a cost of one proton translocated across the inner membrane per one electron transferred ultimately to Fe(III) best matched the line of optimality in the PhPP analysis (see Figure 3 C, Figure S1). Further, this cost is very close to 0.3 ATP per electron transferred ultimately to Fe(III), as the ATP synthase in the cell converts protons to ATP at a ratio of 3.33 protons per ATP (see Text S1). Further analysis of this modeling approach with phenotypic data on different electron donors (butanol, ethanol, and pyruvate) yielded the same cost of external electron transfer (see Table S3). Thus, it was hypothesized that this is the approximate cost for external electron transfer to iron and the reactions and costs were built into the iAF987 reconstruction as such. This cost can now be further validated for different external electron transfer processes that G. metallireducens is known to carry out.
The work presented here demonstrates how constraint-based modeling and reconstruction can be applied to generate hypotheses that can be tested experimentally. Specifically, modeling revealed a non-obvious culturing condition where carbon fixation could be directly examined. Further, the cost of external electron transfer could be quantified using an iterative and systematic approach. Carbon sequestration is of great biotechnological interest [32]. By understanding the mode of growth for CO2 fixation, computational predictions can be used to guide genetic modifications which enhance the rate of CO2 fixation. Specifically, this could be in the form of reaction knockouts or identification of genes which could be targeted for overexpression which are predicted to enhance CO2 fixation. The reconstructed model advances our knowledge for this unique species and provides a platform for further analysis and hypothesis formulation for environmental and biotechnology applications.
Curation of the genome annotation of G. metallireducens was continued after the initial publication [14] with additional insights from curation of the genome annotations of Geobacter bemidjiensis [33], Pelobacter carbinolicus [34], and other species (M. Aklujkar, unpublished), with extensive reference to the MetaCyc database [35]. The updated annotation was submitted to NCBI with reference numbers CP000148 and CP000149.
The reconstruction was generated in a four-step process. First, the updated genome annotation for G. metallireducens was entered into the UCSD SimPheny (Genomatica, San Diego, CA) database [14]. Next, the existing G. metallireducens reconstruction [13] was mapped to the updated genome using the existing gene-protein-reaction associations [8] and all of the pathways excluding membrane lipid biosynthesis, lipopolysaccharide biosynthesis, murein biosynthesis and degradation, and transport were entered into the SimPheny framework if an exact match for the reaction was present in the UCSD SimPheny database. If an exact match for the reaction did not exist in the UCSD SimPheny database on the level of metabolites participating in the reaction, they were manually evaluated for inclusion (see below). Next, a comparison of the metabolic content included in the updated genome annotation (Dataset S1) that was not in SimPheny was performed. Manual evaluation of new content or disagreements from the annotation and existing genome-scale reconstruction consisted of gathering genetic, biochemical, sequence, and physiological data and reconciling this information to determine the likelihood of each reaction being present in the organism. This manual curation process has been described and reviewed several times [8], [9]. In the manual review process, the KEGG database (www.genome.jp/kegg/), the ModelSEED database [20], and primary literature (see Dataset S2) were used extensively in the manual curation process. Confidence scores were given for each reaction along with noteworthy evidence used to justify inclusion of a given reaction.
The BOFs were formulated using a previous template [36] and are included in Dataset S3. The biomass content previously determined for the close species Geobacter sulfurreducens was used to determine the breakdown of macromolecules [12] except that for total carbohydrate as the distribution in the murein, lipopolysaccharide, and cytosolic fractions was not indicated. Further, the genome annotation [14] was used for the breakdown of chromosome bases, a study on lipid and lipopolysaccharide chain length was used for the breakdown of acyl chain length [37], and the remaining content was approximated using the full profile presented for the gram-negative bacterium E. coli [36]. It should be noted that prediction of growth rate and unmeasured uptake rates are relatively insensitive realistic variations in biomass macromolecular weight fractions [36]. The BOF components included in the core BOF were extrapolated from the core BOF formulated for E. coli as Geobacter sp. have the same gram-negative cell morphology.
The COBRA Toolbox 2.0 [38] and the SimPheny framework (Genomatica, Inc., San Diego, CA) were used for simulations. Constraints used to simulate growth in ferric citrate medium are presented in Text S1. For the evaluation of the maintenance energies, the best-fit values of acetate and fumarate were set to the exact values.
Growth screens were performed in triplicate using cultures in 125 mL serum bottles under anoxic conditions using Fe(III) or fumarate (with the dcuB mutant studies) as an electron acceptor. The composition of the fumarate medium and ferric citrate medium were the same as previously described [39] and [40], respectively. The concentrations of the electron donors utilized in the experiments were 15 mM acetate, 20 mM ethanol, and 10 mM butanol. For the growth screens, the cells were passed in the media in which they were being tested for at least three passages before the growth screen was performed. For the dcuB mutant strain, the correlation between OD and biomass that was used was 0.4561 gDW L-1 OD600-1. This value was determined by growing the dcuB cells in freshwater medium, taking OD measurements at various time points, and weighing the dried biomass at various points of the growth curve. The cells were dried overnight in an oven and weighed on filter paper, with a correction for the amount of weight lost by a filter paper that did not contain cells. Analytes were quantified by HPLC using an Aminex 87-H ion exchange column at 65° C. The mobile phase was 5 mM H2SO4 at an isocratic flow of 0.5 mL/minute. Sample injection volume was 10 µL. Products were identified by retention time using ultraviolet detection at 210 nm and refractive index detection at 30° C internal temperature and 45° C external temperature and quantified by relating peak area to those of standards. Fe(II) was measured using the ferrozine assay as described [40]. Growth rates were calculated by determining the exponential growth phase region from a series of samples taken for each growth screen (typically, 7–10 samples were taken over the entire screen). For this corresponding exponential growth phase region, the ratio of an analyte to the gDW of the sample was determined using a linear fit obtained through the least-squares method (‘regress’ function in MATLAB). This value (mmol gDW-1) was then multiplied by the growth rate to get a corresponding uptake or production rate. The ratios were subsequently multiplied by the growth rate to get the uptake or production rates. Averages and standard deviations were reported and calculated from three biological replicates for each experiment. For the G. metallireducens GS-15 dcuB stain, the fumarate and succinate rates were independently calculated and then averaged for each replicate as there is a 1∶1 ratio for the dcuB exchange and experimental accuracy for each analyte differed slightly. The fumarate uptake rate was calculated using the net fumarate and malate concentrations measured [41]. Malate was observed when growing with ethanol and butanol as electron donors with the dcuB strain. For the acceptor rates in these two conditions, the succinate production rate was used as the rate of the donor to minimize compounding measurement error (although the effective fumarate uptake rate was very similar, <27% difference in any individual replicate).
Chemostat cultures of G. metallireducens GS-15 wild-type and dcuB cultures were grown at 30°C in anaerobic continuous culture vessels as previously described [42]. The GS-15 wild-type strain was grown under donor limiting conditions with 5 mM acetate and 55 mM ferric citrate, the dcuB strain was grown under acetate limiting conditions with 5 mM Acetate and 27.5 mM fumarate. A protein conversion value of protein content (ug/mL) = 0.4344*dry-cell-weight (ug/mL) was used to calculate rates.
To determine if G. metallireducens was capable of growth with formate, cells were adapted to 20 mM formate in a ferric-citrate medium [40] and not provided with any other electron donor. After the third transfer with formate as the sole electron donor, samples were withdrawn anaerobically and substrate consumption was monitored. Formate was detected via high performance liquid chromatography. Reduction of Fe(III) to Fe(II) was monitored with the ferrozine assay. Experiments were run with triplicate cultures. The number of cells was monitored over time until the cultures reached a plateau of Fe(II) production. To fix cells, 900 µL culture was withdrawn anaerobically and mixed with 100 µL glutaraldehyde (25%). Fixed cells were stored at −20°C until processed. All samples were defrosted, placed on filters, filtered and washed with sterile water prior to staining with acridin orange for 3 minutes. Filters were washed and dried. Cell counts were carried out for three biological samples by manually counting at least 5 microscopic fields (Area 5826 µm2 per field) per sample, with the help of a cell counting application in ImageJ (http://rsbweb.nih.gov/ij/). Control cultures consisting of (i) cells in ferric-citrate medium with no formate, and (ii) ferric-citrate medium with formate and no cells (i.e., cell-free) were also performed.
For the microarray experiments performed, G. metallireducens was grown anaerobically with Fe(III) citrate (55 mM) as the electron acceptor and one of the following electron donors: acetate (10 mM), benzoate (1 mM), phenol (0.5 mM), or toluene (0.5 mM). Cells were grown in 1 L bottles and harvested during mid-exponential phase by centrifugation. The cell pellet was immediately frozen in liquid nitrogen and stored at −80 °C. RNA was isolated from triplicate cultures grown on each electron donor with a modification of the previously described method [43]. Briefly, cell pellets were resuspended in HG extraction buffer [44] pre-heated at 65 °C. The suspension was incubated for 10 minutes at 65 °C to lyse the cells. Nucleic acids were isolated with a phenol-chloroform extraction followed by ethanol precipitation. The pellet was washed twice with 70% ethanol, dried, and resuspended in sterile diethylpyrocarbonate-treated water. RNA was then purified with the RNA Clean-Up kit (Qiagen, Valencia, CA, USA) and treated with DNA-free DNASE (Ambion, Woodward, TX, USA). The RNA samples were tested for genomic DNA contamination by PCR amplification of the 16S rRNA gene. cDNA was generated with the TransPlex Whole Transcriptome Amplification Kit (Sigma).
Whole-genome microarray hybridizations were carried out by Roche NimbleGen, Inc. (Madison, WI, USA). Triplicate biological and technical replicates were conducted for all microarray analyses. Cy3-labeled cDNA was hybridized to oligonucleotide microarrays based on the G. metallireducens genome and resident plasmid sequences (accession number NC007515 and NC007517 at GenBank). All the microarray data has been deposited with NCBI GEO under accession number GSE33794.
For each metabolic shift (benzoate vs acetate, toluene vs acetate, phenol), the fold-changes in expression level and p-value (t-test) were computed using ArrayStar 4.02 (DNASTAR, Madison, WI, USA). This was used in conjunction with the iAF987 metabolic model to predict the metabolic adjustment using the MADE algorithm [21]. Additionally, the constraints for the iAF987 metabolic model were set to simulate growth on the respective substrates. MADE analysis was implemented using the TIGER toolbox [45]. The output of this analysis consisted of the genes predicted by MADE algorithm to significantly change in expression during the concerned metabolic shift.
|
10.1371/journal.pntd.0001457 | In Vitro and In Vivo Efficacy of Monepantel (AAD 1566) against Laboratory Models of Human Intestinal Nematode Infections | Few effective drugs are available for soil-transmitted helminthiases and drug resistance is of concern. In the present work, we tested the efficacy of the veterinary drug monepantel, a potential drug development candidate compared to standard drugs in vitro and in parasite-rodent models of relevance to human soil-transmitted helminthiases.
A motility assay was used to assess the efficacy of monepantel, albendazole, levamisole, and pyrantel pamoate in vitro on third-stage larvae (L3) and adult worms of Ancylostoma ceylanicum, Necator americanus and Trichuris muris. Ancylostoma ceylanicum- or N. americanus-infected hamsters, T. muris- or Ascaris suum-infected mice, and Strongyloides ratti-infected rats were treated with single oral doses of monepantel or with one of the reference drugs.
Monepantel showed excellent activity on A. ceylanicum adults (IC50 = 1.7 µg/ml), a moderate effect on T. muris L3 (IC50 = 78.7 µg/ml), whereas no effect was observed on A. ceylanicum L3, T. muris adults, and both stages of N. americanus. Of the standard drugs, levamisole showed the highest potency in vitro (IC50 = 1.6 and 33.1 µg/ml on A. ceylanicum and T. muris L3, respectively). Complete elimination of worms was observed with monepantel (10 mg/kg) and albendazole (2.5 mg/kg) in A. ceylanicum-infected hamsters. In the N. americanus hamster model single 10 mg/kg oral doses of monepantel and albendazole resulted in worm burden reductions of 58.3% and 100%, respectively. Trichuris muris, S. ratti and A. suum were not affected by treatment with monepantel in vivo (following doses of 600 mg/kg, 32 mg/kg and 600 mg/kg, respectively). In contrast, worm burden reductions of 95.9% and 76.6% were observed following treatment of T. muris- and A. suum infected mice with levamisole (200 mg/kg) and albendazole (600 mg/kg), respectively.
Monepantel reveals low or no activities against N. americanus, T. muris, S. ratti and A. suum in vivo, hence does not qualify as drug development candidate for human soil-transmitted helminthiases.
| Soil-transmitted helminthiases affect more than one billion people among the most vulnerable populations in developing countries. Currently, control of these infections primarily relies on chemotherapy. Only five drugs are available, all of which have been in use for decades. None of the drugs are efficacious using single doses against all soil-transmitted helminths (STH) species and show low efficacy observed against Trichuris trichiura. In addition, the limited availability of current drug treatments poses a precarious situation should drug resistance occur. Therefore, there is great interest to develop novel drugs against infections with STH. Monepantel, which belongs to a new class of veterinary anthelmintics, the amino-acetonitrile derivatives, might be a potential drug candidate in humans. It has been extensively tested against livestock nematodes, and was found highly efficacious and safe for animals. Here we describe the in vitro and in vivo effect of monepantel, on Ancylostoma ceylanicum, Necator americanus, Trichuris muris, Strongyloides ratti, and Ascaris suum, five parasite-rodent models of relevance to human STH. Since we observed that monepantel showed only high activity on one of the hookworm species and lacked activity on the other parasites tested we cannot recommend the drug as a development candidate for human soil-transmitted helminthiases.
| The hookworm species Ancylostoma duodenale and Necator americanus, the whipworm Trichuris trichiura, the threadworm Strongyloides stercoralis, and the roundworm Ascaris lumbricoides are soil-transmitted helminths (STH) of great public health importance. Cumulatively, these parasites affect more than one billion people globally, particularly in developing regions of Asia, Africa, and Latin America [1], [2]. If untreated, infections with STH are present for years and patients suffer from moderate to severe intestinal disturbances, anemia, nutrient loss and profound physical and mental deficiencies [3], [4].
Helminth control relies primarily on the regular administration of anthelmintics, typically carried out within the framework of school-based deworming programs, once or twice a year [5]–[7]. Five drugs are currently available for the treatment of infections with STH (albendazole, mebendazole, pyrantel pamoate, levamisole, and ivermectin), all of which have been registered for human use before or during the 1980's [8], [9]. No new anthelmintic drug for human use has reached the market since then. Moreover, none of these drugs are efficacious using single doses on all STH species, with particularly low efficacy observed on T. trichiura [10]. Relying on only a handful of drugs is a precarious situation, in the light of a possible emergence of drug resistance [10].
Since drug resistance to nematodes of veterinary importance is widely spread and increasing in frequency, most of the anthelmintic drug research and development efforts are motivated by veterinary needs [11]. For example, albendazole, mebendazole, and pyrantel pamoate were originally developed for livestock and pets [12].
Monepantel (AAD1566) belongs to a new class of veterinary anthelmintics, the amino-acetonitrile derivatives. It has been proposed that monepantel interferes with nematode-specific acetylcholine receptor subunits, leading to body wall muscle paralysis and subsequent death of worms. Due to its unique mode of action, the drug has proven efficacy against nematodes infecting livestock which are resistant to current anthelmintic drugs [13]. Monepantel has been extensively tested on different nematode isolates. Administered at a single oral dose of 2.5 mg/kg, it was found to be safe, well-tolerated by ruminant hosts, and showed high cure rates on fourth stage larvae and adult worms of 15 nematode species [13]–[16]. Due to its high and broad nematocidal activity, monepantel was considered to be a candidate for a human health directed program.
The aim of the present investigation was to study the activity of monepantel, compared with the reference drugs, albendazole, levamisole, and pyrantel pamoate, in five parasite-rodent models, that correspond to important human STH and in vitro. Ancylostoma ceylanicum and Necator americanus were both adapted using eggs from infected dogs or humans to an unnatural host, the golden hamster, and represent robust rodent models for hookworm infections [17], [18]. The Trichuris muris mouse model is an excellent model for trichuriasis [19], [20]. Strongyloides ratti in rats is a commonly used murine model for strongyloidiasis [21]. Finally, murine infection with Ascaris suum is a model which mimics the early infection of A. lumbricoides. The survival of different larval stages and adult worms of A. ceylanicum, N. americanus, and T. muris was evaluated in vitro, following monepantel incubation using a motility assay. The in vitro activity of monepantel on S. ratti has been described recently [22]. In vivo, we studied worm burden reductions and for A. ceylanicum, N. americanus, and T. muris, worm expulsion rates were also measured.
Monepantel was kindly provided by Novartis Animal Health, St-Aubin, Switzerland. Albendazole and pyrantel pamoate were purchased from Sigma-Aldrich (Buchs, Switzerland), and levamisole-hydrochloride from Fluka (Buchs, Switzerland).
For the in vitro studies, stock solutions of the drugs were prepared in 100% DMSO (Fluka, Buchs, Switzerland) and stored at 4°C. For the in vivo studies, drugs were suspended in 7% (v/v) Tween 80% and 3% (v/v) ethanol or DMSO/PEG shortly before treatment.
Three-week-old male Syrian Golden hamsters were purchased from Charles River (Sulzfeld, Germany). Four-week-old female NMRI mice and 3-week-old female C57Bl/6J mice were purchased from Harlan (Horst, The Netherlands). Three-week-old female Wistar rats were purchased from Harlan (Horst, The Netherlands).
All animals were kept in macrolon cages under environmentally-controlled conditions (temperature: 25°C, humidity: 70%, light/dark cycle 12 h/12 h) and had free access to water and rodent food (Rodent Blox from Eberle NAFAG, Gossau, Switzerland). They were allowed to acclimatize in the animal facility of the Swiss Tropical and Public Health Institute (Swiss TPH) for 1 week before infection. The current study was approved by the local veterinary agency based on Swiss cantonal and national regulations (permission no. 2070).
Ancylostoma ceylanicum third-stage larvae (L3) were kindly provided by Prof. J. M. Behnke (University of Nottingham). The A. ceylanicum life cycle [23] had been maintained at the Swiss TPH since June 2009 [17]. To maintain the life cycle, hamsters were treated orally 1 day before infection and then twice weekly with 3 mg/kg hydrocortisone (Hydrocortone®, MSD) or with 1 mg/l dexamethasone (dexamethasone water-soluble, Sigma-Aldrich) in the drinking water. They were orally infected with 150 A. ceylanicum L3, which had been harvested less than 1 month before infection and had been assessed microscopically for viability. Animals assigned to in vivo studies were not treated with hydrocortisone and were infected with 300 L3.
Infective N. americanus L3 were the gift of Prof. S. H. Xiao (National Institute for Parasitic Diseases, Shanghai). Hamsters were immunosuppressed with dexamethasone as described above and were infected subcutaneously with 250 viable N. americanus L3.
Embryonated T. muris eggs were kindly obtained from Prof. J. M. Behnke and Prof. H. Mehlhorn. The life cycle had been maintained at the Swiss TPH since January 2010 as described elsewhere [19]. Briefly, T. muris eggs were evaluated for embryonation under the microscope (magnification 80–160×, Carl Zeiss, Germany). NMRI mice were orally infected with 400 embryonated eggs. Mice were treated either subcutaneously (s.c.) 1 day before infection and then every second day between days 5 and 15 with 15 mg hydrocortisone (Hydrocortisone 21-hemisuccinate sodium salt, Sigma-Aldrich) in 0.9% NaCl solution, or with 8 mg/l dexamethasone in the drinking water until the end of the experiment.
The S. ratti life cycle had been maintained over decades at the Swiss TPH, by serial passage through rats. Rats were infected subcutaneously with 735 freshly harvested S. ratti L3.
Infective A. suum eggs were obtained from Prof. S. M. Thamsborg, University of Copenhagen and Prof. G. Cringoli, University of Naples. Briefly, C57Bl/6J mice were orally infected with 500 embryonated eggs, according to a procedure described elsewhere [24].
The larval or adult motility assay is currently the method of choice to evaluate drug sensitivity of different nematode species [25]–[27]. Non-motile worms were considered as dead and the percent viability or survival in each well was calculated.
The average of motility scores for one drug was calculated for each concentration and normalized into percentage, relative to control. IC50 values were expressed based on the median effect principle using CompuSyn (version 1.0). The r value represents the linear correlation coefficient of the median-effect plot, indicating the goodness of fit, hence the accuracy of the IC50 [33]. Variance analysis in the ovicidal activity studies was performed with the Fisher's exact test, using StatsDirect (version 2.4.5; StatsDirect Ltd; Cheshire, UK). The worm burden reductions were determined by comparing the mean number of adult worms in the intestine of a treated group with the mean numbers of worms in the control group. Means and standard deviations were calculated using Microsoft® Excel 2003. The worm expulsion rates were calculated by dividing the number of expelled worms of a treatment group by the group's total worm burden. The Kruskal-Wallis test and the Mann-Whitney U test were used to assess the statistical significance of the worm burden reduction, using StatsDirect.
The effects of monepantel and reference drugs on L3 and adult worms of A. ceylanicum and T. muris after 72 h of exposure in vitro are presented in Table 1.
To date, only five drugs are included in the WHO model list of essential medicines to treat infections with human STH. Most of these anthelmintics were discovered before the 1980s. Though there is no evidence yet for emerging resistance to any of these drugs in human helminth populations, there are worrying signs that anthelminthic efficacy may be declining [34], [35]. In addition, the increased frequency of reported low cure rates, in particular against T. trichiura and hookworm infections, highlight the need to find alternative drugs [10].
A. ceylanicum, N. americanus, T. muris, S. ratti, and A. suum are five well-established laboratory parasite-rodent models of relevance to human STH. The aim of the present study was to determine their sensitivities to monepantel, a broad spectrum and safe drug used for livestock which recently entered the market for veterinary use. It is one of the few available drug candidates eligible for rapid transitioning into development for human STH infections [36].
Monepantel activates signaling via nematode-specific DEG-3 subtype nicotinic acetylcholine receptors (nAChRs), causing a hypercontraction of the body wall muscles leading to paralysis and hence, death of the worm [13]. ACR-23 protein, a member of the DEG-3 group in Caernohabtitis elegans, and its homolog MPTL-1 in Haemonchus contortus, another model for gastrointestinal nematodes, are major targets of monepantel. The absence of MPTL-1, observed in some nematode species resulted in reduced drug sensitivity [22].
Ancylostoma ceylanicum adult worms were found to be highly sensitive to monepantel in vitro, in contrast to the third-stage larvae, but the drug lacked ovicidal activity. Hamsters harboring adult A. ceylanicum were cleared from the worms following a 10 mg/kg single oral dose of monepantel. N. americanus was not affected by the drug in vitro and only moderately susceptible in vivo to 10 mg/kg or higher doses. These findings suggest a relative stage and species specificity, which might be explained by the absence of a functional MPTL-1 homolog in A. ceylanicum L3 and possibly in L3 and adult N. americanus.
Trichuris muris third-stage larvae were only moderately sensitive to monepantel after incubation for 72 h in vitro, whereas adult stages were not affected, neither in vitro nor in vivo. Monepantel had already been reported to lack activity against T. ovis, a minor parasite of sheep [15], [16], a finding that is in accordance with our data.
In addition, monepantel lacked activity in S. ratti-infected rats, a result in line with a recent investigation, which revealed that S. ratti third-stage larvae were not affected by monepantel after 72 h of incubation [22]. For comparison, a complete elimination of adult worms was achieved with ivermectin (0.5 mg/kg) in S. ratti-infected rats [32]. Strongyloides ratti has a remote homolog of DES-2 and ACR-23/MPTL-1 only, which is not targeted by monepantel [22].
Finally, although only one high dosage was tested, our data indicate that A. suum is not affected by treatment with monepantel in vivo, whereas albendazole reduced the worm burden of A. suum in mice at the same dose.
Like H. contortus, A. ceylanicum and N. americanus are members of the nematode clade V, whereas S. ratti belongs to clade IV, A. suum, to clade III, and T. muris to clade I [37]. One could hypothesize that only clade V species exhibit sensitivity to monepantel, whereas those that diverged from this lineage of the evolutionary tree earlier (clades I to IV) might not have evolved homologous receptors. As available for A. suum [38], further genome sequencing of A. ceylanicum, N. americanus and Trichuris spp. remains to be performed in order to extend current knowledge about evolutionary and functional relationships of receptors involved in sensitivity to monepantel.
In the present investigation, albendazole, levamisole, and pyrantel pamoate have been extensively studied in vitro and in vivo. The results obtained are in agreement with earlier in vivo [29], [39]–[41] and in vitro [41] work using A. ceylanicum and N. americanus. In addition, to our knowledge, the in vitro and in vivo sensitivities of these three drugs against T. muris are presented for the first time. In line with human efficacy data [10], albendazole showed highly potent activity against A. ceylanicum, N. americanus and A. suum, yet much less pronounced activity against T. muris in vivo. A similar trend was observed for pyrantel pamoate, which achieved a moderate effect against A. ceylanicum but lacked activity in T. muris-infected mice. These results on pyrantel pamoate are compatible with cure rates reported in clinical trials [10]. On the other hand, levamisole was highly efficacious in our ancylostomiasis and trichuriasis rodent models, while low to moderate cure rates have been recently reported in humans [10], [42].
Interestingly, contradictory results were obtained with albendazole, levamisole, and pyrantel pamoate against adult A. ceylanicum in vitro and in vivo. In addition, albendazole showed excellent activity in the N. americanus hamster model but lacked activity in vitro. This finding might be partially explained by the presence of active metabolites, since for example albendazole and levamisole are rapidly metabolized in vivo [43]–[45]. In addition, large differences in sensitivity between larval and adult hookworm stages were observed with levamisole (and albendazole for A. ceylanicum). It is commonly accepted that the benzimidazoles tend to be lethal to developing stages but not always to adult worms. Developing cells are obviously more harmed by the benzimidazoles, as the utilization of tubulin in the mitotic cycles is affected [46].
In conclusion, to our knowledge, we have for the first time analyzed the efficacy of monepantel in animal models corresponding to human intestinal helminthiases. A recently developed target product profile suggested that a drug development candidate for the treatment of infections with STH should ideally target all stages (at least adult and ova) and species of the major geohelminths such as Ascaris, Trichuris, both hookworm species and Enterobius [36]. Hence, based on our results, established in nematode-rodent models, monepantel does not fulfill the required minimal product characteristics for a new intestinal anthelmintic.
|
10.1371/journal.pcbi.1003810 | Inferring on the Intentions of Others by Hierarchical Bayesian Learning | Inferring on others' (potentially time-varying) intentions is a fundamental problem during many social transactions. To investigate the underlying mechanisms, we applied computational modeling to behavioral data from an economic game in which 16 pairs of volunteers (randomly assigned to “player” or “adviser” roles) interacted. The player performed a probabilistic reinforcement learning task, receiving information about a binary lottery from a visual pie chart. The adviser, who received more predictive information, issued an additional recommendation. Critically, the game was structured such that the adviser's incentives to provide helpful or misleading information varied in time. Using a meta-Bayesian modeling framework, we found that the players' behavior was best explained by the deployment of hierarchical learning: they inferred upon the volatility of the advisers' intentions in order to optimize their predictions about the validity of their advice. Beyond learning, volatility estimates also affected the trial-by-trial variability of decisions: participants were more likely to rely on their estimates of advice accuracy for making choices when they believed that the adviser's intentions were presently stable. Finally, our model of the players' inference predicted the players' interpersonal reactivity index (IRI) scores, explicit ratings of the advisers' helpfulness and the advisers' self-reports on their chosen strategy. Overall, our results suggest that humans (i) employ hierarchical generative models to infer on the changing intentions of others, (ii) use volatility estimates to inform decision-making in social interactions, and (iii) integrate estimates of advice accuracy with non-social sources of information. The Bayesian framework presented here can quantify individual differences in these mechanisms from simple behavioral readouts and may prove useful in future clinical studies of maladaptive social cognition.
| The ability to decode another person's intentions is a critical component of social interactions. This is particularly important when we have to make decisions based on someone else's advice. Our research proposes that this complex cognitive skill (social learning) can be translated into a mathematical model, which prescribes a mechanism for mentally simulating another person's intentions. This study demonstrates that this process can be parsimoniously described as the deployment of hierarchical learning. In other words, participants learn about two quantities: the intentions of the person they interact with and the veracity of the recommendations they offer. As participants become more and more confident about their representation of the other's intentions, they make decisions more in accordance with the advice they receive. Importantly, our modeling framework captures individual differences in the social learning process: The estimated “learning fingerprint” can predict other aspects of participants' behavior, such as their perspective-taking abilities and their explicit ratings of the adviser's level of trustworthiness. The present modeling approach can be further applied in the context of psychiatry to identify maladaptive learning processes in disorders where social learning processes are particularly impaired, such as schizophrenia.
| The process of how we represent others' intentions is an important determinant of social exchange. This inferential process becomes even more crucial when we need to rely on other people's advice regarding a course of action. Credibility can be inferred from another's reputation, which is in turn developed through recursive social interactions [1], [2]. But since advice is motivated by unknown goals, which may also change in time, we are constantly challenged by the question of how accurately we represent others' intentions.
As agents' intentions are hidden from observers, they have to be inferred from their actions. The monitoring of other agents' intentions represents a particular aspect of “theory of mind” [3]–[5]. Different cognitive frameworks for understanding this process have been suggested, e.g. action understanding vs. mentalizing (attribution of mental states) [6]–[8]. Bayesian models in particular provide a formal account of how observers build models of other agents and use them to predict their desires or intentions. One important approach is to formulate social cognition in terms of a partially observable Markov decision process (POMDP) that describes the relations between environmental states (accessible to the observer) and another agent's (unobservable) mental states [9]–[11]. This conceptualization, however, tends to be normative and does not usually emphasize individual variability in social inference. Another framework proposes that theory of mind can be understood in terms of recursive thinking, and focuses on identifying the depth of reasoning that leads to optimal inference [2], [12], [13]. Importantly, so far both types of approaches have been applied to situations where the other agents' intentions are stable over time.
In the present study we build on these previous computational treatments of how humans infer on the intentions of others by considering the additional challenge of detecting how quickly they change in time, i.e. volatility. To this end, we propose novel generative models of how humans may infer on volatile intentions of others and apply these models to behavioral data from a new experimental paradigm. The models we employ are conceptually similar to previous POMDP models, but emphasize individual approximations to Bayes-optimality, as described below.
Specifically, we addressed the following two questions by comparing the explanatory power of alternative computational models that were fitted to the observed behavior: (i) Are humans able to deploy hierarchically structured learning during social interactions and simultaneously predict the accuracy of advice and the stability of the adviser's intentions? (ii) Would humans rely more on social advice (with potentially high information but also unknown degree of uncertainty) or on non-social information that is potentially less accurate but had a known outcome distribution (i.e., risk)?
To address these questions, we designed an interactive and deception-free economic game that involved situations of both aligned and conflicting interests between participants (all male) who were randomly assigned to a “player” or an “adviser” role. In this social exchange paradigm, which builds on a previous task by Behrens et al. (2008), participants received distinct information about the probability of two possible outcomes. The player had to predict the outcome of a binary lottery whose true probability distribution was displayed as a pie chart. The adviser issued an additional recommendation (advice) to the player. The information available to the adviser was still probabilistic, but with a larger and constant probability (80%) as it was generated after the outcome had been drawn (Figure 1).
Importantly, the adviser's payment was structured such that his incentive to provide valid or misleading advice varied during the game and introduced temporal variations in aligned and conflicting interests between player and adviser. This required the player to detect changes in the adviser's intentions and adapt his own decision-making accordingly. It is not clear, however, what exact mechanism underlies adaptive behavior in this scenario: would players only track trial-wise changes in advice accuracy, or would they invoke a more complex hierarchical model, which also assumes that players track the volatility of the advisers' intentions (see [14])? Furthermore, even if the latter was the case, would volatility estimates only serve to optimize inference and learning, or would they directly impact on trial-by-trial variability of decisions?
To address these questions, we considered different explanations (hypotheses) for the behavior displayed by our participants, each of which was formalized as a two-component model. The first component of each model represented the player's belief updating about the causes of the advisor's behavior; we refer to this component as the “perceptual model”. The second component is the “response model”, which maps the current belief to the player's actual decision (see [15], [16]). We constructed a factorially-structured set of 12 different models (model space) by systematically combining different perceptual and response models (see Figure 2), as described in detail in the Methods section. We then fitted these models to the trial-by-trial responses of each subject using Bayesian model inversion and formally compared the plausibility of all 12 models by random effects Bayesian model selection (BMS). Altogether, this corresponds to a “meta-Bayesian” approach [15], i.e., a Bayesian treatment of Bayesian models of cognition, also known as a “doubly Bayesian” [17] or “ecumenical Bayes” [18] approach. This enabled us to identify a hierarchical generative model, which may underlie social inference in our paradigm, and whose parameter estimates predicted independent behavioral data, such as explicit ratings of the players, self-reports on strategy used by the advisers and questionnaire scores.
All participants gave written informed consent before the study, which had received ethics approval by the local responsible authorities (Kantonale Ethikkommission, KEK 2010-0312/3).
Thirty-two healthy male adult volunteers (age range: 19-30 years; median age = 22) participated in the study. Only men participated in this study to avoid potential gender-related confounds in the pairings of advisers and players, such as gender differences in the perception of trustworthiness (with women being perceived as generally more trustworthy than men [19]).
Participants with previous neurological or psychiatric history or who were taking medication at the time were excluded from the study. Three days before the testing session, participants received a battery of psychological questionnaires, which they had to fill out online. This included the Temperament and Character Inventory (TCI-K) [20] to measure personality traits and the Interpersonal Reactivity Index (IRI) [21] to measure empathy, perspective-taking, and theory of mind traits.
Inspired by the paradigm of Behrens et al. (2008), we developed a deception-free and interactive economic game for monetary rewards. This paradigm involved pairs of volunteers (randomly assigned to a “player” and “adviser” role) who met each other for the first time on the day of the experiment. The player had to perform a standard probabilistic reinforcement learning task and was provided with truthful information about the a priori probabilities of trial-wise outcomes by a visual pie chart. The outcome was either green or blue, and all trials contained one of 6 cue types (blue:green pie charts: 75∶25, 65∶35, 55∶45, 45∶55, 35∶65, and 25∶75) (Figure 1a). The adviser, however, received more accurate information: once the outcome was determined (according to the probabilities of the visual pie chart), he was informed about the result with a constant accuracy of 80%. Based on this information, the adviser issued a recommendation to the player on which option to choose. To signal his suggestion, the adviser held up a blue or a green card (Figure 1b); these recommendations were recorded, using a video camera, for use as stimuli in future experiments. Throughout the experiment, both the player and the adviser sat across from each other and were not allowed to interact in any other way than the adviser holding up a card to indicate his suggestion.
Notably, as detailed below, the adviser's pay-off was structured such that his motivation to provide valid or misleading information varied across the game. The player therefore needed to learn about the time-varying intentions of the adviser in order to decide whether to trust him or not on any given trial. In addition to computational modeling of trial-wise choices, we obtained an explicit readout of the player's estimates by requiring him, on 8 out of the total of 200 trials, to characterize the advisers' intentions as “helpful”, “misleading”, or “uninformative”. The timing of these questions and the order of the options was randomized, but they were presented at the same times across subjects.
The player's final payment was proportional to his total score, plus a potential bonus if his score ended in a predefined silver or gold range (see Figure 1). He could track the accuracy of his predictions by monitoring a progress bar at the bottom of the screen, which increased with every correct prediction and decreased with every missed or incorrect response by 1 point. By reaching the silver or the gold target, he could win CHF 10 or CHF 20, respectively (Figure 1 and Table 1). The player was informed before the experiment that the adviser had incentives that were not necessarily aligned with his own and could vary throughout the experiment.
The adviser was able to monitor the player's progress, and was simultaneously shown his own opportunities to gain monetary rewards (i.e., gold and silver ranges, which were unknown to the player). Critically, the targets of the player and the adviser were arranged to create situations of shared and conflicting interests: the gold range of the adviser preceded the silver target of the player, and the silver range of the adviser also ended before the onset of the gold target of the player (see Figure 1 and Table 1).
A typical interaction between the two participants during this game unfolded in the following manner (compare Figure 1): the adviser initially had an incentive to assist the player until the latter reached the adviser's gold range. Once the players' score was within the adviser's gold range, the advisers' incentive to provide misleading advice increased. Once the player recognized this hidden change in intention and either ignored the advice or decided to bet on the opposite color, the player's progress bar was likely to exceed the adviser's gold range. Consequently, if the adviser was unable to confine the player to his (the adviser's) gold range, the next-best strategy for the adviser was to help the player with correct advice again and aim to push him into his (the adviser's) silver range. Once the player reached the adviser's silver range, the adviser had an additional incentive to mislead the player again to prevent the player from moving out of his (the adviser's) silver range.
To distinguish general inference processes under volatility from inference specific to intentionality, each pair of participants also performed a control task. To exclude temporal order effects, the sequence of the two tasks was counterbalanced across participants. In the control task, the adviser was blindfolded and issued his recommendation by picking a card from 6 separate decks placed before him by the experimenter. The blindfolding removed any intentionality by preventing that the adviser could influence what advice he was giving the player; furthermore, the adviser was unable to witness trial outcomes. The predictive accuracy of the six decks of cards was either 80% or 20%. The players were informed in advance that the card decks varied in their predictive accuracy, but not what the probabilities were nor that they were constant per each deck. However, the players could observe from which deck the card was sampled. This control condition thus closely corresponded to the main task, except for the role of intentionality: the player was required to track advice accuracy under volatility (induced by the adviser blindly switching between decks with different accuracy) and had to make trial-wise decisions how to combine the veridical information from the visual pie chart with the more informative (but volatile) advice.
Both tasks included 192 trials (plus the 8 rating trials) with an equal number of 6 cue target types (75∶25, 65∶35, 55∶45, 45∶55, 35∶65, and 25∶75 blue: green pie charts). The trial outcome was randomly drawn from these probability distributions. At the end of the study, all participants were debriefed and asked to describe the strategy that they employed during the game.
In the present study, we examined how subjects updated their beliefs about others' intentions and chose to follow or disregard their advice. For this purpose, we applied two cognitive models (which we here refer to as “perceptual models”): (i) the Hierarchical Gaussian Filter (HGF), a generic Bayesian model of learning under perceptual uncertainty and environmental volatility [22], and (ii) the Rescorla-Wagner (RW) model [23], a commonly used reinforcement learning model. In order to verify whether players really deploy hierarchical learning and infer on the volatility of the adviser's intentions, we also included a reduced (non-hierarchical) version of the HGF as control; this alternative model contained only two levels of learning (see Table 2).
Furthermore, in order to link trial-by-trial beliefs to the observed decisions (and thus enable model inversion), we considered several alternative response models, which differed with regard to whether participants incorporated social and/or non-social sources of information. Together, this resulted in a factorial model space (see Figure 2), which is described in more detail below.
The response model describes how the agent's beliefs (the result of perceptual inference) map onto choices (actions). In our task, subjects can integrate social and non-social information, or use either source of information exclusively. Specifically, the pie chart indicates the true a priori probability about the outcome as non-social information that is directly accessible to the player without need for inference. By contrast, the (uncertain) social information corresponds to the player's belief that the adviser gives correct advice on the current trial. In the HGF, this belief corresponds to the logistic sigmoid transform of the predicted tendency (the posterior from the previous trial) of the adviser to give correct advice (see Eq. (12)):(12)
The response model describes how the player bases his decision on a weighted average of the two sources of information. Taking as the weight of the social information, we obtain the integrated belief that the advice is accurate:(13)
For the RW model, is replaced by .
The probability of the player following the advice (i.e., making decision as opposed to for going against the advice) is then described by a sigmoid function, which maps the unit interval onto itself for a given decision noise parameter (note that this function differs from the logistic sigmoid above which maps the whole real line onto the unit interval).(14)
In systematic model comparisons, we compared variations in Eqs. (13) and (14), examining whether (i) subjects were more likely to integrate social and non-social information or used either source of information exclusively, and whether (ii) the decision noise in the mapping from beliefs to decisions (i.e., in Eq. (14)) was fixed or varied in time as a function of the estimated adviser's volatility. These variations are detailed in the section on “Model space” below.
Priors of the model parameters, namely for all models as well as for the HGF and for the RW model are listed in Table 2. We defined the priors based on the experimental design and pilot data. For parameters that were strictly bounded between 0 and 1, we chose the prior mean to be 0.5. For real-valued parameters, we chose prior means that represented values under which an ideal Bayesian agent would experience the least surprise about its sensory inputs (see the functions tapas_fitModel.m and tapas_bayes_optimal_binary_config.m in the HGF toolbox). The priors were chosen to be relatively uninformative (with large variances) to allow for substantial individual differences in learning and advice weighting.
In the HGF models, we also estimated participants' initial beliefs about the advice accuracy and the adviser's volatility, as well as their uncertainty about these two quantities. Parameters and states are estimated in spaces where they are unbounded. For example, parameters confined to the [0,1] interval are log-transformed and thus also estimated in an unbounded space. Given the priors over parameters and the input sequence, maximum-a-posteriori (MAP) estimates of model parameters were calculated using the HGF toolbox version 2.1. The code used is freely available as part of the open source software package TAPAS at http://www.translationalneuromodeling.org/tapas.
Optimization was performed using a quasi-Newton optimization algorithm [33]–[36]. The objective function for maximization was the log-joint posterior density over all perceptual and observation parameters, given the data and the generative model. To exclude the possibility that our Gauss-Newton gradient descent optimization could have been influenced by local minima of the log-joint objective function, we used two additional global optimization methods, a Gaussian Process optimization algorithm (GPO) [37] and a Markov chain Monte Carlo [38] sampling scheme.
Overall, our model space was structured hierarchically, as shown in Figure 2. We combined three alternative perceptual models with four potential response models, constituting a total of 12 models, , which are described in more detail below.
Although the assumptions of hierarchically coupled-learning were well founded, we also considered that participants' decisions could be explained by simpler non-hierarchical models. To examine this hypothesis, we included two model classes, which were both non-hierarchical. The first was a simplified version of the HGF (), in which the volatility at the third level was fixed to its prior mean and did not evolve over time (see Table 2 for the prior values used). This model assumed that participants ignored the instructions that the advisers' intentions might change in time, expecting negligible changes in log-volatility at the third level. The second model class was the classical RW model (), which assumed a fixed learning rate.
Concerning the response models, the key question was whether participants integrated social and non-social sources of information or relied exclusively on one of the two sources of information. Herein, we included two (reduced) response models, which proposed that participants considered either the advice alone or the cue alone when predicting the outcome. The first model was defined by setting to 1 (see Eq. (15) and Table 2), whereas the second only included the displayed winning probabilities with fixed to 0 (see Eq. (16) and Table 2).(15)(16)
Notice that the latter response model is not coupled to any of the perceptual models, because it suggests that participants do not learn about the validity of the advice and intentions of the adviser: on the contrary, they base their decisions only on the displayed winning probabilities.
Furthermore, we assessed two potential mechanisms of belief-to-response mapping (see Eq. 14), by including models which either assumed that (i) participants responded in accordance to their belief about advice accuracy but tainted by decision noise (“Decision noise” model family for models ), or that (ii) participants' decisions were based on their estimates of the volatility of the adviser's intentions (“Volatility” model family for models ).
The “Decision noise” model refers to parameter in Eq. (14), which represents the inverse of the decision temperature: as , the sigmoid function becomes steeper, approaching a step function (no decision noise) at . By contrast, the “Volatility” model family contains a time varying mapping of beliefs onto decisions. In this model set, the decision temperature parameter varies with the estimate of adviser volatility or . Hence, as the estimated volatility of the adviser's intentions decreases, the sigmoid function becomes steeper. This predicts that on trials when the player infers that the adviser's intentions are stable, he responds in accordance to his beliefs. As the volatility increases, the player becomes more uncertain of the adviser's intentions, and thus behaves in a more exploratory manner, resulting in a noisier mapping of belief-to-response probabilities. It is important to note that the mapping of beliefs onto actions is updated trial-wise, unlike in the case of the “Decision noise” response models, in which the link from beliefs to decisions is determined by a fixed, subject-specific parameter . Please see the video S1 for a demonstration of how the states of the perceptual model map onto decisions using equations (13) and (14), given all possible ranges of response model parameter .
Thanks to our factorial model space, we used family-level inference [39] to (i) determine the most likely class of perceptual models pooling across all response models, and (ii) the most likely response model class pooling across all perceptual models.
Before inferring on the model parameters, we evaluated the model space using Bayesian model selection (BMS). This procedure rests on computing an approximation to the model evidence or the probability of the data y given a model m [40]. The model evidence is the integral of the log-joint over the entire parameter space, which cannot be evaluated analytically. However, one can approximate the log model evidence with a lower-bound, the so-called (negative) free energy .
Alternative models can then be compared via the ratio of their respective evidences, i.e. their Bayes factor or equivalently, the difference in their log-evidences. At the group level, a group Bayes factor (GBF) can be computed by multiplying Bayes factors across subjects. The disadvantage of this procedure, however, is that it rests on the fixed-effects assumption that all participants' data are generated by the same model and variation is simply due to measurement noise [41]. This is not appropriate for our paradigm, as it emphasizes individual differences in social learning (e.g., by letting advisers choose their own strategy). This requires a random-effects BMS approach, where the model becomes a random variable in the population.
The random-effects BMS approach we use here rests on a hierarchical scheme introduced by Stephan et al. (2009), which estimates the parameters of a Dirichlet distribution of the probabilities of all models considered; in turn, these probabilities inform a multinomial distribution over model space. This makes it straightforward to compute the posterior probability that a given model generated the data for any randomly selected participant, relative to all other models considered (for details see [41]). Similarly, one can compute the “exceedance probability” that a particular model is more likely than any other model in the comparison set. In other words, the exceedance probability represents the amount of evidence that, in the population studied, a given model is more frequent than the others.
In case no model really stands out as a “winner” (i.e., no high exceedance probability), we can partition the model space, pool evidence over subsets of models that share a common feature (e.g., with and without a hierarchical level) and thus compare model subspaces or families, instead of single models (see [39] for details). This idea is essentially similar to factorial experimental designs in psychology where data from all cells are used to assess the strength of main effects and interactions. It amounts to specifying a partition , which splits the entire model set into subsets (model families). The subset contains all models belonging to family where there are models in the th subset. Due to the agglomerative property of the Dirichlet distribution, for any partition of model space into families, it is straightforward to define a new Dirichlet density reflecting this split (see Eq. 18 in [41]). The family probabilities are then given by:(17)where is the probability of each family occurring in the population. Exceedance probabilities can then be computed for each family, in the same way as for single models. They correspond to the probability that family is more frequent than any other family (of all families considered), given the data from all subjects:(18)
Because the conditional model probabilities sum to one over all models considered, this equation becomes particularly intuitive when model space is split into 2 families:(19)
To test the performance of BMS for our particular case, we generated 20 datasets for each of the 4 perceptual models considered under realistic levels of decision noise and using the prior means as parameter values. Thus, we augmented the softmax function in Eq. (14), which describes the mapping of beliefs onto actions, with a noise parameter :(20)
We then performed model inversion using the quasi-Newton optimization algorithm (in the same way as for the other models in this paper) and summarized the performance of BMS in terms of a confusion matrix (Figure S1). This matrix depicts the frequency of “correct cases” (where the model which generated the data has the highest exceedance probability of all models tested); here, rows denote the model that generated the data, and columns the model that was inverted. Thus, off-diagonal matrix elements indicate the frequency with which one generative model is “confused” with another.
As described in the Methods section, we examined the robustness of our BMS results in two ways. First, we used three different optimization schemes for inverting subject-wise models (quasi-Newton, MCMC, Gaussian process optimization). As shown by Table 3 (which lists the posterior probability of all 12 models under each optimization scheme), BMS results were consistent across all schemes. Second, the simulation results suggest that in the large majority of cases, the perceptual models that generated synthetic data could be recovered by model selection (Figure S1).
When comparing all 12 models against each other, random effects BMS showed that the three-level HGF augmented by the “Volatility” response model ( performed significantly better than the rest of the models in the majority of participants (exceedance probability = 0.99; Table 4 and S2–S3). Across the perceptual model family, the three-level HGF family outperformed non-hierarchical models () including the reduced HGF and RW models ( = 0.99; Table 5; Figure 4a). Taken together, these findings indicate that participants infer on two quantities, the advice accuracy and the volatility of the adviser's intentions, and incorporate the time-varying volatility estimates of the advisers' intentions into their learning about the advice.
The HGF quantifies subject-specific learning rates at distinct temporal scales. As an example, Figure 5 contains the learning rates for one individual subject. Here, the learning rate at the second level (transformed according to Eq. 9) increases as the reliability of the advice unexpectedly changes (blue line in Figure 5). The learning rate at the third level (green line; see Eq. 11), however, fluctuates more slowly, and increases when the adviser's intentions change from being consistently helpful to being misleading. This illustrates that the learning rate adapts to fluctuations in trial-by-trial advice reliability as well as to slower fluctuations in adviser intentionality. By contrast, the RW model assumes a constant learning rate α over trials. Figure 5 contrasts this estimate of α with the trial-wise learning rates provided by the HGF. This comparison suggests that, in a volatile environment, such a constant learning rate is necessarily too high on many trials.
The family of response models proposing that participants integrate both social and non-social sources of information (i.e., ) best explained participants' choices ( = 0.99; see Figure 4b and Table 6). That is, to predict the winning color, most participants relied on both the uncertain advice and the known outcome probabilities indicated by the visual pie chart. However, the posterior parameter estimate of was, on average, significantly smaller than 0.5 (; see Table 7), suggesting that participants weighted the visual pie chart information more than the advice.
As explained above, we considered two possibilities vis-à-vis how the player's beliefs might determine his actions trial-by-trial. A standard approach is to use a sigmoidal function (softmax or exponentiated Luce choice rule; see Eq. 14), which conveys decision noise whose amount is fixed across trials. This approach is used in our “Decision noise” response models with a fixed, subject-specific parameter . Alternatively, however, decision noise might vary dynamically across trials as a function of higher-order beliefs, such as the player's estimates of the adviser's volatility. This is represented in our “Volatility” response model family, which postulates that when the player estimates the adviser's intentions to be stable, he responds in close accordance to his beliefs. On the other hand, when the player's estimates of the adviser's volatility increase, he behaves in a more exploratory manner, resulting in a less deterministic (noisier) link between beliefs and responses. Our model comparisons indicated that the second perspective provided a better account of the data: the “Volatility” response model family clearly outperformed the “Decision noise” model family (; = 0.99; Figure 4c).
We used both classical multiple regression and variational regression to examine whether model parameter estimates of the winning model () predicted scores of relevant psychological traits, as measured by questionnaires, which the subjects completed three days prior to the experimental session. Model parameter estimates and predicted players' scores on the IRI (R2 = 50%, F = 6.03, ; log Bayes Factor (full versus null model) = 15.16; Table 8). Thus, participants with a stronger tendency to take into account the perspective of others during social interactions showed (i) stronger coupling between inference on advice accuracy and adviser volatility and (ii) more stable belief updates about advice reliability (and adviser trustworthiness) (Figure 6a). Notably, this link between parameter estimates and independent questionnaire scores for model was absent for the other models ( for the HGF with “Decision noise” and for the RW model).
Performance accuracy averaged at 73%±5% (mean ± standard deviation), indicating that, on average, the players reached the silver target and received CHF 10 bonus payment at the end of the game. Perceptual model parameter and response model parameter predicted participants' performance accuracy (R2 = 40%, F = 9.41,; log Bayes Factor (full versus null model) = 17.59). Taken together, these results reflect that participants who perceived the adviser's intentions to be more stable and who weighted the social information more during decision-making performed better in the task (Table 8; Figure 6b–c).
Advisers also reached, on average, the silver target and received CHF 10 payment at the end of the game. Across advisers, their recommendation was correct in 74%±9.8% of all trials.
On debriefing, 4 of the 16 advisers reported a general intention to help the players during the task; these advisors provided correct recommendations on 85%±9.2% of the trials (note that the information available to the advisers predicted wins with 80% accuracy). The majority of the advisers (9 out of the 16), however, aimed to increase their final pay-off and provided correct recommendations on only 74%±1.6% of the trials. On the one hand, the players who interacted with more helpful advisers weighted the advice more as indexed by larger values and perceived the advisers' intentions as more stable as indexed by reduced values (; Figure 6c–d). On the other hand, the players who interacted with advisers whose intentions changed over the course of the game exhibited significantly larger values than the rest (Figure 6e). This suggests that there was a more pronounced coupling between the two learning levels (advice accuracy and adviser volatility) during interactions with advisers whose intentions were changing.
To demonstrate the interpretability of our model parameter estimates, we asked each player eight times at random points during the game to explicitly rate the advisers as “helpful”, “uninformative” or “misleading”. These ratings were coded such that “helpful” corresponded to a probability of accurate advice of 1, “uninformative” to a probability of 0.5, and “misleading” to a probability of 0. To relate the participants' ratings to the estimates of advice reliability as inferred from the model, we used each player's ratings as the outcome variable in a general linear model with the explanatory variable being the prediction about the advice reliability (. This proved to be highly significant (t = 5.92, ) in a second level random effects regression analysis (see Figure S4). In brief, the state estimates of our model correspond well to the players' overtly expressed beliefs about the adviser's intentions during the game. Notably, the same analysis using the value of the advice as estimated by the RW model did not yield significant results (). Altogether, this corroborates our model comparison results and provides construct validity for our model.
Distinct learning performance was observed in the control task (where the adviser was blindfolded and presented his advice by holding up a card sampled from a series of card decks, each of which was, on average, either 80% or 20% accurate). The players performed significantly worse in this task compared to the socially interactive task (t (15) = 5.48, ), with performance accuracy averaging at 64%±2.6%.
In this task, the BMS yielded different results compared to the socially interactive task (see Table 9). More precisely, the three-level HGF family () still outperformed non-hierarchical models (), such as the reduced HGF and the RW model ( = 0.98), suggesting that participants did incorporate time-varying estimates of volatility (resulting from the switches among card decks) into their beliefs about the advice accuracy. Furthermore, the integrated response model family (, which proposed that participants weigh both social and non-social sources of information, explained participants' responses better than reduced response models () according to which subjects relied on one source of information only ( = 0.99). In contrast to the social setting, in the control task, the response model prescribing volatility-driven mapping of beliefs to decisions did not differ from the model that utilized a single decision noise parameter ( = 0.54). In other words, unlike in the social task, decision noise might not change across trials as a function of adviser volatility estimates.
With respect to the posterior parameter estimates (see Table 7), there were notable differences between the two tasks: In the control task, parameter averaged at 0.28±0.11; this was significantly lower than in the social task (t (15) = 2.44, ), indicating that the players weighted the social (but unintentional) advice significantly less than in the social task. That is, although the card decks were more informative (80% predictability of wins/losses) than the non-social cue (55–75% predictability), the players relied more on the binary lottery information to predict the outcome. The difference in the performance and the response model parameters of these two tasks suggests that participants performed better and relied more on the adviser's recommendations when the adviser intentionally issued the advice.
In each participant, the model parameters describe an individual learning trajectory (see Figure 7). As we debriefed each adviser explicitly about the strategy that he employed during the task, we were able to use these debriefings to examine how model-based quantification of individual learning and inference reflected the players' adaptive responses to the advisers' behavior during the game.
Four out of the 16 advisers provided accurate advice throughout the game with advice reliability averaging at 85%±9.2%. Upon debriefing, they reported that they aimed for the silver range from the beginning, as they deemed it fair for both participants to reach the silver target. The players who interacted with this subset of advisers perceived their intentions to be stable over time and weighted the advice more, as indexed by larger values (see Figure 6d). An example of such a player is given in Figure 7a (subject SL_010), where the trajectory of estimated advice reliability indicates that this player's estimate of advice accuracy stayed close to 90% throughout the game. For this subject, the estimate of was 0.54, indicating that he relied more on the advice than the non-social cue when making his predictions.
By contrast, another 3 of the 16 advisers were consistently uninformative, exhibiting an average of only 56%±6.6% advice reliability. Indeed, when they were debriefed, they described that in order to reach the gold range, they attempted to confuse the player from the beginning, preventing his progress bar to increase significantly throughout the game and maximising their own chances to reach gold. This turned out to be a successful strategy, as the advisers who used this strategy were the only ones who reached the gold target. The players who interacted with this subset of advisers showed a very distinct trajectory of learning from those discussed above. First, was low in all these players averaging at 0.19, indicating that they (rightfully) assigned little weight to the advice and relied exclusively on the pie chart. A representative subject is shown in Figure 7b (subject SL_005), for whom the estimate of was 0.12. Furthermore, a high value of meta-volatility indicated that this participant perceived the adviser's intentions as becoming more and more volatile over the course of the interaction. Additionally, high levels of indicated fast updating of beliefs about advice correctness over trials and independently of estimates of adviser volatility. Again, this is a sign of adaptive behavior: because the adviser is consistently uninformative (random) in his advice, a high tonic learning rate for updating estimates about advice validity in is appropriate. In other words, this scenario describes an agent who perceives the adviser's intentions as stochastic because the advice is uninformative. In this scenario, the participant necessarily performs poorly because he must base his decisions almost exclusively on the non-social cue.
The largest subset of advisers (9 out of 16) used a strategy, which reflected the change in incentives induced by the payoff scheme: advisers were helpful at the beginning of the game until the players' progress bar reached their gold range. From this point on, advisers began to mislead the players, preventing them from moving beyond the gold range. Once the players detected this change in intentionality, their score began increasing again. This elicited another switch in the advisers' strategy, who now resumed a helpful attitude in order to at least reach the silver range. On average, the recommendations of these advisers were 74%±1.6% accurate. Players who interacted with these advisers exhibited a learning trajectory that reflected the advisers' dynamic incentives (see Figure 7c). For example, for the representative subject shown in Figure 7c (subject SL_013), steadily reached 0.9 after the first 80 trials, with a concomitant decrease in estimates of the adviser's volatility . Once the adviser's intentions changed, the player updated his beliefs accordingly, as reflected by an increase in the learning rate in and larger updates of . In this scenario, the player's adaptive behavior takes into account both the volatility of the adviser's intentions and the accuracy of his advice. This is reflected by high values of the estimates for κ, ω, and .
Finally, to illustrate the capacity of our model-based approach for characterizing individual differences, we show an unusual subject in Figure 7d (subject SL_015). This player did recognize a change in the adviser's intentions halfway through the game but was much slower in updating his estimates of advice accuracy and adviser's volatility than the subject discussed above. This is because his prior beliefs were close to how the adviser actually behaved for the first half of the game and because low estimates of meta-volatility prevented a rapid response to the change in the adviser's intentions. In other words, this participant remained relatively confident about his prior estimate of the adviser's volatility and expected to see little change over the course of the social interaction.
The question of how we infer on others' intentions is a fundamental computational problem during social transactions. To examine this process, we extended a paradigm introduced by [14], turning it into an interactive social decision-making game in which each participant was assigned to a “player” or an “adviser” role. Critically, the game was designed to ensure that the adviser's incentives to cooperate or deceive the player varied, thus making his intentions volatile.
While our paradigm was inspired by the previous work of Behrens and colleagues [14], [42], it introduced two important advances. First, whereas Behrens et al. made subjects believe that the computer-generated advice was provided by a human being, our paradigm used real participants without any deception. This provides ecological validity and eschews potential ethical concerns, which makes the recorded trials from this paradigm more widely applicable, e.g., for future patient studies. Secondly, our paradigm allows for a wide range of interactions between agents, as both the adviser and the player are not restricted to employ specific strategies during their interaction. The player can rely on the binary lottery information (the non-social cue), the advice, or both when selecting his choices. Furthermore, on every trial, the adviser can also choose to provide either helpful or misleading advice, depending on whatever strategy he may be employing; in turn, these differences in strategy across advisers elicit differences in the adaptive behavior of the players.
To explain the ensuing variability in adaptive behavior across subjects, we modeled the players' learning using a systematic set of alternative models that factorially combined different models of learning behavior (“perceptual models”) and decision-making (“response models”). Using Bayesian model selection, we demonstrated that a hierarchical Bayesian model (the hierarchical Gaussian filter, HGF) with three levels best described the players' learning in the task. This suggests that participants updated their beliefs about advice reliability depending on an ongoing estimate of the volatility of the adviser's intentions, and that this estimate of volatility directly informed the trial-wise decisions. This three-level HGF outperformed simpler non-hierarchical models (such as Rescorla-Wagner), indicating that during social exchanges, participants employ a multilevel model of their environment and are capable of learning how others' intentions to be helpful or misleading fluctuate over time. These higher-order expectations are in turn exploited to update trial-by-trial predictions about advice reliability.
An important contribution of this paper is the translation of a recent Bayesian framework for comparing alternative cognitive models [15], [16], [22] to the domain of social interactions. The implementation of this framework in the present study, however, has one significant limitation: The present models aimed to explain only the players' learning during the game, and not the advisers'. That is, they neglected the recursive process of perspective-taking (in other words, the player's belief about the adviser's belief about the player's belief etc.), which occurs in many social situations. Recent studies (see [2], [12], [13], [43]) used recursive theory-of-mind models to explain social inference in cooperative or multi-round trust games. These models propose that the expected value of a given action (e.g., to cooperate or to compete and to choose equitable or unfair offers) is a function of the other agent's strategy. Thus, players optimize their strategies or their depth of recursive reasoning by taking into account their opponents' future actions. For example, Yoshida and colleagues (2010a) showed that players who employ higher-order strategies, which take into account the opponents' future actions, forgo immediate rewards for options that lead to higher pay-off but require multi-player cooperation.
One important future direction of our work is to extend the modeling of hierarchically coupled beliefs to take the depth of recursive perspective-taking into account. Having said this, the recursive depth of social inferences is typically limited [2], [13]. For example, Xiang et al. (2012) classified the depth of subjects' reasoning during an investor-trustee game. Approximately half of 195 investors were classified as strategic level 0 players, suggesting that they do not simulate their partner's play, while the other half employed either level 1 or level 2 depth-of-reasoning. Yoshida et al., 2010a also reported significant variability in recursive depth across individuals. In the present study, as described in the Results section, more than half (9/16) of the advisers reported to have adopted a fixed strategy (either consistently helpful or consistently random throughout the game) and are thus unlikely to have engaged in recursive perspective-taking. For the remaining advisers, it is possible that modeling their belief updating processes (in addition to those of the players) would lead to an even better prediction of the players' behavior. We will test this possibility in future work, contrasting models with and without the representation of recursive interactions.
As it did not incorporate recursive perspective-taking, our study focused on modeling the downstream consequences of the differential strategies that advisers employed. The players' belief-updating process reflected the advisers' policy and determined how much they were willing to take the advisers' suggestions into account during decision-making. We found that players who interacted with consistently helpful advisers perceived their intentions to be stable over time and thus weighted their advice more when predicting the outcome, as reflected by reduced values in the meta-volatility parameter and larger values, respectively. Furthermore, players who interacted with advisers, whose intentions were changing to maximize their own winnings, showed more pronounced values. This result suggests that the two hierarchical learning levels were more strongly coupled in this subset of participants, and that the volatility estimate was used to update the beliefs about advice accuracy. Unlike in the case of consistently misleading advisers, in this particular social exchange, the volatility of the advisers' intentions was more traceable. Thus, players could benefit from inferring on the volatility of the advisers' intentions to predict the advice accuracy.
Beyond reflecting the adviser's policy in the parameter estimates, our model exhibited construct validity in two ways: First, its posterior parameter estimates predicted participants' scores on the IRI, a questionnaire, which they completed prior to participation in the study. Players, who described themselves as proficient perspective-takers, exhibited a more stable model of the adviser as reflected by the less pronounced tonic component of the learning rate. Second, the model's posterior parameter estimates also predicted the participants' explicit ratings of the advisers' helpfulness throughout the game. Notably, this relationship was specific for the hierarchical Bayesian model while the parameter estimates from the competing RW model did not show this predictive capacity.
As described above, model comparison indicated that the participants' behavior was best explained by a hierarchical model in which estimates of volatility (of the adviser's intentions) played a key role for belief updating. Furthermore, beyond inference and with respect to the translation of beliefs to decisions, we found that a response model in which participants' estimates of the volatility of the advisers' intentions determined their trial-wise decisions explained participants' choice behavior best. That is, the mapping from beliefs to choices was increasingly deterministic the more the player considered the adviser's intentions to be currently stable. By contrast, when the player's estimates of the adviser's volatility increased, the relation between beliefs and decisions became more stochastic and the player exhibited a more exploratory behavior. This result demonstrates the direct relevance of volatility estimates for determining trial-by-trial variability of decisions; note that this is distinct from (and complementary to) our findings on the role of volatility for learning and inference, described in the context of comparing different perceptual models above. The finding that volatility is an important factor determining trial-by-trial choice variability goes beyond previous studies, which examined the impact of volatility with respect to inference only (e.g., [14], [22], [25], [44]). Moreover, in the context of social learning, these results stress the deployment of a hierarchical model and a key role of volatility estimates for both inference and decision-making. These are important in that they complement concepts of social learning, which emphasize the role of simple heuristics (e.g., [45]) or refer to non-hierarchical reinforcement learning (e.g., [46]).
Similar to what Behrens and colleagues (2008) observed, we found that participants did not base their decision on a single source of information, but integrated the advice with information from the visual pie chart, which was also probabilistic but had a known outcome distribution. That is, the uncertainty of the information provided by the pie chart was directly given on each trial, whereas the uncertainty of the advice had to be estimated online. This can be related, to some degree, to the distinction between risk and ambiguity [47]–[49].
Our modeling results show that participants were able to trade-off between these different forms of uncertainty depending on the type of adviser they faced (see Figures 6d–e and 7): when interacting with generally helpful advisers, most players considered the advice strongly because, on average, it was more accurate than the visual pie chart. However, when they interacted with advisers who deliberately showed consistently uninformative (random) behavior, participants tended to discount their recommendation and relied more strongly on the visual pie chart. This is remarkable since it means that players did not display a uniform tendency to avoid ambiguity; instead, ambiguity aversion was restricted to interactions with an unhelpful adviser.
Additionally, we found that the different sources of information (cue and advice) did not receive equal weight during decision-making. Consistent with previous findings [50], we observed that participants relied more on the non-ambiguous information (i.e., the non-social cue) compared to the advice. Previous models [51], [52] describing how people integrate social and non-social sources emphasized the importance of ambiguity that is intrinsic to social exchange: We are uncertain about how uncertain our appraisal of the other agent's intentions is.
Previous work on uncertainty in repeated advice taking showed that, surprisingly, decision-makers do not become more confident in their choices with increasing advice accuracy [53]. Although we did not explicitly ask subjects to rate confidence or uncertainty, our modeling results did take into account how their inferred estimates of uncertainty (about the adviser's intentions) informed their trial-wise decisions.
Furthermore, analysis of a control condition in which trial-wise advice was randomly sampled (by a blindfolded adviser) from several decks of cards with either 80% or 20% accuracy suggested that participants relied more on the advice when it was intentional, as opposed to unintentional. This behavior was observed even though it was perfectly possible to extract predictive information from the card decks with the same accuracy as from a helpful adviser. Since players based their decisions more on the visual pie chart and did not take advantage of the advice, their performance was significantly lower than in the social condition.
Beyond the results per se and their implications for concepts of social learning, the modeling approach in this paper, with its emphasis on inter-individual variability in inference and decision-making, may serve useful for future studies of social learning. To facilitate this, the HGF and the BMS routines are freely available as open source MATLAB code (the HGF can be found at www.translationalneuromodeling.org/tapas; the BMS routines are part of the SPM software package: www.fil.ion.ucl.ac.uk/spm).
Finally, we believe that the approach presented here has potential for characterizing mechanisms of maladaptive behavior in individual patients. The present study in healthy volunteers provides a proof of concept how individual mechanisms can be elucidated in the context of social interactions, a domain where many psychiatric disorders, including schizophrenia, are characterized by particularly salient deficiencies [54]. For example, many patients with schizophrenia exhibit a negative attribution bias about others' intentions, which reflects the finding that negative information is perceived as more diagnostic of another person's true character than positive information [55]–[57]. One attractive option is to use models as the one described in this study for computational phenotyping of patients from heterogeneous disorders [58]–[60]. For example, patients may show a diminished ability to dynamically infer on the intentions of others for different reasons: they may have overly tight prior beliefs about others' motivations, or they may suffer from an abnormality in belief updating mechanisms, which in turn could be due to aberrant computations of prediction error, precision or both (see Eq. 6 above). In other words, models of cognition such as the one introduced here and in previous studies may prove useful to propose potential nosological dimensions with mechanistic interpretability and disambiguate alternative mechanisms in individual patients through model selection [61]. This study serves as a precursor for future neuroimaging studies, in which we hope to investigate neuronal mechanisms of social learning and tracking the volatility of another agent's intentions.
|
10.1371/journal.pgen.1006311 | The Multivesicular Bodies (MVBs)-Localized AAA ATPase LRD6-6 Inhibits Immunity and Cell Death Likely through Regulating MVBs-Mediated Vesicular Trafficking in Rice | Previous studies have shown that multivesicular bodies (MVBs)/endosomes-mediated vesicular trafficking may play key roles in plant immunity and cell death. However, the molecular regulation is poorly understood in rice. Here we report the identification and characterization of a MVBs-localized AAA ATPase LRD6-6 in rice. Disruption of LRD6-6 leads to enhanced immunity and cell death in rice. The ATPase activity and homo-dimerization of LRD6-6 is essential for its regulation on plant immunity and cell death. An ATPase inactive mutation (LRD6-6E315Q) leads to dominant-negative inhibition in plants. The LRD6-6 protein co-localizes with the MVBs marker protein RabF1/ARA6 and interacts with ESCRT-III components OsSNF7 and OsVPS2. Further analysis reveals that LRD6-6 is required for MVBs-mediated vesicular trafficking and inhibits the biosynthesis of antimicrobial compounds. Collectively, our study shows that the AAA ATPase LRD6-6 inhibits plant immunity and cell death most likely through modulating MVBs-mediated vesicular trafficking in rice.
| Plants have evolved sophistical immunity system in fighting against pathogenic micro-organisms including bacteria, fungi and oomycetes. Upon perception of pathogens, the immune system activates rapid cell death, characterized as a form of hypersensitive response typically in and around the infection sites to restrict pathogen invasion and prevent disease development. Recent studies have suggested that MVBs-mediated vesicular trafficking might play key roles in plant immunity and cell death. However, the molecular regulation is poorly known. By using the lesion resembling disease (lrd) mutant, lrd6-6, which exhibits autoimmunity and spontaneous cell death, we characterized LRD6-6 as a MVBs-localized AAA ATPase. We found that the ATPase LRD6-6 was required for MVBs-mediated vesicular trafficking and inhibited the biosynthesis of antimicrobial compounds for immune response in rice. Both the ATPase activity and homo-dimerization of LRD6-6 were essential for its inhibition on immunity and cell death. The catalytically inactive ATPase, LRD6-6E315Q, played dominant-negative effect on inhibition of immunity in plants. In addition, the LRD6-6 protein co-localized with the MVBs-spread marker protein RabF1/ARA6 and also interacted with ESCRT-III components OsSNF7 and OsVPS2. In summary, our study has shown that the AAA ATPase LRD6-6 inhibits plant immunity and cell death most likely through modulating MVBs-mediated vesicular trafficking in rice.
| Plants are exposed to a vast diversity of micro-organisms such as bacteria, fungi and oomycetes. To protect themselves from pathogenic plant–microbe interactions, plants have developed a sophisticated innate immunity system [1]. Pattern triggered immunity (PTI) and effector triggered immunity (ETI) are two major layers of an immunity system that shares many common responses to pathogen infection including protein phosphorylation, hormonal change, ion fluxes change, production of reactive oxygen species (ROS), synthesis of antimicrobial compounds, transcriptional activation of pathogenesis-related (PR) genes and cell-wall reinforcement via oxidative cross-linking of cell-wall components and deposition of lignin [2–4]. Cell death, which plays a central role in many plant processes, has been observed in both PTI and ETI [3, 5, 6]. Upon perception of pathogens, the immunity system activates rapid cell death, characterized as a form of hypersensitive response (HR) typically in and around the infection sites to restrict pathogen invasion and prevent disease development [7, 8].
The lesion resembling disease (lrd) mutants carry a cell death phenotype that mimics HR without pathogen attack and are useful tools for studying immunity and cell death [9, 10]. A large number of lrd mutants characterized by enhanced immunity and cell death have been identified in maize [11], Arabidopsis [12], rice (Oryza sativa) [13], barley [14] and Brassica oleracea [15]. In rice, more than 10 genes encoding different proteins have been cloned. These include the heat stress transcription factor SPL7 [16], E3 ubiquitin ligase SPL11 [17], zinc finger protein OsLSD1 [18], hydroperoxide lyase OsHPL3 [19], kinase OsPti1a [20], MAPKKK OsEDR1 [21], NPR1-like protein OsNPR1 [22], acyltransferase-like protein SPL18 [23], cytochrome P450 family protein SPL1 [24], fatty-acid desaturase OsSSI2 [25], clathrin-associated adaptor protein complex 1 medium subunit μ1 (AP1M1), SPL28 [26], coproporphyrinogen III oxidase RLIN1 [27], putative splicing factor 3b subunit 3 (SF3b3) protein SPL5 [28] and double-stranded RNA binding motif containing protein OsLMS [29]. Some of them have been studied in molecular regulation of immunity and cell death, including Spl11, which encodes an E3 ubiquitin ligase and is associated with SPIN6 and OsRac1 to negatively modulate immunity and cell death [17, 30]. However, the mechanisms of immunity and cell death deployed by lrd mutants remain largely unknown in rice.
Previous studies have shown that protein trafficking mediated by multivesicular bodies (MVBs) is associated with immunity in plants [31, 32]. Upon perception of ligand flagellin flg22, the Arabidopsis immune receptor FLAGELLIN SENSING 2 (FLS2) present at the plasma membrane is internalized under regulation of the ESCRT-I components VPS37-1. These results suggest that the protein endocytic sorting at the MVBs is critical for FLS2-mediated immunity [33]. Rice SPL28 inhibits immunity and cell death likely through regulation of post-Golgi trafficking [26]. When Spl28 is disrupted, rice plants display enhanced immunity and exhibit cell death constitutively [26]. These studies indicate that MVBs-mediated vesicular trafficking may participate in regulation of immunity and cell death in plants.
The AAA (ATPase associated with various cellular activities) ATPase family proteins contain conserved ATPase domains spanning 200–250 residues which cover the Walker A, Walker B and the SRH (Second Region of Homology) motifs that distinguish them from classic p-loop NTPases [34–36]. In the process of MVBs biogenesis, the AAA ATPases are used to disassociate the ESCRT-III complex from the membrane by providing required energy [37, 38]. These ATPases participate in diverse cellular processes including membrane fusion, proteolysis and DNA replication, and MVBs-mediated vesicular trafficking [34]. Recent studies have determined that AAA ATPases are also involved in immunity in both mammals and plants. For example, the human AAA ATPase p97/valosin-containing protein (VCP) is an important host factor in antiviral immunity [39]. The human VPS4A functions as a tumor suppressor in hepatoma cells [40] and VPS4B is involved in drug resistance in multiple myeloma cells [41]. The tobacco AAA ATPase NtAAA1 negatively regulates defense response against the invasion of Pseudomonas syringae [42, 43]. In Arabidopsis, the AAA ATPase AtOM66 functions as a positive regulator in immunity and cell death [44]. AtSKD1, homologous to VPS4A and VPS4B, has been reported to contribute to vacuolar maintenance and MVBs-mediated vesicular trafficking [45] and likely regulates immunity in Arabidopsis [46]. However, little is known about the role of AAA ATPases in immunity in rice.
In this study, we report the identification and characterization of the rice lrd6-6 mutant, which shows enhanced immunity and spontaneous cell death. Map-based cloning reveals that Lrd6-6 encodes an AAA ATPase, and disruption of the AAA ATPase LRD6-6 leads to autoimmunity and spontaneous cell death in the lrd6-6 mutant. The ATPase activity and homo-dimerization of LRD6-6 is essential for its inhibition of immunity and cell death in rice. A catalytically inactive mutation, LRD6-6E315Q, plays dominant-negative effect in plants. The LRD6-6 protein mainly spreads on MVBs and interacts with ESCRT-III components OsSNF7 and OsVPS2. Further analysis reveals that biosynthesis of antimicrobial metabolites, including lignin and phytoalexins, is highly activated and the process of the MVBs-mediated vesicular trafficking is largely dysregulated in the lrd6-6 mutant, suggesting that the accumulation of antimicrobial metabolites resulting from the disruption of the LRD6-6 ATPase is tightly linked with the disordered processes of MVBs-mediated vesicular trafficking. Collectively, our study reveals that the AAA ATPase LRD6-6 regulates immunity and cell death likely by modulating the MVBs-mediated vesicular trafficking process.
The lrd6-6 mutant was generated from tissue culture of rice cv. Kitaake. Plants of lrd6-6 exhibit reddish-brown lesion spots on the leaves about two weeks after sowing (Fig 1Aa) and the lesion spots expand through the entire plants along with development (Fig 1Ab and 1Ac). The lrd6-6 plants also exhibit lesion spots when grown in sterile ½ Murashige and Skoog (MS) medium in Solo cups (S1 Fig), suggesting that the occurrence of lesion spots in lrd6-6 is spontaneous in the absence of any biotic or abiotic stresses. The leaves of lrd6-6 shaded with silver paper also show lesion spots (S2 Fig), indicating that lesion spots formation in lrd6-6 is light independent. The lesion spots lead directly to the decrease of photosynthetic pigments in the lrd6-6 mutant because the contents of chlorophyll a (Chla), chlorophyll b (Chlb) and carotenoid (Car) are dramatically reduced (S3 Fig) in lrd6-6 after lesion appearance.
To determine if cell death occurred in the lrd6-6 mutant, we stained rice leaves using trypan blue. The blue staining spots were present in 8-d-old lrd6-6 plants even before initiation of lesion spots, and more blue staining spots accumulated along with plant development (Fig 1B, upper panel). However, no blue staining spots appeared in leaves of Kitaake (Fig 1B, upper panel). Further, DAB (3, 3’-diamiobenzidine) analysis detected extensive stains in leaves of lrd6-6 before lesion spots appearance. On the contrary, almost no staining occurred in Kitaake along with plant development (Fig 1B, lower panel), indicating that excess hydrogen peroxide had accumulated in the lrd6-6 mutant compared with Kitaake. We then sectioned the leaves of lrd6-6 and Kitaake, and observed under transmission electron microscopy. The subcellular structures were severely degraded in the lesion-spotted parts of leaves but not in other leaf parts in absence of lesion spots in lrd6-6 or the equivalent parts of Kitaake (S4 Fig). These results suggest that spontaneous cell death occurs in the lrd6-6 mutant, which results in the formation of lesion spots.
Previous studies have shown that cell death in plants is usually mediated by enhanced immunity [12, 47, 48]. We therefore determined the expression levels of immunity related genes, such as the PR genes, Betv1, OsNPR10, OsPR1a and 04g10010 [30, 49, 50]. The expression of these PR genes all increased in lrd6-6 compared with Kitaake plants, predominantly after the presence of cell death (Fig 1C). We then challenged lrd6-6 plants with the fungal pathogen Magnaporthe oryzae (M. oryzae) and the bacterial pathogen Xanthomonas oryzae pv oryzae (Xoo), which cause blast and bacterial blight diseases, respectively. The disease lesion length and the number of lesions on leaves inoculated with M. oryzae strains (ZB25, Zhong1 and ZE-1) compatible with Kitaake, were all dramatically reduced in the lrd6-6 mutant compared with Kitaake (Fig 1D). When inoculated with compatible Xoo strains (P2, P4, P5, P6 and Xoo-4), the lrd6-6 mutant also exhibited enhanced resistance, showing much shorter disease lesion length than Kitaake plants (Figs 1E and S5). Collectively, our results indicate that the rice lrd6-6 mutant possesses enhanced immunity, which may be mediated by the cell death in the lrd6-6 mutant.
Since the lrd phenotype caused by cell death in lrd6-6 did not co-segregate with the hygromycin (Hyg) gene (S6 Fig), we presumed that the spontaneous cell death in the lrd6-6 mutant might have resulted from tissue-culture induced somatic mutation during transformation. We thus developed three F2 populations and performed genetic analysis on the lrd6-6 locus. Genetic analyses showed that the lrd6-6 phenotype was controlled by a single recessive nuclear locus (S1 Table). Next, map-based cloning of the Lrd6-6 gene was performed using 344 F2 individuals with the spontaneous cell death phenotype from the cross between the lrd6-6 mutant and rice 02428. The Lrd6-6 locus was first mapped in the interval with a physical distance of 2.54 Mb between the InDel markers I4-2 and I8 on chromosome 6 (Fig 2Aa). By using more markers to analyze 2175 F2 individuals with the cell death phenotype, the Lrd6-6 locus was then delimited on the genomic region within 93 kb between RM8075 and RM587 (Fig 2Aa). Next, we sequenced the genomic DNA sample bulked with 30 BC2F3 individuals with cell death phenotypes using a whole-genome resequencing approach. The genomic DNA of Kitaake was also sequenced as a control. When comparing the sequences of the 93 kb between the bulked DNA sample and the Kitaake control DNA, we found that a 1446 bp DNA fragment from gene LOC_Os06g03940 (RGAP ID from http://rice.plantbiology.msu.edu, abbreviated to Os06g03940 hereafter) that spans four exons was tandemly repeated in Os06g03940 in the bulked DNA sample (Fig 2Ab). Sequencing of cDNA revealed that this 1446 bp tandem repeat resulted in an insertion of 534 bp in the protein-coding sequence of Os06g03940 that might disrupt the gene function in the lrd6-6 mutant (S7 Fig). These results indicate that the spontaneous cell death phenotype in the lrd6-6 mutant was likely resulted from the insertion of 1446 bp tandem repeat in the Os06g03940 gene, and thus Os06g03940 encoding an AAA ATPase is likely the target gene of Lrd6-6.
We carried out a knockdown experiment to confirm that the lrd6-6 mutation was caused by Os06g03940. We amplified a unique segment (Seg I, 445 bp) covering nucleotides 557 to 1001 of the Os06g03940 open reading frame (ORF) (S8 Fig) and used this segment to create an RNA interference (RNAi) construct, pANDA–Os06g03940Ri. This segment shows only approximately 20.12% identity with the closest homologous gene Os01g04814 (S8 Fig). This construct was then introduced into Kitaake through Agrobacterium-mediated transformation. We found that all 33 transgenic plants with suppressed Os06g03940 expression displayed spontaneous cell death similar to the lrd6-6 mutant (Figs 2B and S9A). In these plants, the expression of Os01g04814, which shared the highest identity with Os06g03940 in cDNA sequence, was not suppressed (Fig 2Bd), suggesting that specific silencing of Os06g03940 resulted in the spontaneous cell death phenotype.
To further verify the result obtained by RNAi analysis, we then cloned an 11.5 kb genomic DNA fragment harboring the native promoter and full coding region of the gene Os06g03940 from rice Nipponbare and placed it in the binary vector pCactN-XG to create the construct pCactN-XG–Os06g03940-11.5kb. The DNA fragment Os06g03940-11.5kb was then introduced into the lrd6-6 mutant through a similar Agrobacterium-mediated transformation approach. All 12 transgenic lines carrying the transgene Os06g03940-11.5kb no longer exhibited the spontaneous cell death phenotype (Figs 2C and S9B). This reveals that the Os06g03940-11.5kb transgene restores the spontaneous cell death phenotype in the lrd6-6 mutant to that of the wild type Kitaake. Taken together, these results clearly demonstrate that Os06g03940 is the target gene of lrd6-6, in which the mutation with the inserted 1446 bp tandem repeat was responsible for the spontaneous cell death phenotype.
Recently, Fekih et al. reported that a G–>A base substitution resulted in a premature translation termination in the Lmr gene (RAP ID Os06g0130000; http://rapdb.dna.affrc.go.jp), which is the same gene Os06g03940 from RGAP according to gene ID conversion through ID Converter (http://rapdb.dna.affrc.go.jp/tools/converter), and also led to the spontaneous cell death phenotype in rice [51]. Similar to lrd6-6, the lmr mutant displayed elevated PR gene expression and enhanced disease resistance compared with wild type rice Hitomebore [51]. Fekih and his colleagues also used the RNAi approach targeting Seg II of the gene, which spans a region different from Seg I used in our study (S8 Fig), and a complementary test by overexpressing the full-length cDNA of Lmr to confirm their results [51]. Together, these results confirm that disruption of Os06g03940 leads to the lrd6-6 phenotype.
To determine the expression pattern of Lrd6-6, we respectively sampled the root, stem, leaf and panicle at the two-, four-, six-leaf and mature stages of Kitaake and determined the transcript levels of Lrd6-6 in these tissues. The quantitative reverse transcription-PCR (qRT-PCR) analysis showed that Lrd6-6 was expressed in all these tissues with predominance in leaves (S10 Fig).
To investigate the subcellular localization of LRD6-6, we generated construct p35S-Lrd6-6–GFP expressing the LRD6-6–GFP fusion protein and performed a transient expression assay using Nicotiana benthamiana through an agroinfiltration approach. The GFP signal was punctate in the cytoplasm in the leaf of N. benthamiana expressing LRD6-6–GFP fusion protein whereas, as expected, the GFP fluorescence was distributed in the cytoplasm and nucleus in the leaf expressing GFP alone (S11A Fig). The distribution of the LRD6-6–GFP fusion protein was reminiscent of the localization of the Arabidopsis LRD6-6 homologous protein AAA ATPase SKD1, which was previously shown to be located in MVBs [38]. By using bombardment-mediated transformation, we also observed a punctate pattern of GFP signals in onion epidermal cells expressing the LRD6-6–GFP fusion protein (S11B Fig). The Lrd6-6–GFP transgene was able to restore the lrd6-6 plants to the wild type level (S12 Fig) suggesting that the LRD6-6–GFP fusion protein functions similarly as LRD6-6. The MVBs-localized pattern of the LRD6-6–GFP protein highly suggests that LRD6-6 localizes in MVBs.
The Rab GTPase RabF1/ARA6 protein has been shown to locate on the peripheral membrane of the MVBs and has been widely used as the specific marker protein for plant MVBs-localization analysis [52, 53]. We transiently expressed the fused proteins, RabF1/ARA6–GFP and RabF1/ARA6–RFP respectively in N. benthamiana cells through the same agroinfiltration approach and found that both RabF1/ARA6–GFP and RabF1/ARA6–RFP are present in punctate patterns in the cells as expected (S13 Fig). When treated with chemical wortmannin which is able to change MVBs into ring-like structures through inhibiting phosphatidylinositol 3-kinase (PI3) activity [54, 55], the punctate GFP or RFP signals in the N. benthamiana cells expressing RabF1/ARA6–GFP or RabF1/ARA6–RFP were respectively converted into ring-like structures (S13 Fig). These results suggested that RabF1/ARA6–GFP and RabF1/ARA6–RFP were MVBs-localized and could be used as control for MVBs-localization analysis on LRD6-6. Then, we co-transformed LRD6-6–RFP (RFP fused on the C-terminus of LRD6-6) with RabF1/ARA6–GFP, LRD6-6–GFP (GFP fused on the C-terminus of LRD6-6) or YFP–LRD6-6 (YFP fused on the N-terminus of LRD6-6), respectively, with RabF1/ARA6–RFP. Consistently with prior results, the punctate fluorescence signals of the fusion proteins, LRD6-6–GFP, LRD6-6–RFP and YFP–LRD6-6, all co-located with fluorescence signals derived from RabF1/ARA6–GFP or RabF1/ARA6–RFP protein but not the signals of chlorophyll (Figs 3 and S14). Together, our results clearly show that the LRD6-6 protein resides mainly on MVBs which is different from the results of Fekih et al. showing that LRD6-6–GFP was localized in chloroplast [51].
To characterize the LRD6-6 protein, multiple sequence alignment was performed on LRD6-6 and homologous proteins reported in different species. The LRD6-6 is clustered in the same clade with proteins AtSKD1 from Arabidopsis [38], ZmSKD1 from Zea mays [56], SKD1 from human [57], and VPS4p from yeast [58] (S15A Fig). The results showed that the LRD6-6 protein also contained the conserved domains (Walker A, Walker B and SRH), typical of previously characterized AAA ATPases [59] (S15B Fig), suggesting that LRD6-6 was an AAA ATPase. Previous reports suggest that the residue lysine (K) at 261st in Walker A motif of LRD6-6 is likely essential for nucleotide binding [37], glutamic acid (E) at 315th in Walker B is likely responsible for ATP hydrolysis [37] and arginine (R) at 372nd in SRH is likely vital for both ATP hydrolysis and oligomerization [59] (S15B Fig). To determine the ATPase activity of LRD6-6, the N-terminally truncated LRD6-6 LRD6-6(125–487) (AAs: 125–487, covering the ATPase domain) and its variants carrying point mutations, LRD6-6(125–487)K261A, LRD6-6(125–487)E315Q and LRD6-6(125–487)R372E, were respectively fused with a His-tag and expressed in Escherichia coli (Fig 4A). The purified proteins were then used for in vitro ATPase activity assay. The results showed that LRD6-6(125–487) was able to hydrolyze ATP but its variants LRD6-6(125–487)K261A, LRD6-6(125–487)E315Q or LRD6-6(125–487)R372E were not (Fig 4B). These results show that LRD6-6 is an active ATPase and that residues K261, E315 and R372 are essential for its ATPase activity.
Previous reports have shown that the AAA ATPase VPS4/SKD1, which is homologous to LRD6-6, can form dimers to function in disassembly of the ESCRT-III complex in regulation of MVBs biogenesis [37, 58, 60]. To test if LRD6-6 possesses the ability of dimerization, we cloned the full-length coding sequences (CDSs) of Lrd6-6, Lrd6-6K261A, Lrd6-6E315Q and Lrd6-6R372E respectively into both pGADT7 and pGBKT7 vectors and used for yeast two hybrid (Y2H) analysis. The CDS of lrd6-6 (Lrd6-6m) was also cloned into these vectors and included in this analysis. The results showed that the protein LRD6-6 interacted with itself (Fig 5A). Variants LRD6-6K261A and LRD6-6E315Q also interacted with themselves whereas variants LRD6-6R372E and LRD6-6m did not (Fig 5A). We then fused these proteins with the split YFP N-half and split YFP C-half, respectively, and performed bimolecular fluorescence complementation (BiFC) assay in N. benthamiana. The results showed that the cells of N. benthamiana co-transformed with LRD6-6–YFPN and LRD6-6–YFPC, LRD6-6K261A–YFPN and LRD6-6K261A–YFPC, or LRD6-6E315Q–YFPN and LRD6-6E315Q–YFPC produced yellow fluorescence signals whereas the cells co-transformed with LRD6-6m–YFPN and LRD6-6m–YFPC, LRD6-6–YFPN and LRD6-6m–YFPC, or LRD6-6R372E–YFPN and LRD6-6R372E–YFPC, exhibited no fluorescence signals (Fig 5B). These results suggest that LRD6-6 is capable of homo-dimerization and–of the three residues, K261, E315 and R372 –only R372 was required for this dimerization. As residues K261 and E315 are essential for ATPase activity but not required for homo-dimerization, these results also indicate that the ATPase activity is not required for homo-dimerization of LRD6-6. However, residue R372 is required for both homo-dimerization and ATPase activity of LRD6-6, suggesting that homo-dimerization may be required for the ATPase activity of LRD6-6.
To determine if the ATPase activity of LRD6-6 is required for its biological function, we created constructs, which carry Lrd6-6K261A, Lrd6-6E315Q or Lrd6-6R372E with the ATPase activity knocked out or compromised, and introduced them individually into the lrd6-6 mutant. Transformation of Lrd6-6 into lrd6-6 was also performed as a positive control. The transgenic plants expressing the wild type LRD6-6 were able to restore the lrd6-6 phenotype to wild type (S16 Fig); however, none of LRD6-6K261A, LRD6-6R372E and LRD6-6E315Q restored the mutant (Fig 6). The lrd6-6 plants expressing the LRD6-6 with catalytically inactive or compromised ATPase activity retained enhanced immunity and spontaneous cell death as the lrd6-6 mutant (Fig 6). Thus, the ATPase activity is essential for LRD6-6 to inhibit immunity and cell death in rice.
The glutamate (E) of hhhhDE sequence (h represents a hydrophobic amino acid) in the Walker B motif is crucial for ATP hydrolysis while the lysine (K) residue in the Walker A consensus sequence GXXXXGK[T/S] (X is any amino acid) of the conserved AAA ATPase family is crucial for ATP binding [36]. Although these two residues are capable of rendering the wild type protein dominant-negative [36], mutation of E in the Walker B motif has been more widely used to create ‘substrate traps’ in yeast [37], mammals [61] and Arabidopsis [38]. Because LRD6-6E315Q could dimerize with LRD6-6 as shown in yeast and in N. benthamiana (Fig 7A), we presumed that LRD6-6E315Q would also play a dominant-negative role for LRD6-6 in rice. To test this hypothesis, we expressed Lrd6-6E315Q in wild type Kitaake. The result showed that the transgenic plants expressing LRD6-6E315Q displayed spontaneous cell death similar to the lrd6-6 mutant plants, while transgenic plants expressing the wild type LRD6-6 did not (Figs 7B and S17). Detection of dead cells using trypan blue and measurement on the expression of PR genes indicated that the transgenic plants expressing LRD6-6E315Q also possessed enhanced immunity and presented spontaneous cell death similarly as the lrd6-6 mutant (Fig 7C). These results show that the LRD6-6E315Q is able to compromise LRD6-6 function likely by forming a functionally inactive homo-dimer, thus acting dominant-negatively. These results also suggest that mutant gene Lrd6-6E315Q may be utilized as a gene trap to suppress the inhibitory regulation of the ATPase LRD6-6 in immunity and cell death to enhance plant disease resistance.
To identify downstream components of the immunity and cell death resulting from disruption of the LRD6-6 ATPase in lrd6-6, we performed a genome-wide transcript comparative analysis on lrd6-6 and Kitaake using an RNA-seq approach. A total of 1223 differentially expressed genes (DEGs) were obtained (S18 Fig). Of these, 980 genes were up-regulated whereas 243 were down-regulated in the lrd6-6 mutant compared to the wild type Kitaake [P < = 0.05, Log2FC (lrd6-6/Kitaake) > 1] (S18 Fig).
Gene Ontology (GO) analysis showed that these DEGs could be classified into different cellular components (S19 Fig and S2 Table). Of them, MVBs-mediated vesicular trafficking associated components were the most enriched, including the GO terms membrane, membrane coat, clathrin coat of coated pit, endoplasmic reticulum membrane and clathrin coat of trans-Golgi network vesicle (S19 Fig and S2 Table). We then randomly selected some of these genes, performed qRT-PCR analysis on them and verified their differential expression between lrd6-6 and Kitaake (S20 Fig). We also investigated the expression of these genes in lrd6-6 plants transformed with the Os06g03940-11.5kb transgene and plants expressing Lrd6-6E315Q by qRT-PCR analysis. The results showed that expression of Os06g03940-11.5kb in lrd6-6 restored the expressions of these genes to the similar levels in the wild type Kitaake plants while expression of Lrd6-6E315Q in Kitaake retained dysregulation of these genes similarly as lrd6-6 (S21 Fig). These results suggested that the dysregulation of these genes in lrd6-6 is indeed the result of loss of function of the Os06g03940 (Lrd6-6) gene. Since a crosstalk exists between the secretory pathway and the early endocytic route in the early MVBs/trans-Golgi network [62], we therefore tested whether some genes associated with both secretory and endocytic trafficking were influenced by dysfunction of the AAA ATPase LRD6-6. Indeed, the expression of gene Os01g74180, encoding the β-subunit of adaptor protein complex 3 (AP-3), whose homologs have been shown to be important regulators of both endocytic and secretory pathways in yeast, mammals and Arabidopsis [63–65], was down-regulated in lrd6-6 plants and the Kitaake plants carrying Lrd6-6E315Q (S20 and S21 Figs, S2 Table). The clathrin heavy chain gene (chcA) was reported to be essential for secretion of lysosomal enzymes in Dictyostelium discoideum [66]. In Arabidopsis, the chc2 single mutant and dominant-negative CHC1 (HUB) transgenic lines were defective in bulk endocytosis and in internalization of prominent plasma membrane proteins [67]. We also found the Os12g01390 gene, coding for the clathrin heavy chain was up-regulated in lrd6-6 based on the RNA-seq data (S2 Table). These results indicate that MVBs-mediated vesicular trafficking is associated with the ATPase LRD6-6 and this trafficking process is dysregulated in lrd6-6 and the Kitaake plants expressing Lrd6-6E315Q.
Previous studies have shown that the Arabidopsis soluble vacuolar Carboxypeptidase Y (AtCPY) is transported from endoplasmic reticulum (ER) to the vacuole through the early secretory pathway mediated by MVBs [68]. Dysregulation of MVBs-mediated vesicular trafficking inhibits the AtCPY–GFP transport in Arabidopsis [45]. To detect whether the MVBs-mediated vesicular trafficking process was dysregulated in lrd6-6, we transiently expressed AtCPY fused to the N-terminal GFP (AtCPY–GFP) in protoplasts prepared from Kitaake and lrd6-6, respectively. The GFP fluorescence signal of AtCPY–GFP in Kitaake protoplasts could be visualized in the vacuole (Fig 8Aa and 8C). By contrast, the fluorescence was prominently presented in cytoplasm in the lrd6-6 protoplasts (Fig 8Ab and 8C), showing that the trafficking of AtCPY–GFP from ER to the vacuole was obviously inhibited in lrd6-6. The transport of AtCPY–GFP in the lrd6-6 plants expressing Os06g03940-11.5kb transgene was restored to the normal level of the wild type Kitaake plants whereas inhibited in the Kitaake plants expressing Lrd6-6E315Q (Fig 8Ac, 8Ad and 8C). These results clearly showed that the MVBs-mediated vesicular trafficking was modulated by the AAA ATPase LRD6-6. To determine whether dysfunction of MVBs could inhibit the transport of AtCPY–GFP in rice, we treated the Kitaake protoplasts transiently expressing AtCPY–GFP with wortmannin. We found that the transport of AtCPY–GFP was inhibited in most protoplasts (about 80%) treated with wortmannin (Fig 8B and 8C). This result clearly shows that transport of rice-expressed AtCPY–GFP from ER to vacuoles is also mediated by MVBs and this transport is completely inhibited in the lrd6-6 mutant.
Taken together, our results suggest that the MVBs-mediated vesicular trafficking is altered in the lrd6-6 mutant and the autoimmunity and spontaneous cell death of lrd6-6 likely results from the dysregulated MVBs-mediated vesicular trafficking.
To determine whether loss-of-function of LRD6-6 affects MVBs in general, we transiently expressed RabF1/ARA6–GFP in the protoplast cells prepared from Kitaake and lrd6-6 respectively. The result showed that localization of RabF1/ARA6–GFP in lrd6-6 was not changed compared to that observed in Kitaake (S22 Fig). This result suggested that LRD6-6 does not affect the machinery of MVBs generally.
Among the DEGs identified with RNA-seq, many were associated with immunity and cell death according to GO biological function analysis. These included PR genes, chitinase, WRKY transcription factors, MPKs and oxidation-related genes (S3 Table). We randomly selected some of these genes and verified their differential expression between the lrd6-6 mutant and wild type Kitaake (S23 Fig).
To explore the downstream events involved in the autoimmunity and spontaneous cell death in the lrd6-6 mutant, we performed pathway analysis on the DEGs–many pathways were likely involved in the immunity and cell death (S4 Table). Of them, three pathways, phenylpropanoid biosynthesis, diterpenoid biosynthesis and phenylalanine, tyrosine and tryptophan biosynthesis, which contribute to innate immunity and cell death according to previous reports [69–71] were highly activated in the lrd6-6 mutant compared with the wild type Kitaake (S24 Fig and S4 Table).
Gene pathway analysis also showed that the antimicrobial metabolites, including serotonin, lignin and phytoalexins might accumulate in the lrd6-6 mutant (Figs 9A and 10A). We thus performed qRT-PCR to analyze the expression of the key enzyme genes involved in biosynthesis of these antimicrobial metabolites. Indeed, we found that expression of the genes aroA and aroC, required for biosynthesis of chorismate from shikimate [72, 73], obviously increased in the lrd6-6 mutant (Fig 9A). Because chorismate is required for biosynthesis of both serotonin and lignin [74, 75], the increased expression of aroA and aroC would lead to accumulation of the antimicrobial metabolites, serotonin and lignin. We then found that expression of the downstream genes essential for respective biosynthesis of serotonin and lignin initiated from chorismate (for biosynthesis of serotonin: As, Igps, Ts and Tdc; for biosynthesis of lignin: Pal, Ptal, 4Cl, Ccr, F5h, Cad and Pox) increased in the lrd6-6 mutant (Fig 9A), supporting our notion that the antimicrobial metabolites, serotonin and lignin, would accumulate in the lrd6-6 mutant. Then we measured the expression of the genes required for the biosynthesis of phytoalexins, including phytocassanes A–E (genes: OsCyc2/OsCps2 and OsDtc1/OsKsl7), oryzalexins A–F (genes: OsCyc2/OsCps2 and OsDtc1/OsKsl10), oryzalexin S (genes: OsCyc1/OsCps4 and OsDtc2/OsKsl8) and momilactones A and B (genes: OsCyc1/OsCps4 and OsDtc1/OsKsl4) [76] between the lrd6-6 mutant and Kitaake (Fig 10A). These genes expressed higher in the lrd6-6 mutant compared with Kitaake (Fig 10A). This result suggested that biosynthesis of these phytoalexins might be highly activated in the lrd6-6 mutant. Previous reports have shown that the WRKY transcription factor OsWRKY14 is required for serotonin biosynthesis through regulating the expression of genes TS (tryptophan synthase) and TDC (tryptophan decarboxylase) [74, 77] and that the MPK genes OsMPK3, OsMPK6 and OsMPK4 regulate the biosynthesis of lignin and phytoalexins [75, 78]. When determining the expression of these genes, with the exception of OsMPK6, we found that the expression of all these genes was highly increased in the lrd6-6 mutant compared with Kitaake (Figs 9B and 10B), further supporting our notion that antimicrobial metabolites accumulated in the lrd6-6 mutant.
To confirm that these antimicrobial metabolites accumulate in lrd6-6, we selectively measured and compared the contents of lignin and phytoalexins between lrd6-6 and Kitaake. The result showed that the amount of total lignin and phytoalexins, momilactones A and B, and phytocassanes A–E in lrd6-6 were indeed largely increased compared with Kitaake (Figs 9C and 10C). Consistently, the contents of phytoalexins in the lrd6-6 plants expressing the transgene Os06g03940-11.5kb was restored to the levels of the wild type Kitaake while highly accumulated in the transgenic Kitaake plants expressing Lrd6-6E315Q which exhibited cell death (S25 Fig). Previous study has shown that salicylic acid (SA) biosynthesis is involved in shikimate-phenylpropanoid pathway [79], which is also involved in lignin biosynthesis activated in lrd6-6. Elevated levels of SA could stimulate SA-mediated defense response leading to secretion of antimicrobial compounds and increased expression of PR genes in plant [80]. To determine whether SA was accumulated in lrd6-6, we sampled the leaf of the lrd6-6 mutant, the lrd6-6 plant expressing the transgene Os06g03940-11.5kb, the Kitaake plant expressing Lrd6-6E315Q and the wild type Kitaake and subjected them to total SA content determination. The result showed that there was no significant difference of SA contents among the samples (S26 Fig). Taken together, these results suggested that the enhanced immunity and cell death in lrd6-6 results from accumulation of phytoalexins. These results also indicate that the accumulation of the antimicrobial metabolites instead of SA in the lrd6-6 mutant likely directly inhibit the infection of pathogens such as M. oryzae and Xoo.
To further dissect the molecular mechanism of LRD6-6 in regulation of immunity and cell death, we performed an Y2H screen using LRD6-6 as bait. The full amino acid coding sequence of Lrd6-6 was cloned in-frame with the GAL4 DNA binding domain of the bait vector pGBKT7. The Y2H screen was conducted using a cDNA library derived from rice Nipponbare. We identified a SNF7 domain containing protein (Os06g40620) and named this protein OsSNF7 because it showed high identity with Arabidopsis SNF7 in amino acid sequence. To confirm its interaction with LRD6-6, we amplified the full-length CDS of OsSnf7 and cloned it into the prey vector to obtain pGADT7-OsSNF7. The Y2H test clearly showed that full-length protein of OsSNF7 interacted with LRD6-6 (Fig 11A). We then fused OsSNF7 with GFP, LRD6-6 with RFP respectively and co-expressed them into N. benthamiana and observed their localizations. The result showed most of the green fluorescence produced by OsSNF7–GFP overlapped with the red fluorescence signals of LRD6-6–RFP (S27 Fig). This result suggests that the OsSNF7 protein also resides in MVBs and co-localizes with LRD6-6. The interaction between OsSNF7 and LRD6-6 was further confirmed by BiFC analysis in N. benthamiana through co-expressing LRD6-6–YFPN/OsSNF7–YFPC (Fig 11B).
To test if OsSNF7 functions in immunity and cell death in plants, we knocked out (KO) OsSnf7 in rice Kitaake using the CRISPR/CAS9 approach [81]. Three independent OsSnf7-KO rice lines with premature stop mutation at different residues were obtained (S28 Fig). Unexpectedly, these OsSnf7-KO plants, which were homozygous for their respective premature stop mutation, displayed no obvious cell death or growth defects compared with Kitaake. The expression of PR genes in these lines remained at the same level as wild type Kitaake (S29 Fig). These results suggested that the knock-out of OsSnf7 did not affect the immune response or cell death in plants. We then analyzed the homologs of OsSNF7 in rice and found four proteins encoded by genes Os12g02830, Os11g03060, Os09g09480 and Os07g30830 that also shared high identities with OsSNF7 (S30A Fig). We then cloned three of these homologs and tested their interactions with LRD6-6. The result showed that at least the protein encoded by Os12g02830 could interact with LRD6-6 (S30B Fig), suggesting that Os12g02830 likely functions redundantly with OsSNF7. Thus, to elucidate the role of OsSNF7 and its homologs in immunity and cell death in rice, we may need to develop double or multiple knock-out mutants of these genes in future studies.
Another core component of the ESCRT-III complex, VPS2, is also MVBs-localized and interacts with the AAA ATPase AtSKD1 in Arabidopsis [82, 83]. Three VPS2 homologs, encoded by Os03g43860, Os07g13270 and Os11g47710, respectively, were predicted in rice, based on alignment analysis of protein sequences [84]. To determine if these rice VPS2s interact with LRD6-6, we cloned full length CDSs of these genes into the vector pGADT7 to obtain pGADT7-Os03g43860, pGADT7-Os07g13270, and pGADT7-Os11g47710, respectively, and used them for Y2H tests. The result showed that the VPS2 homologous protein (we named it OsVPS2) encoded by Os03g43860 interacted with LRD6-6 (S31A Fig). BiFC assay showed that the N. benthamiana cells co-expressing OsVPS2 and LRD6-6 displayed punctate fluorescence signals with a similar pattern as LRD6-6 (S31B Fig) which further confirms the interaction between LRD6-6 and OsVPS2. These results also confirm that LRD6-6 localizes in MVBs.
Rice lrd mutants usually exhibit immunity-mediated cell death [48]. The lrd6-6 mutant possessed highly activated biosynthesis of antimicrobial metabolites with accumulated lignin and phytoalexins, which have been proven to fight against pathogens in rice and other plants [71, 80, 85, 86] (Figs 9, 10, S24 and S25). Consistently, like most lrd mutants, lrd6-6 displayed accumulation of hydrogen peroxide and spontaneous cell death (Figs 1B and S4) which are usually associated with the immune response. Indeed, the immunity was enhanced in the lrd6-6 mutant as the expression of the genes associated with immunity, including the PR, chitinase, WRKY transcription factors, MPKs and oxidation-related genes, were remarkably upregulated compared with wild type Kitaake (Figs 1C and S23, S3 Table). The enhanced immunity and cell death in the lrd6-6 mutant thus led to increased resistance to the pathogens M. oryzae and Xoo (Figs 1D, 1E and S5). Further analyses identified a novel AAA ATPase LRD6-6, whose disruption resulted in the lrd phenotype in the lrd6-6 mutant (Figs 2 and S9). Thus, the ATPase LRD6-6 negatively regulates immunity and cell death in rice. A very recent study reported cloning of the Lmr gene, the same gene as Lrd6-6, which harbors a G to A base substitution resulting in a premature translation termination in the lmr mutant [51]. The lmr mutant also showed cell death and displayed enhanced disease resistance to pathogens like the lrd mutant. However, this previous study did not characterize this protein biochemically or elucidate the underlying regulatory mechanism [51].
LRD6-6 contained Walker A, Walker B and SRH motifs, which have been characterized in typical AAA ATPases [34–36] (S15 Fig). High identities in amino acid sequences shared by LRD6-6 with the previously characterized AAA ATPases, SKD1 from human [57], Vps4p from yeast [58], mcSKD1 from ice plant [87] and At-KD1 from Arabidopsis [38] (S15 Fig), and the ATPase activity possessed by the recombinant LRD6-6 protein purified from E. coli further showed that LRD6-6 is an active AAA ATPase (Fig 4). Unlike the wild type LRD6-6, none of its ATPase catalytically inactive or impaired variants—LRD6-6K261A, LRD6-6E315Q or LRD6-6R372E—could inhibit the immunity-mediated cell death in the Lrd6-6-disrupted lrd6-6 plants (Fig 6), indicating that the ATPase activity was required for LRD6-6 to function. Interestingly, LRD6-6 was able to homo-dimerize in both yeast and plants (Fig 5). The ATPase catalytically inactive variant LRD6-6 E315Q was able to exert a dominant-negative effect through dimerization (Fig 7). This E (Glu) residue is widely conserved among AAA ATPases of the SKD1 subfamily, including human SKD1, yeast Vps4p, ice plant mcSKD1 and Arabidopsis AtSKD1, all of which are highly homologous to LRD6-6 (S15 Fig). The equivalent EQ mutants, with the conserved residue E (Glu) replaced by Q (Gln), SKD1E235Q [57], Vps4pE233Q [58], mcSKD1E231Q [88] and AtSKD1E232Q [38], are also able to abolish their respective ATPase activity and exert their regulatory roles dominant-negatively. Thus, these AAA ATPases share conserved biochemical characteristics and may function similarly biologically.
Previous studies have isolated about 14 genes responsible for lrd phenotypes and exhibit immunity-mediated cell death. Of them, only SPL28 is associated with protein trafficking. However, the Golgi apparatus-localization suggests that SPL28 may specifically be involved in the post-Golgi trafficking process [26]. Thus none of these proteins are associated with ATPase, MVBs-mediated vesicular trafficking as is LRD6-6. Together, these suggest that the molecular regulation of immunity and cell death in these lrd mutants are very complicated and the regulatory machinery in the lrd6-6 mutant differs from those in other lrd mutants characterized previously.
Many AAA ATPases belong to the same subfamily as LRD6-6, including the mammal SKD1 [57], yeast Vps4p [58], ice plant mcSKD1 [87] and Arabidopsis AtSKD1 [38], which spread mainly on MVBs and are required for MVBs biogenesis, and its mediated vesicular trafficking [60]. In the regulation of MVBs biogenesis, the AAA ATPase of this subfamily is recruited to the MVBs membrane by ESCRT-III subunits. Then, the AAA ATPase provides the energy to disassociate the ESCRT-III complex from the membrane, and this has been characterized as the last essential step of MVBs biogenesis [60]. When the AAA ATPase is mutated or its function is rendered dominant-negative by interaction with its catalytically inactive EQ mutant, both the MVBs biogenesis and its mediated vesicular trafficking are largely blocked [60]. This can even be lethal in Arabidopsis [38, 45], and suggests the conserved roles of these AAA ATPases in diverse species [36]. A recent study reported that LMR, encoded by the same gene Lrd6-6, localizes in chloroplasts for its function [51]. However, our study revealed that the LRD6-6 protein spreads on MVBs (Fig 3), which is in agreement with the MVBs-localization of the AAA ATPases from this family [38, 57, 58, 88]. Firstly, the protein LRD6-6 fused with GFP and RFP on its C-terminus and with YFP on its N-terminus all displayed a punctate distribution and co-localized with the MVBs-localized marker protein RabF1/ARA6 but did not overlap with the chlorophyll auto-fluorescence (Figs 3 and S14). Based on our results, the subcellular localization pattern of LRD6-6 is similar to its Arabidopsis homolog AtSKD1, previously shown to be MVBs-localized [38]. Secondary, LRD6-6 co-localized and interacted with ESCRT-III components OsSNF7 and OsVPS2 (Figs 11, S30 and S31), which are respectively homologous to the MVBs-localized AtSNF7 and AtVPS2 from Arabidopsis [82, 84]. Taken all together, we thus conclude that LRD6-6 mainly localizes on MVBs.
The MVBs-localization suggests that the LRD6-6 AAA ATPase is associated with MVBs-mediated vesicular trafficking in rice. Indeed, although this trafficking machinery is not generally affected (S22 Fig), it is dysregulated in the lrd6-6 mutant as revealed by whole transcriptome expression analyses and the inhibited trafficking of the soluble vacuolar cargo AtCPY from ER to vacuoles (Figs 8, S20 and S21, S2 Table). Both the genes that likely encoding components of secretory and endocytic trafficking are co-regulated. For example, the AP-3β-coding gene Os01g74180 was down-regulated; the clathrin heavy chain coding gene Os12g01390 was up-regulated in both lrd6-6 and Kitaake plants expressing Lrd6-6E315Q (S20 and S21 Figs, S2 Table). It is thus likely that both these two trafficking pathways are dysregulated in plants with loss of function of the AAA ATPase. Thus, our study demonstrates the essential roles of the AAA ATPase LRD6-6 in MVBs-mediated vesicular trafficking similar to SKD1 [57] and its homologous proteins AtSKD1 from Arabidopsis [38, 45], VPS4p from yeast [58]. Our study also supports the notion that the AAA ATPases of the SKD1 subfamily are conserved in function in diverse species [60]. Importantly, our study reveals that the AAA ATPase LRD6-6 inhibits the immune response, suggesting that the LRD6-6-mediated modulation of immunity and cell death is associated with MVBs-mediated vesicular trafficking in rice, which is not reported previously for SKD1 or any other homologous proteins. Although the dysregulated MVBs-mediated vesicular trafficking may cause many biological defects, such as impeded cell proliferation, adhesion and drug resistance in humans [41], there are few reports concerning their regulation of immunity and cell death. Previous studies have implied that cell death may occur in plants with AtSKD1 knocked out or a catalytically inactive variant dominant-negative mutant AtSKD1E232Q in Arabidopsis [38, 45]. However, no direct evidence was provided from their studies because AtSKD1 knock-out or expression of AtSKD1E232Q is lethal to plants [38, 45]. Interestingly, the rice plants with disrupted LRD6-6 or dominant-negatively regulated by the catalytically inactive variant LRD6-6E315Q still grow well, with the only visible defect of spontaneous cell death phenotype (Figs 1A and 7B). Thus, our discovery that the AAA ATPase LRD6-6 is essential for MVBs-mediated vesicular trafficking provides a valuable way to study its regulation on immunity and cell death. Of great importance, if introducing the catalytically inactive variant LRD6-6E315Q with dominant-negative effect engineered under the pathogen-inducible promoter into rice plant, we may obtain the rice with enhanced disease resistance without affecting rice yield.
MVBs-mediated vesicular trafficking has attracted much attention because of its role observed in immunity recently [31, 32]. The animal pattern recognition receptor TLR4 (toll-like receptor 4) mediates perception of bacterial-derived lipopolysaccharides, and undergoes internalization upon activation with its cognate ligand through MVBs-mediated vesicular trafficking [89]. In Arabidopsis, a transmembrane leucine-rich repeat receptor kinase FLS2 that recognizes bacterial flagellin, similarly exhibits ligand-stimulated endocytosis [46, 89]. The trans-Golgi network/early endosomes component KEG is reported to play a role in plant immunity by regulation of intracellular trafficking processes, and the secretion of apoplastic defense proteins [90]. VPS35B in Arabidopsis is part of the retromer complex, which functions in endosomal protein sorting and vesicular trafficking, contributing to TIR-NB-LRR and CC-NB-LRR protein-mediated autoimmunity and HR cell death [91]. However, it remains unclear how these immune proteins are transported through MVBs-mediated vesicular trafficking. In rice, trafficking is essential for the OsCEBiP/OsCERK1-OsRacGRF1-OsRac1 module to regulate immunity [92, 93], while little is known about the regulation of MVBs-mediated vesicular trafficking on immunity. Our study uncovered that the inhibitory regulation of AAA ATPase LRD6-6 on MVBs-mediated vesicular trafficking may be associated with plant immunity and cell death. This discovery indicates that certain immune response-associated protein(s), including OsCEBiP in PTI, NBS-LRR proteins in ETI, and important regulators downstream, may not be sorted or transported properly due to dysregulated MVBs-mediated vesicular trafficking. The disordered sorting or failure in the transport of these immune response-associated proteins may then activate the immune response without pathogen infection and results in spontaneous cell death in rice. Alternatively, it is also likely that the LRD6-6 mediated-MVBs trafficking potentially is guarded by certain NBS-LRR proteins and constitutes a downstream component of the ETI pathway. The dysfunction of this process may bypass the activation of NBS-LRR and trigger the NBS-LRR-mediated ETI response, resulting in spontaneous cell death and enhanced disease resistance similarly as reviewed previously [94]. In addition, the MVBs pathway is positively regulated by pathogen responsive MPK3/6 through phosphorylation of LIP5, an ESCRT component [46]. As MPK3/6 is part of PTI and ETI responses, these results seem to support the notion that regulation of MVBs trafficking is part of the ETI/PTI pathways. It is unclear how and where serotonin, lignin, and phytoalexins accumulate in the lrd6-6 mutant. It is possible that the blockage of transport of these antimicrobial compounds to vacuoles prevents their turnover in vacuoles and results in buildups in other compartments of the cell. Our results suggest MVBs-mediated trafficking may be essential for accurate delivery of these antimicrobial compounds.
Previous studies have reported that other AAA ATPases of different subfamilies to LRD6-6 are also associated with immunity in mammals and plants. The human AAA ATPase p97/VCP regulates antiviral immunity through binding directly to multi-ubiquitin chains and unfolding ubiquitin-fusion degradation substrates, such as the larger substrate adenovirus particle [39]. Overexpression of the mitochondrial outer membrane-localized AAA ATPase AtOM66 can constitutively induce salicylic acid-related defense response and cell death in Arabidopsis [44]. The tobacco AAA ATPase NtAAA1 inhibits innate immunity by regulating ethylene- and salicylic acid-mediated defense response through interaction with a small GTPase, NtARF [42, 43]. However, our present study reveals that dysregulation of MVBs-mediated vesicular trafficking by disruption of the AAA ATPase LRD6-6 results in accumulation of antimicrobial metabolites which then leads to activation of immunity and cell death in rice (Fig 12). This differs from the molecular regulation of immunity mediated by those AAA ATPases reported previously and defines a novel regulatory machinery of immunity in plant. Thus, our discovery provides novel insights into immunity regulated by the AAA ATPase LRD6-6 likely through MVBs-mediated vesicular trafficking in rice and possibly other species.
The lrd6-6 mutant was obtained from the tissue cultured rice, Kitaake. The lesion spot appeared about 15 days after sowing. Husked seeds of Kitaake and the lrd6-6 mutant were sterilized in 30% bleach for 30 min followed by rinsing three times with sterile ddH2O. To investigate whether the lesion spot phenotype occurs under sterile conditions, the sterilized seeds were germinated on ½ Murashige and Skoog (MS) medium in SoLo cup and incubated in growth chamber until presence of lesion spots on leaves. For other phenotypic characterizations and map-based cloning, the plants were grown in the fields at Sichuan Agricultural University in Wenjiang, Chengdu, or Lingshui, Hainan, China.
The leaves of both the lrd6-6 mutant and Kitaake plants were shaded by bandaging with silver paper before initiation of lesion spot in lrd6-6 until clear presence of lesion spots on the part of leaf without bandage.
The leaves from the lrd6-6 mutant and Kitaake plants were collected at three days before and after lesions appearance, respectively. Samples with 200 mg leaf tissue of each were soaked in 30 ml 80% acetone for 48 h in dark until the disks became colorless. Chlorophyll concentration was measured with four experimental repeats following the method as described previously [95].
Trypan blue staining assay was performed on fresh leaves following the method as described previously [96]. In brief, samples were submerged in lactic acid-phenol-trypan blue solution (2.5 mg/ml trypan blue, 25% (w/v) lactic acid, 23% water-saturated phenol and 25% glycerol in H2O) and were boiled in water for 2 min, then de-stained with solution containing 30% (w/v) chloral hydrate for 3 days with multiple exchanges of the solution. After distained completely, the samples were then equilibrated with 50% glycerol for five hours followed by photo-picture taken. Detection of H2O2 accumulation was carried out using DAB staining method as described previously [97]. Briefly, leaf samples were immersed in 1 mg/ml DAB containing10 mM MES (pH 6.5) for 12 h in the dark at 30°C. Then the leaf samples were transferred to solution containing 90% ethanol and 10% glycerol at 90°C until chlorophyll was completely removed. Cleared leaves were examined and photographed using an Olympus anatomical lens.
The samples were collected from the leaves of the lrd6-6 mutant and Kitaake plants before and after appearance of lesion spot in lrd6-6. The samples were then prefixed with a mixture solution of 3% glutaraldehyde. Subsequently, post-fixed in 1% osmium tetroxide, dehydrated in series acetone, infiltrated in Epox 812 for 4 hours, and embedded [98]. The sections were stained with methylene blue and ultrathin sections were cut with diamond knife, stained with uranyl acetate and lead citrate. Sections were examined under a transmission electron microscope (TEM; HITACHI, H-600IV, Japan).
The mRNA samples were extracted using TRIzol (Invitrogen, Carlsbad, CA, USA) following the procedures as described by the manufacturer. The mRNA was treated with DNase I according to the manufacturer’s instructions (Invitrogen, Carlsbad, CA, USA) and was subject to reverse transcription to synthesize first-strand cDNA. Oligo (dT) was used as primer and SuperScript II (Invitrogen, Carlsbad, CA, USA) was used as reverse transcription enzyme.
The qRT-PCR was conducted using a Bio-Rad CFX96 Real-Time System coupled to a C1000 Thermal Cycler (Bio-Rad, Hercules, CA, USA). The reference gene Ubiquitin 5 (Ubq5) [99] was used as control for qRT-PCR experiments. The sequences of the primers were listed in S6 Table.
Ten-days-old seedlings of Kitaake and the lrd6-6 mutant were used for inoculations with Magnaporthe oryzae (M. oryzae). The M. oryzae strains, ZB25, Zhong1 and ZE-1, which are compatible with Kitaake, were used for inoculation. Spore concentration was adjusted to 2×105 spores/ml with a hemacytometer before spraying [100]. The disease lesion length and the sum of lesions were acquired at day seven after inoculation.
The Xoo strains, P2, P4, P5, P6 and Xoo-4, which are compatible with Kitaake, were used for inoculation. Xoo bacterial suspensions with 0.5 of OD600 were used to inoculate by using the scissors-dip method as described previously [101]. Disease lesion lengths were determined at day 15 post inoculation. The lesion lengths and bacterial populations were determined at day 0, 7, 14 post inoculation with the strain P2.
Statistical analyses were performed using SPSS version 19.0 (Released 2010. IBM SPSS Statistics for Windows, Version 19.0. Armonk, NY: IBM Corp).
Three F1s and three F2 populations derived from the crosses of Jodan × lrd6-6, lrd6-6 × Jodan and 02428 × lrd6-6 were respectively used for the genetic analysis. The F2 population derived from the cross of 02428 × lrd6-6 was used for mapping of the gene, lrd6-6. Bulk segregation analysis (BSA) was first used to rapidly locate the locus of lrd6-6 on a chromosome. For the BSA analysis, equal amount of leaf blades from six F2 plants with the lesion spot phenotypes or six F2 plants with the wild type phenotypes were collected for DNA extraction to construct the mutant and the wild type DNA pools, respectively. The physical linkage map was then constructed using additional markers nearby the locus of lrd6-6.
The SSR primers were synthesized according to the information from Gramene database (http://www.gramene.org/microsat). InDel markers were designed based on the alignment analyses on the reference japonica rice Nipponbare (http://www.rgp.dna.affrc.go.jp) and indica rice 93–11 (http://www.rise.genomics.org.cn) genome sequences at the target location. Primers were designed using Primer 5 software. The specificity of each primer in the rice genome was confirmed by BLAST and PCR analyses.
For analysis of the PCR product, amplified products by the primer pairs of the markers, I4-2, RM19274, RM19320, I8, were separated by 3.5% agarose gel electrophoresis in 1×TAE buffer, and visualized and photographed under UV light. The amplified products by RM8075 and RM587 were separated by 6% denaturing polyacrylamide gel electrophoresis and visualized by silver staining according to the method described previously [102].
For whole genomic resequencing, lrd6-6 was backcrossed with Kitaake for twice and self-crossed to produce BC2F2 progenies homozygous for the locus of lrd6-6. Equivalent total DNA of 30 BC2F2 plants homologous for lrd6-6 locus were pooled and sequenced at Beijing Genomics Institute (BGI, Beijing, China). For genome sequence comparison, the Kitaake genome DNA was also sequenced.
For RNA interference (RNAi) construct, a unique cDNA fragment of Os06g03940 from 557 bp to 1002 bp was amplified and put into the pANDA vector [103] to create RNAi construct, pANDA-Os06g03940Ri, using LR recombination enzyme (Invitrogen, Carlsbad, CA, USA).
For genetic complement construct, the genomic DNA fragment of 11.5 kb containing the native promoter (1960 bp upstream of ATG), entire coding region of LOC_Os06g03940 (Os06g03940) and a fragment with 2295 bp downstream of TAG, was cloned into the binary vector pCactN-XG following the steps: I, Four DNA fragments covering the 11.5 kb genome sequence from Nipponbare genome were amplified with KOD DNA polymerase (Toyobo, Osaka, Japan) using the primers, lrd6-6cp-P1, lrd6-6cp-P2, lrd6-6cp-P3, lrd6-6cp-P4, listed in S5 Table. These fragments were then cloned into the vector pEASY Blunt Simple (TransGen Biotech, Beijing, China) to create pSimple-CP1, CP2, CP3, CP4 respectively followed by sequencing verification (Sangon Biotech, Shanghai, China). II, the DNA fragment of CP4 was released from pSimple-CP4 under digestion with the enzymes, NdeI and SalI, and put into pSimple-CP3 pre-digested by the same enzymes to generate pSimple-CP3-CP4. III, the DNA fragment of CP1 was cut from pSimle-CP1 with BamHI and XhoI, and put into pSimple-CP2 predigested to create pSimple-CP1-CP2. IV, the DNA fragment of CP1-CP2 was digested from pSimle-CP1-CP2 with the enzymes, BamHI and SmaI, and put into TSK108 vector predigested to get TSK108-CP1-CP2. V, the DNA fragment of CP3-CP4 was released from pSimle-CP3-CP4 by cutting with SmaI and SalI and cloned into TSK108-CP1-CP2 to generate an intermediate vector TSK108-CP1-CP2-CP3-CP4 (TSK108-Os06g03940-11.5kb). VI, the DNA fragment of Os06g03940-11.5kb was then cut from TSK108-Os06g03940-11.5kb with BamHI and SalI and cloned into the predigested pCactN-XG to get the destination construct pCacTN-XG-Os06g03940-11.5kb. The construct pCacTN-XG-Os06g03940-11.5kb was verified by sequencing before used for transformation (Sangon Biotech, Shanghai, China). For other constructs, the intended segments were cloned into the vector pEASY Blunt Simple and then sub-cloned into the destination vectors. All the constructs information and the primers used were listed in S5 Table.
The Agrobacterium-mediated transformation was used for rice genetic transformation according to the method described previously [104]. The transgenic plants obtained were selected with Hygromycin or G418 during regeneration. The positive transgenic plants were then verified by PCR-based genotyping using the primer pairs, specific for hygromycin gene or Neomycin phosphotransferase II gene (NPT II) selected with G418, respectively.
For subcellular localization in Nicotiana benthamiana (N. benthamiana) and onion epidermal cells, the plasmid DNA of p35S:Lrd6-6–GFP was introduced into N. benthamiana leaf by agroinfiltration [105] and onion epidermal cells using a bombardment-mediated gene transformation [106], respectively. The N. benthamiana leaf and onion epidermal cells were transformed with p35S:GFP as controls. Fluorescence was examined under a confocal microscopy (NiKon A1 i90, LSCM, Japan) 36 h post transformation of N. benthamiana and 16 h post transformation of onion epidermal cells, respectively.
For subcellular localization in rice protoplasts, the plasmid, pBI221–AtCPY, was transformed into protoplasts prepared from Kitaake and lrd6-6 mutant seedlings following the method as described previously [107]. For a control, the protoplasts were transformed with pBI221. Fluorescence was examined under a confocal microscopy (NiKon A1 i90, LSCM, Japan) 16 h after transformation.
The N. benthamiana leaf expressing the RabF1/ARA6–GFP or RabF1/ARA6–RFP fusion protein was immersed into solution containing 33 μM wortmannin (Selleckchem) for 40 min and then examined under a confocal microscopy similarly as described above.
For treatment of the rice protoplasts expression AtCPY–GFP, wortmannin (33 μM) was added into the incubation buffer 4 hours after transformation and then incubated at 28°C until examination.
The N.benthamiana leaves expressing the interested proteins were respectively harvested 48 hours post transformation. Protein extraction and immunoblotting analysis with corresponding antibodies as indicated were performed according to the method described previously [105].
For determination of the protein express in yeast, the total protein extract was prepared and analyzed with anti-Myc and anti-HA antibodies, separately, following the Yeast Protocols Handbook from Clonetech (Otsu, Shiga, Japan).
To obtain the mutants, K261A, E315Q and R372E of LRD6-6, the full-length coding sequence of Lrd6-6 was cloned into the vector pEASY Blunt Simple. Mutagenesis was then performed with the specific primers (Listed in S5 Table) using QuikChange site-directed mutagenesis kit following the manual (Stratagene, La Jolla, California, USA) and the intermediate cloning and final constructs harboring the desired mutation sites were verified by sequencing.
The truncated LRD6-6 (AAs: 125–487, covering the ATPase domain) and its mutation versions were cloned into the pET28a vector (Novagen), and expressed in the E.coli strain BL21. Bacteria contain the plasmids were grown in Luria-Bertani (LB) medium containing 100 μg/ml and kanamycin at 37°C to OD600 = 0.6, induced by addition of isopropyl β-D-1-thiogalactopyranoside (IPTG) to final concentration of 1 mM and incubated at 28°C for 6 hours respectively. Cells were pelleted by centrifugation, re-suspended in lysis buffer (20 mM Tris-HCl PH 7.4, 0.1 M NaCl, 10 mM imidazol) and sonicated. After the cell debris removed by centrifugation (12000 g, 10 min, 4°C), the supernatant was loaded onto a Ni-NTA-agarose column (GE Healthcare, Buckinghamshire, United Kingdom), washed with washing buffer (20 mM Tris-HCl PH 7.4, 0.1 M NaCl, 20 mM imidazol) and eluted with elution buffer (20 mM Tris-HCl PH 7.4, 0.1 M NaCl, 200 mM imidazol).
ATPase activity of LRD6-6(125–487), LRD6-6(125–487)K261A, LRD6-6(125–487)E315Q and LRD6-6(125–487)R372E were measured by the malachite green-based colorimetric method using the ATPase/GTPase activity assay kit (Sigma-Aldrich, St. Louis, MO, USA). The elution buffer was used as negative control. One unit is termed as the amount of enzyme that catalyzes the production of 1 μM of free phosphate per minute under the assay conditions.
Yeast two-hybrid assay was performed using the GAL4-based Matchmaker Gold Yeast Two-Hybrid System (Cat. No. 630489, Clontech, Otsu, Shiga, Japan), in which the bait protein is expressed as a fusion to the Gal4 DNA-binding domain in pGBKT7 vector and the prey protein is expressed as fusion to the Gal4 activation domain in pGADT7 vector [108, 109]. Full-length amino acids of the proteins tested were cloned in frame with the Gal4 DNA-binding or activation domain using genes specific primers to obtain the constructs (Listed in S5 Table). After sequence verified, the constructs were co-transformed by pair into the yeast stain Y2HGold via polyethylene glycol (PEG)/LiAc-based method provided by the Yeastmaker Yeast Transformation System 2 (Cat. No. 630439, Clontech). The transformed Y2HGold cells were plated on DDO media (double dropout medium: SD/–Leu/–Trp) and the cell clones were diluted for 3 gradients on the QDO/A media (quadruple dropout medium: SD/–Ade/–His/–Leu/–Trp supplemented with Aureobasidin A) for interaction test.
After confirmation that no autoactivation and toxicity exist in the Y2HGold cells co-transformed with pGBKT7-LRD6-6 and the null pGADT7 vector, yeast two hybrid screening was performed by co-transformation with the pGBKT7-LRD6-6 plasmid DNAs (10 μg) and cDNA library DNA (fused into the pGADT7 vector) (10 μg) were co-transformed into the Y2HGold cells following the procedure in the manual of Yeastmaker Yeast Transformation System 2 (Cat. No. 630439, Clontech). Plasmids of the positive clones grew on the selective QDO/A media were rescued and were respectively co-transformed with pGBKT7-LRD6-6 or the empty bait vector pGBKT7 into Y2HGold cells to validate the interactions. After verification of the interactions, the plasmid DNAs of clones were subject to sequencing.
The full-length coding sequences of Lrd6-6, lrd6-6, Lrd6-6K261A, Lrd6-6E315Q, Lrd6-6R372E and OsSnf7 without stop codon were amplified using specific primers (Listed in S5 Table) and cloned into the psPYNE and/ or psPYCE vectors through XbaI and BamHI to generate their respective YFPN and/ or YFPC fusion protein expressing constructs [110]. Equivolume suspensions of different Agrobacterium strains carrying different constructs were mixed prior to infiltration and co-infiltrated into N.benthamiana leaves following the method described previously [105]. Fluorescence was examined under a confocal microscopy (NiKon A1 i90, LSCM, Japan) 36–48 hours post transformation.
When lesion spot initiated on leaf of the lrd6-6 mutant, the leaf tissues from lrd6-6 and the equivalent part of Kitaake as indicated (S18 Fig) were sampled for RNA extraction. RNA-seq was performed by CapitalBio Corporation at Beijing, China. Cuffcompare [111] was used to compare the assembled transfrags of each library to the reference annotation, and build up a non-redundant transcripts data set among the libraries. Then Cuffdiff [111] was used to identify DEGs. Transcripts data set were first compared with Kyoto Encyclopedia of Genes and Genomes database (KEGG, release 5.8) [112] using BLASTX [113] at E values < = 1e-10. A Perl script was used to retrieve KO information from blast result and then established pathway associations between transcripts and database. InterPro domains [114] were annotated by InterProScan (release 4.8) [115] and functional assignments were mapped by using GO (Gene Ontology) analysis [116]. Expression verification of the selected DEGs was conducted by using q-RT-PCR analysis. The primer pairs used were listed in S5 Table.
The lignin content and total SA content were determined according to the acid detergent lignin method and the ultrahigh performance liquid chromatography–triple quadrupole mass spectrometry (UPLC-MS/MS) method as described previously [117, 118].
For determination of phytoalexins, tissue samples (ca. 10 mg fresh weight for each) were harvested from the lrd6-6 mutant and Kitaake. The samples were soaked in 1 ml extraction solvent (MeOH/H2O, 80:20 [v/v]) and incubated at room temperature overnight. Then, 5 μL of the extract was subjected to phytoalexin measurement by LC-ESI-MS/MS. An Agilent 1200 separation module (Agilent Technologies, Palo Alto, CA, USA) equipped with a CAPCELL CORE C18 column (50 mm long, 2.1 mm in diameter; Shiseido, Tokyo, Japan) was used for HPLC analysis. The mobile phase consisted of 0.05% AcOH in H2O (solvent A), and 0.05% AcOH in MeCN (solvent B). Elution was conducted using a linear gradient from 40% to 60% solvent B over 10 min at a flow rate of 0.2 mL/min, and the eluate monitored by Agilent 6460 Triple Quadrupole mass spectrometer (Agilent). The ionization mode was electrospray. All phytoalexins were analyzed in positive ion mode. Electrospray conditions were as follows: capillary voltage, 3500 V; drying gas flow, 5 L/min nitrogen; drying gas temperature, 300°C; nebulizer pressure, 45 psi; sheath gas temperature, 350°C; and sheath gas flow, 11 L/min. The multiple reaction monitoring (MRM) mode was used in ESI-MS/MS. Phytoalexins were detected with MRM transitions of m/z 315.2/271.1 for momilactone A; m/z 331.2/269.1 for momilactone B; m/z 317.2/299.1 for phytocassanes A, D and E; m/z 335.2/317.2 for phytocassane B; m/z 319.2/301.2 for phytocassane C. Collision energy was 25 V for momilactones A and B; 15 V for phytocassanes A, D and E; 21 V for phytocassane B; 17 V for phytocassane C. Dwell time and fragmentor voltage were 200 ms and 135 V, respectively, for all phytoalexins.
Genes reported in this article can be found in the GenBank/RGAP databases under the following accession numbers: LRD6-6 (LOC_Os06g03940); RabF1/ARA6 (AT3G54840); OsSKD1 (LOC_Os01g04814); AtSKD1 (AT2G27600); ZmSKD1 (XP_008655625.1); VPS4A (NP_037377.1); VPS4B/SKD1 (NP_004860.2); VPS4p (NP_015499.1); FtsH (AAA97508.1); NSF-1 (NP_524877.1); AtCPY(At3g10410); OsSNF7 (LOC_Os06g40620); OsVPS2 (LOC_Os03g43860); CHMP4b (NP_789782.1); SHRUB (NP_610462.3); CHMP4a (NP_054888.2); scVPS32 (NP_013125.1); SNF7.1 (AT4G29160) and SNF7.2 (AT2G19830).
|
10.1371/journal.ppat.1002917 | Phase Variable O Antigen Biosynthetic Genes Control Expression of the Major Protective Antigen and Bacteriophage Receptor in Vibrio cholerae O1 | The Vibrio cholerae lipopolysaccharide O1 antigen is a major target of bacteriophages and the human immune system and is of critical importance for vaccine design. We used an O1-specific lytic bacteriophage as a tool to probe the capacity of V. cholerae to alter its O1 antigen and identified a novel mechanism by which this organism can modulate O antigen expression and exhibit intra-strain heterogeneity. We identified two phase variable genes required for O1 antigen biosynthesis, manA and wbeL. manA resides outside of the previously recognized O1 antigen biosynthetic locus, and encodes for a phosphomannose isomerase critical for the initial step in O1 antigen biosynthesis. We determined that manA and wbeL phase variants are attenuated for virulence, providing functional evidence to further support the critical role of the O1 antigen for infectivity. We provide the first report of phase variation modulating O1 antigen expression in V. cholerae, and show that the maintenance of these phase variable loci is an important means by which this facultative pathogen can generate the diverse subpopulations of cells needed for infecting the host intestinal tract and for escaping predation by an O1-specific phage.
| The O1 serogroup of Vibrio cholerae is the most common cause of the potentially fatal diarrheal disease cholera, which remains a significant global health burden worldwide. The O1 antigen is a constituent of the lipopolysaccharide portion of the outer membrane, and its location on the bacterial surface makes it a major target of both the immune system and bacteriophages. We used an O1-specific bacteriophage as a tool to understand if, and how, V. cholerae can alter O1 antigen expression. We discovered that two genes, which are critical for O1 antigen biosynthesis, are subject to phase variation. Additionally, one of the phase variable genes we identified was not previously known to play a role in O1 antigen biosynthesis in V. cholerae. Phase variation is a well-recognized mechanism many other bacterial pathogens use to generate variable expression of surface components, and this is generally thought to help these organisms evade the immune system. Phase variation has not previously been described as a widespread mechanism used by V. cholerae, furthermore, this is the first report that V. cholerae O1 is capable of generating diverse populations with variable and unique O1 antigen expression.
| Lipopolysaccharide (LPS) is a prominent constituent of the outer membrane of Gram-negative bacteria. The LPS molecule is divided into three components; lipid A, core oligosaccharide and O-specific polysaccharide (or O antigen). The structure of the O antigen typically defines the serogroup of an organism, and over 200 serogroups of Vibrio cholerae are currently recognized [1]. Interestingly, the O1 serogroup has been and continues to be the dominant cause of both endemic and epidemic cholera throughout the world, though the reasons for this are unknown. The incidence of cholera worldwide is steadily increasing, and the cumulative number of reported cases in 2010 was nearly double what it was in 2009 [2]. When considering the level of gross under-reporting, the actual global disease burden is estimated to be 3–5 million cases and more than 100,000 deaths [3], [4]. The observed increase in reported cases in 2010 is largely due to a recent outbreak of an O1 strain that started in Haiti: Even more concerning is the observation that 53% of the global total of the number of reported deaths from cholera in 2010 occurred in Haiti in a period of only 70 days [2]. These observations highlight the fragile nature of impoverished and tragedy-struck nations to the rapid onset of cholera epidemics.
The V. cholerae O1 antigen is composed of 12–18 repeating units of α(1,2)-linked d-perosamine (4-amino-4,6-dideoxy d-mannose) residues, the amino groups of which are acylated with tetronate (3-deoxy-l-glycero-tetronic acid) (Fig. S1) [1], [5]–[8]. The genes currently described as being required for the synthesis of the O1 antigen are located on chromosome 1 of the V. cholerae O1 N16961 genome between open reading frames (ORFs) VC0240 (gmhD) and VC0264 (rjg) (Fig. 1A) [9]. This region (the wbe or rfb region) was originally identified through the heterologous expression of the V. cholerae O1 antigen in Escherichia coli K-12 [10]. Additional genes required for the synthesis of the O1 antigen in V. cholerae were subsequently identified [11], however all O1 antigen biosynthetic genes studied to date have been between the gmhD and rjg flanking genes. The genes responsible for O1 antigen biosynthesis have been placed into the following five groups according to putative function: perosamine biosynthesis (VC0241–VC0244) [12]; O antigen transport (VC0246–VC0247) [13]; tetronate biosynthesis (VC0248–VC0252) [14]; O antigen modification (VC0258) [15], [16]; and additional genes essential for O antigen biosynthesis (VC0259–VC0260, VC0263) [11] (Fig. 1A). A putative pathway for the biosynthesis of perosamine has been proposed by Stroeher et al. [12] (Fig. 1B). In this pathway, which is based solely on homology comparisons, the first step is the conversion of fructose-6-phosphate (F6P) to mannose-6-phosphate (M6P) by ManC (a predicted type II phosphomannose isomerase [PMI]). M6P is then converted to mannose-1-phosphate (M1P) by ManB, and then to GDP-mannose by ManC. GDP-mannose is converted to GDP-4-keto-6 deoxymannose by WbeD and then GDP-perosamine by WbeE. PMIs (E.C. 5.3.1.8) catalyze the reversible isomerization of M6P to F6P and are divided into three families on the basis of amino acid sequence [17]. Type I PMIs are monofunctional enzymes and include proteins from humans to bacteria including E. coli [18] and Salmonella enterica serovar Typhimurium [19]. Type II enzymes are bacterial bifunctional enzymes possessing both PMI and guanosine diphospho-d-mannose pyrophosphorylase (GMP) activity (for the conversion of M1P into GDP-mannose) in distinct catalytic domains [20]. PMIs play critical roles in mannose catabolism and in the supply of GDP-mannose, which is necessary for the mannosylation of various structures including LPS.
The distal location of the O antigen extending outward from the bacterial surface positions it at the interface between the bacterium and its environment. As such, the O antigen is important for protection from various environmental stresses including antibiotics and the host immune response [21], [22]. The O antigen is also consequently the target of both the immune system and bacteriophages, which can independently apply powerful selective forces. As such, cell surface structures, including the O antigen, are frequently observed to exhibit high levels of variation [23]. Examples of phase variable surface structures are abundant in bacterial pathogens and include Haemophilus influenzae lipooligosaccharide (LOS) [24]–[26], Neiserria meningitidis LOS [27], Helicobacter pylori LPS [28], [29] and Campylobacter jejuni LOS [30], [31]. The loci responsible for phase variable expression of these structures, often referred to as contingency loci, are thought to offer a preemptive strategy to increase diversity necessary for bacterial adaptation in unpredictable environments [32]. Phase variation can be mediated by DNA polymerase slipped-strand mispairing across simple sequence repeats, and when located in coding sequences, can lead to a frameshift mutation resulting in the production of truncated, often nonfunctional, peptide. Homopolymeric nucleotide tracts are one subset of simple sequence repeats commonly observed to undergo frequent expansion and contraction resulting in reversible heritable phenotypic variation [23], [32].
One variation of the V. cholerae O1 antigen that has been demonstrated and which defines the two serotypes, Ogawa and Inaba, is the presence or absence, respectively, of a terminal methyl group [16]. The two serotypes can undergo serotype conversion during epidemics or in endemic areas [33]–[38]. Spontaneous mutations in the predicted methylase wbeT (VC0258) are linked to this switching phenotype [15] and may be involved in immune evasion as cross-serotype protection is limited. In contrast to the immune pressure being somewhat specific for a given serotype, bacteriophages that target the O1 antigen for use as a receptor may not be serotype-specific. In this regard, it has recently been reported that the absorption of several different O1-specific phages to V. cholerae can be modulated by its cyclic AMP (cAMP)-cAMP receptor protein regulatory system, suggesting that regulatory pathways may exist that alter O1 antigen abundance or surface organization [39]. We recently described ICP1, an O1-specific, but serotype nonspecific V. cholerae phage that is prevalent in cholera patient stool samples in the cholera endemic region of Bangladesh [40]. ICP1 is likely related to the previously described phage JSF4 [41]. We sought to investigate the mechanisms employed by pathogenic V. cholerae O1 to resist ICP1 infection and discovered two phase variation mechanisms by which V. cholerae O1 displays intra-strain O antigen heterogeneity. This heterogeneity is mediated by two contingency loci involved in tetronate biosynthesis (wbeL) and a previously unrecognized PMI (manA) critical for perosamine biosynthesis.
Plaques resulting from the infection of a wild type V. cholerae O1 strain with the O1-specific phage ICP1 were routinely observed to have colonies growing in the center indicating the presence of phage resistant isolates. Four independent phage resistant isolates were subjected to whole genome resequencing and the majority of the isolates (three of four) had single nucleotide deletions that mapped to homonucleotide (poly-A) tracts within two genes. Two mutants were found to have a deletion in the poly-A (A8) tract starting at nucleotide position 108 in wbeL, a gene that is predicted to be required for tetronate synthesis [14]. The full-length WbeL protein is 471 amino acids and a single nucleotide deletion within the poly-A tract results in the production of a truncated peptide of 42 amino acids due to a premature stop codon 12 nucleotides downstream of the poly-A tract (Fig. S2). wbeL is unique to V. cholerae O1 strains, and the poly-A tract is 100% conserved in all 37 V. cholerae O1 strains available for analysis through the National Center for Biotechnology Information (NCBI) DNA sequence database.
Purified LPS from a wbeL* (A7) phase variant shows a distinct lower molecular weight pattern on a silver stained SDS-PAGE gel (Fig. 2), although the strain exhibits a normal slide agglutination phenotype with anti-Ogawa typing serum (Table 1). The wbeL* strain is completely resistant to infection with ICP1, and accordingly purified LPS from this strain shows no inhibition of ICP1 (Table 1). To test the possibility that the wbeL* A7 frame shift exerts a polar effect on downstream genes, which also contribute to tetronate biosynthesis (Fig. 1B), we performed complementation analysis and found that wild type LPS production and ICP1 sensitivity are restored when the wbeL* mutant is complemented with wbeL in trans (Fig. 2). This indicates that the observed phenotypes are a consequence of the loss of WbeL expression and not due to polar effects of the wbeL* mutation. To further confirm this, an in-frame deletion in wbeL was constructed and complementation analyses were performed. The ΔwbeL strain is devoid of O1 antigen (Fig. 2) and accordingly exhibited no agglutination with anti-Ogawa typing serum (Table 1), and these phenotypes can be complemented with wbeL in trans (data not shown). Since the phenotype of the in-frame deletion mutant is not consistent with the original wbeL* mutation, we hypothesized that the wbeL* allele maintains some O1 antigen biosynthetic function in the cell. Consistent with this, when the ΔwbeL strain is complemented with the wbeL* allele in trans, agglutination in the presence of anti-Ogawa typing serum is restored while the strain maintains complete resistance to ICP1, just like the original wbeL* strain (Table 1). The ΔwbeL mutant expressing wbeL* in trans also produces a small amount of lower molecular weight LPS like the original wbeL* strain (data not shown). To further address the biosynthetic function of the wbeL* allele, we explored two possible explanations for the phenotype of the wbeL* phase variant. First, that the presence of the truncated 42 amino acid peptide produced by the wbeL* allele is necessary and sufficient to allow V. cholerae O1 to elaborate the lower molecular weight LPS observed (Fig. 2). To test this hypothesis we constructed a deletion in the remainder of the coding sequence downstream of the premature stop codon in wbeL*. Purified LPS from a wbeL* strain expressing only the 42 amino acid peptide lacks O1 antigen substituted LPS (Fig. 2), which rules out this hypothesis. The second hypothesis that we tested to account for the biosynthetic role of the wbeL* allele is that it allows some functional WbeL protein to be made because the wbeL* allele is subject to nonstandard decoding (ribosomal frame shifting or transcriptional slippage at the A7 tract). This in turn would allow for a small amount of tetronate modification to occur, resulting in a lower molecular weight but still compositionally (and antigenically) normal O antigen (which is consistent with the observation that the wbeL* phase variant agglutinates in the presence of anti-Ogawa typing serum (Table 1). To test this hypothesis, we made silent point mutations within the A7 tract in wbeL* (A7 to TAAGAAA) designed to prevent non-standard decoding as well as prevent further slipped-strand mispairing during replication and characterized the ability of this strain (referred to as wbeL* PL for phase-locked) to produce O1 antigen substituted LPS. The point mutations in wbeL* PL abolished the ability of this strain to make O antigen substituted LPS, which supports the conclusion that the wbeL* allele maintains biosynthetic function by allowing some functional WbeL protein to be produced through nonstandard decoding, and that this is dependent on the A7 tract. The mechanism responsible is not currently known, nor is it known if there are other sequence motifs within the wbeL* allele that facilitate the +1 frameshifting needed to restore the reading frame. It is important to emphasize that the observed lower molecular weight pattern of purified LPS from wbeL* is not due to reversion of A7 to A8 in the genome of a substantial subset of the population, because if that were the case we would observe a small amount of wild type length O1 antigen and not the unique species observed for the wbeL* strain. These results further suggest that tetronate acylation of the perosamine backbone is necessary for incorporation of the O1 antigen into the LPS molecule, and this may be required for recognition and subsequent transport of the undecaprenyl-linked O antigen polymer to the periplasm by the ABC transporter and/or for efficient ligation of the O antigen polymer to the lipid A-core by WaaL ligase [42]. Importantly, these data demonstrate that the lower molecular weight O1 antigen produced by the wbeL* mutant somehow endows V. cholerae with full resistance to ICP1.
The second mutation identified in ICP1-resistant colonies from the center of plaques also mapped to a poly-A tract, and this localized to VC0269, which we designate manA for reasons explained below. manA has two poly-A (A9) tracts and, as illustrated in Fig. 1A, is located approximately 4 kbp downstream of the right junction gene (rjg = VC0264), which was thought to delineate the end of the O1 antigen biosynthetic locus in V. cholerae. ManA shows homology to type I PMIs that catalyze the reversible isomerization of F6P to M6P, which is the first step in perosamine biosynthesis (Fig. 1B). Like wbeL, manA is also specific to V. cholerae O1 strains: All 37 V. cholerae O1 strains available for bioinformatic analysis have manA and it is highly conserved between these strains with the exception of two strains which have the manA* (A8) allele resulting from a single nucleotide deletion in the first poly-A tract (these latter strains are designated 2740–80 and HC-61A1). The full length ManA protein is 399 amino acids, and a single nucleotide deletion within the first poly-A tract is predicted to result in a truncated peptide 81 amino acids long, while a single nucleotide deletion within the downstream poly-A tract (designated manA*) produces a truncated peptide 207 amino acids long (Fig. S3). LPS purified from manA* after overnight growth looks identical to the parental strain (Fig. 2). However, purified LPS from the manA* strain does not bind and inhibit ICP1 as efficiently as a equal amount of LPS purified from the parental strain (Table 1), which is consistent with this mutant being partially phage resistant (Fig. 2 and Table 1). These results leave open the possibility that while manA* produces wild type length O1 antigen, the overall abundance of O1 antigen substituted LPS is less than wild type, and this translates into fewer available receptors for ICP1. Phage infection is a complex process, which can require sequential receptor binding steps. For example T4 infection is initiated by the reversible attachment of at least three of the six long tail fibers to the outer core of the LPS, but does not result in DNA injection unless the six short tail fibers successfully engage their receptors on the inner core of the LPS [43]–[45]. Throughout the course of this study isolates were obtained with a frameshift in either poly-A tract in manA and these isolates were phenotypically indistinguishable from one another with regard to phage sensitivity and agglutination with anti-Ogawa typing serum (data not shown). Similarly, a strain harboring an in-frame deletion of manA was phenotypically indistinguishable from the manA phase variants (Fig. 2 and Table 1), indicating that both frame-shift mutations function as manA nulls.
manA does not appear to be required for producing full length O antigen as indicated by SDS-PAGE and silver staining of LPS (Fig. 2), and yet the manA* strain is partially resistant to ICP1, which specifically requires the O1 antigen for infection. These observations led us to speculate that there is likely another gene that can contribute to the conversion of the F6P to M6P in V. cholerae O1. We noted the presence of another annotated type I PMI in the V. cholerae O1 genome, VC1827, hereafter designated manA-2. manA-2 is located immediately downstream of a mannose permease encoded by VC1826 [46], but is not linked to the O1 antigen biosynthetic cluster. manA and manA-2 are 65% identical at the nucleotide level over 70% of their sequence, and at the protein level, they are 59% identical over 97% of their sequence. manA-2 has a higher GC content than manA (46.1% compared to 42.1%, respectively) and its GC content is much closer to the overall GC content of the entire V. cholerae N16961 genome (∼47% [47]) suggesting manA may have been recently horizontally acquired. Unlike manA, manA-2 is found in non-O1 V. cholerae. In other Gram-negatives, including E. coli and S. Typhimurium, the manA gene generally maps as an independent gene not associated with the LPS gene cluster due to its role in mannose metabolism [48], [49], although in these organisms the PMI activity of the unlinked manA is required for O antigen synthesis [50], [51]. In organisms that do not metabolize mannose, the manA gene is generally absent [49]. Some bacteria have a bifunctional type II PMI-GMP (typically referred to as ManC) encoded in the LPS biosynthetic cluster. Monofunctional and bifunctional forms of ManC are not clearly distinguishable on the basis of size or even sequence similarity [49]. Indeed ManC (VC0241) in V. cholerae O1 has the bioinformatic designation of a type II PMI, however our results below suggest this enzyme lacks PMI activity.
The presence of two putative type I PMIs in V. cholerae O1 led us to investigate the phenotypes of single and double mutants with regards to O antigen biosynthesis. A mutant lacking manA-2 was indistinguishable from wild type with regard to phage sensitivity and LPS pattern on a gel (Fig. 2 and Table 1), indicating that manA-2 is not important for O antigen biosynthesis under the conditions tested. However, in the absence of manA, manA-2 becomes important since the manA* ΔmanA-2 double mutant is completely phage resistant and produces very little fully length LPS (Fig. 2). Another potential source of M6P for perosamine biosynthesis is through the conversion of exogenously acquired mannose to M6P by a hexokinase, which led us to determine if trace mannose present in the growth media (Luria-Bertani [LB] broth) contributed to the small amount of O1 antigen still visible in the manA* ΔmanA-2 double mutant. Indeed, this small amount of O1 antigen substituted LPS is absent when the double mutant is grown in M9-glucose, but is present when the double mutant is grown in M9-glucose plus mannose (Fig. 2), demonstrating that exogenous mannose is responsible for the small amount of O1 antigen substituted LPS in the double mutant. Similar results have been observed in E. coli and S. Typhimurium manA mutants, which are unable to synthesize O antigen without the inclusion of mannose in the growth media [50], [51]. Also consistent with these results, we observed that our double manA* ΔmanA-2 mutant is unable to grow in M9-mannose owing to the critical nature of type I PMIs in the conversion of M6P to F6P as a substrate for glycolysis. These results indicate that manC in V. cholerae O1, which was hypothesized to catalyze the first reaction in biosynthesis of perosamine (Fig. 1B) [12], is not active as a bifunctional PMI-GMP, and is likely only important for the later steps in perosamine biosynthesis for converting M1P to GDP-mannose. As type II PMIs possess two catalytically distinct domains for each PMI and GMP activity [20], attempts were made to construct mutant manC alleles that should be defective for PMI activity alone (data not shown) but all constructs resulted in complete depletion of the O1 antigen, suggesting that these mutations had an inadvertent negative impact on the GMP activity of the protein. The reason behind V. cholerae O1 possessing two type I PMIs is thus not clear, although it suggests that they each have their primary roles: manA in O antigen biosynthesis and manA-2 in mannose metabolism, though it is unclear what factors define those functional roles. A search for other bacteria harboring multiple annotated type I PMIs reveals a limited number of organisms including strains of Vibrio vulnificus (Accession No. NC_005140), Vibrio parahaemolyticus (NC_004605), and Yersinia enterocolitica (NC_015224); however, to our knowledge, the functional roles that these enzymes have in these other organisms are not known.
As mentioned previously, the O1 antigen biosynthetic cluster was originally identified through the heterologous expression of the V. cholerae O1 antigen in E. coli [10], and all O1 antigen biosynthetic genes studied thus far have been between the gmhD and rjg flanking genes. manA was likely not identified as part of this biosynthetic pathway because its function was complemented by the E. coli manA gene, permitting expression of the O1 antigen in this host. Additional genes required for O1 antigen biosynthesis were subsequently identified following the initial report by Manning et al. [11], and were similarly likely missed because the phenotype was masked in E. coli. Blokesch and Schoolnik [52] provided some additional support that further extends the O1 wbe region downstream of rjg. They observed that when serogroup conversion of V. cholerae O1 to O139 occurred through uptake of O139 donor DNA during natural transformation, the crossovers were often localized within or downstream of VC0271 at the right junction, and the location of the left junction was within or upstream of gmhD. These results coupled with our observation that manA participates in O1 antigen biosynthesis suggests that the wbe region extends approximately 8 kbp downstream of rjg (Fig. 1B). There are six genes currently annotated in addition to manA in this region (Fig. 1A), however it remains to be seen if these other genes do in fact participate in O1 antigen biosynthesis.
The ability of wbeL* and manA* phase variants to colonize the small intestine was assessed in competition assays in the infant mouse model. The wbeL* strain is attenuated over 1000-fold (Fig. 3). We did not anticipate such a high level of attenuation given that this strain still elaborates LPS (although it is a lower molecular weight form, Fig. 2). Interestingly, the altered LPS produced by the wbeL* strain does provide some advantage over not having any O1 antigen, as is apparent by the significantly lower competitive index (CI) for the ΔwbeL strain (p<0.05, Mann-Whitney U test). To rule out secondary mutations, we chose to use revertant strains in vivo to avoid potential complications concerning plasmid loss and non-wild type gene expression levels during infection. The colonization defect observed for the wbeL* strain is absent when the poly-A (A7) tract is reverted to wild type length (A8) (Fig. 3), demonstrating that the virulence defect is due to the wbeL* allele. The manA* phase variant, which produces an apparently full length LPS but which is partially resistant to ICP1, is over ten-fold attenuated for colonization, and this defect is absent when the A8 tract is reverted to wild type length (A9) (Fig. 3).
We anticipated that a manA* ΔmanA-2 double mutant would be significantly more attenuated owing to a major reduction in O1 antigen produced by the strain (Fig. 2). However, although we did observe a further decrease in the CI for manA* ΔmanA-2 compared to manA*, it was not significant (p = 0.42) (Fig. 3). To try and explain this, we hypothesized that the selective pressure exerted on the manA* ΔmanA-2 double mutant in the small intestine is sufficient to select for spontaneous revertants which would potentiate the observed moderate drop in CI. To test this, we patched manA* ΔmanA-2 isolates recovered from mouse intestines onto M9-mannose agar plates. Consistent with our hypothesis, roughly half of isolates recovered from mice after 24 h of infection had regained the ability to grow on mannose, and subsequent sequencing revealed that these strains had reverted to the wild type, in-frame poly-A tract in manA. Conversely, we were unable to detect spontaneous revertants of the wbeL* or manA* strains among the isolates recovered from mouse intestines infected with those strains, suggesting that the A7 wbeL* allele is less prone to reversion by slipped-strand mispairing, and that the A8 allele in manA* is not under enough selective pressure to revert as long as ManA-2 is functional. Our inability to detect wbeL* revertants is likely due to the experimental limitation that there are so few wbeL* isolates recovered from infected mice that we were only able to test ∼100 CFU.
The manA revertants recovered from intestines of mice infected with the manA* ΔmanA-2 strain were effectively equivalent to ΔmanA-2, which we previously observed has no effect on LPS biosynthesis or virulence (Fig. 2 and Fig. 3). Confirming this, we also competed the double mutant with in-frame deletions in both manA and manA-2 and observed the anticipated significant increase in attenuation (>2000-fold) compared to a manA* phase variant alone (p<0.05 Mann-Whitney U test) (Fig. 3). Furthermore, the double ΔmanA ΔmanA-2 mutant is significantly more attenuated when grown in M9-glucose prior to infection than when it is grown in LB which contains trace mannose (p<0.05 Mann-Whitney U test). Consistent with this, a comparable CI was observed for the two strains that are devoid of all O1 antigen prior to infection (ΔmanA ΔmanA-2 grown without exogenous mannose, and ΔwbeL) (Fig. 3). In vitro control competitions revealed that the observed defects were specific to the in vivo environment, with the exception of the ΔwbeL mutant, which has a ∼10-fold defect in vitro compared to the wild type (data not shown). The nature of this defect is not known, but it could be due to the build-up of O antigen and/or LPS intermediate products in this strain which, unlike the wbeL* phase variant, is unable to elaborate any O1 antigen on the surface. Consistent with this, mutants harboring transposon insertions in wbeL, which also showed a decreased colonization phenotype in infant mice, also exhibited a decreased CI in in vitro control competitions [53].
Human intestinal epithelial cells produce antimicrobial peptides that are critical components of the host innate defense mechanism [54]. Antimicrobial peptides are inherently structured to target the membrane of bacteria because they are highly basic and have a substantial portion of hydrophobic residues [55], [56]. The net positive charge of these peptides facilitates their electrostatic interaction with negatively charged phospholipid groups or the lipid A anchor of LPS on the Gram-positive or Gram-negative bacterial membranes, respectively, allowing them to induce lysis and bacterial cell death. In order for an antimicrobial peptide to gain access to the Gram-negative outer membrane it must first traverse the barrier presented by the sugar chains of the O antigen layer. Perturbations to the LPS have been shown previously to alter resistance of V. cholerae to antimicrobial peptides; specifically, Matson et al. [57] and Hankins et al. [58] have investigated the structural importance of lipid A with regards to peptide resistance, and Nesper et al. [59] suggested that mutations affecting the LPS core oligosaccharide have a more dramatic affect on antimicrobial peptide resistance than mutations affecting O1 antigen biosynthesis (although only a rough strain was tested in those experiments).
We investigated the susceptibility of wbeL* and manA* phase variants and their trans-complemented derivatives to the antimicrobial peptide polymyxin B by determining their survival in killing assays. We observed that the wbeL* mutant exhibited a very low level of survival (Fig. 4A). This phenotype could be complemented and survival levels could be restored to wild type when this mutant was expressing wbeL in trans, again supporting the previous data suggesting there is no polar effect of wbeL* on genes downstream. With regards to the manA* phase variant, we observed polymyxin sensitivity, but in a growth phase-dependent manner. There was intermediate sensitivity of this strain to polymyxin B when the inoculum used for the killing assay was grown up to early exponential phase (OD600 = 0.15) (Fig. 4A). In contrast, when manA* was grown up to mid-exponential phase prior to treatment with the peptide, wild type levels of survival were observed (Fig. 4A). The phenotype observed at early exponential phase with the manA* phase variant could be fully complemented by expressing manA in trans (Fig. 4A), again indicating the observed phenotypes are due to the loss of ManA. We had also observed that phage sensitivity of the manA* phase variant was growth phase-dependent, and these results parallel the observations in the antimicrobial peptide assay, that is at OD600 = 0.15, manA* is completely resistant to ICP1 and at OD600≥0.2, turbid plaques result from ICP1 infection (data not shown). To address this puzzling observation, we purified LPS from manA* and wild type at early and mid-exponential growth phase (OD600 = 0.15 and OD600 = 0.5, respectively). In contrast to the seemingly wild type appearance of purified LPS from manA* after overnight growth (Fig. 2) and at mid-exponential phase, the LPS pattern of manA* isolated at early exponential growth phase showed very little O1 antigen substituted LPS (Fig. 4B). Since analysis of the double manA* ΔmanA-2 mutant indicated that manA-2 is important for O1 antigen synthesis only in the absence of manA, we interpret these results to suggest the compensatory activity of ManA-2 is incomplete during early exponential phase growth in LB broth. The reason for this is not known, but may relate to differences in expression, activity or localization of ManA-2. In any event, a complete functional redundancy between ManA and ManA-2 would have been at odds with the observation that all V. cholerae O1 strains have a phase variable manA gene: Specifically, the evolution of a contingency locus would be futile if the encoded protein exhibited complete functional redundancy with a non-phase variable gene.
In general, the results of the polymyxin B killing assays (Fig. 4A) parallel the observed in vivo colonization defects (Fig. 3). Strains that produce less O1 antigen substituted LPS (such as wbeL* and ΔmanA ΔmanA-2) are highly susceptible to polymyxin B and are more severely attenuated in vivo than manA*, which accordingly is less susceptible to polymyxin B. These data suggest that the phase variants are defective for in vivo colonization because they are more susceptible the antimicrobial peptides present in the intestinal tract.
V. cholerae O1 persists in the environment as a member of the aquatic ecosystem where it is thought to associate with and use the chitinous exoskeletons of zooplankton as a nutrient source [60]. The levels of V. cholerae O1 phages in the environment, including potentially ICP1, have been shown to inversely correlate with disease burden suggesting that phage predation in the natural environment may contribute to the collapse of a given cholera epidemic [41]. We investigated the potential for phage resistance to develop in a simulated natural environment comprised of chitin and pond water. A thousand CFU from independent cultures of wild type V. cholerae O1 were inoculated into pond water with chitin, and ICP1 was added at an MOI of 0.01. After 24 hours at 30°C, we observed that in all 11 pond microcosms to which ICP1 was added, the phage titer increased at least one million-fold (data not shown). Bacterial levels in the uninfected control had increased 10,000-fold, while the bacterial levels in the infected microcosms varied substantially from below the detection limit to levels nearly comparable to that observed in the uninfected control (Table 2). All isolates recovered from environments to which phage was added were resistant to ICP1 infection. Furthermore, the majority of isolates (63 out of 80 isolates, 79%) from independent microcosms that had become resistant to ICP1 had the wbeL* frameshift allele (Table 2). We only observed heterogeneity in the resistance mechanism for isolates from one microcosm (Table 2, number 10), suggesting that in most cases phage predation resulted in the clonal expansion of a single resistant mutant. We did not observe any mutations in the poly-A tracts of manA, likely owing to the incomplete ICP1 resistance afforded to V. cholerae with those alleles. Control experiments with manA* and wbeL* variants as input strains in the absence of phage confirmed that neither mutant exhibit a decreased ability to grow in the simulated pond microcosm (data not shown). Isolates from all microcosms to which phage were added were also replica plated onto LB agar containing 350 µg/mL polymyxin B (a concentration that permits growth of the wild type parent but not wbeL* or manA*), and all isolates were unable to grow. This indicates that the mutations that occurred less frequently and that did not map to wbeL also likely affected LPS biosynthesis. These results show that in a simulated natural environment phage predation can occur with consequent selection for bacteria with altered O1 antigen, and that the dominant mechanism by which mutational escape is achieved is through mutations in the poly-A tract in wbeL.
We also investigated the diversity of phage resistant mutants that appeared in the center of plaques on LB plates by similarly sequencing the poly-A tracts in wbeL and manA in 83 phage resistant isolates. In contrast to the experiment designed to mimic the natural aquatic environment, during selection in LB soft agar overlays in which diffusion of phage and bacteria are relatively limited, the mode of phage resistance is much more varied although a substantial portion can be attributed to phase variation in wbeL and manA (∼20%); we found that six out of 83 isolates were wbeL* phase variants and 11 out of 83 were manA* phase variants (six of the 11 had a deletion that mapped to the first poly-A tract, and the other five to the second poly-A tract). The reason for this difference in frequency of wbeL* and manA* occurrence compared to the pond microcosm is not known. However, as was mentioned previously (and will be addressed below) the manA* mutation was observed in sequenced V. cholerae O1 isolates; therefore it is apparent there are relevant circumstances in which manA* phase variants are selected for.
We wanted to determine if O antigen heterogeneity exists in the population of V. cholerae excreted along with ICP1 phage from patients during natural infection. To do this we obtained three ICP1-positive stool samples collected from three patients admitted to the International Centre for Diarrhoeal Disease Research, Bangladesh (ICDDR,B) during a cholera epidemic in 2001. Several representative isolates from all three stool samples were analyzed and were found to be V. cholerae O1 Inaba and were sensitive to ICP1. We tested the ability of these stool isolates to become phage resistant by collecting phage resistant mutants from the centers of plaques resulting from ICP1 infection and readily isolated manA* and wbeL* phase variants (data not shown), showing that these clinical isolates can phase vary at these contingency loci. Next we screened many thousands of colonies from each of these archived stool samples for the presence of manA* and wbeL* phase variants, which we hypothesized might have arisen as a result of ICP1 phage pressure during the patient infection. Since we showed above that manA* and wbeL* phase variants are attenuated in the infant mouse model, we expected that their frequency in the human stool samples would be low or perhaps even undetectable due to decreased fitness during infection of the human small intestine. As we had done previously, we took advantage of the observation that both phase variants exhibit increased sensitivity to polymyxin B compared to the parental strain. We determined that both the wbeL* and manA* phase variants obtained in vitro from these clinical isolate strain backgrounds failed to grow on LB agar plates containing 800 µg/ml polymyxin B after replica plating, while the parental strain maintained its ability to grow. We screened approximately 5,000 colonies from each stool sample by replica plating and did not observe any isolates with increased polymyxin B sensitivity.
In the above analysis we were limited by the number of available archived stool samples, since routine practice is to purify a single colony from a stool sample and store that for further examination. While such single colony isolate collections cannot be used to answer the question of whether O antigen heterogeneity exists within the population of V. cholerae excreted from a single patient, it does allow us to determine if O antigen heterogeneity exists between isolates recovered from different patients. We evaluated phage sensitivity of approximately 50 isolates recovered from cholera patients at the ICDDR,B between 2001 and 2005. Three of these isolates displayed resistance to ICP1 and were characterized further. One of the clinical isolates has a mutation in the second poly-A tract in manA. The other two resistant isolates did not have mutations that mapped to the poly-A tracts in either manA or wbeL, and examination of purified LPS from these strains showed that they produce very little full length O-antigen substituted LPS (data not shown), however the nature of the observed defects is not known.
From these results it appears that, despite the hypothesized ICP1 phage pressure during infection and despite pressure from the host immune system to reduce or alter the O1 antigen (O1 antigen is the dominant antigen [61], [62]), the intra-patient and inter-patient O1 antigen variability is quite low. In agreement with this, previous studies have detected LPS mutants in mice following coinfection of phage and V. cholerae O1, but at very low frequencies. Zahid et al., [63] estimated that the frequency of such mutants was on the order of 10−8, and in accordance with our results, failed to detect phage-resistance heterogeneity among V. cholerae O1 directly from human stools. Additionally, despite relatively high levels of O1-specific phage in the stool samples, V. cholerae O1 isolated from the same stool samples remain completely susceptible to phage lysis [63]–[65].
Previous studies of V. cholerae O antigen negative strains or strains with altered LPS structures were also found to be defective in colonization of infant mice [53], [59], [63], [66]–[69]. These results support the assertion that O1 antigen-deficient V. cholerae would be selected against during human infection due to the high fitness cost associated with mutational escape. When put into the context of our findings, it is fitting that mutational escape is frequently conferred through phase variability, as a key feature defining this mode of variability is its reversible nature [23], [32], [70]. Inherently each cell will retain its ability to switch between expression states (the switching rates of phase variable genes are typically between 10−2 to 10−5 [32]), and therefore the phenotype of a clonal population of bacteria capable of phase variation will vary as a function of selection. Predation of V. cholerae O1 in the environment by O1 antigen-dependent lytic phage may rapidly select for the subpopulation with altered O1 antigen mediated by manA* and wbeL* frame-shifted contingency loci, and even though those subpopulations are less suited for life in the intestinal tract, positive selection (for example when this mixed population is ingested by a human) results in enrichment of the subpopulation with full O1 antigen expression. We were able to experimentally confirm the reversibility of frameshift mutations occurring at the poly-A tract in manA, however, we were unable to do so for wbeL likely due to experimental limitations (discussed above). It is possible that the observed mutational escape mediated by the poly-A tract in wbeL is a function of hypermutation and not phase variation (if it is not reversible). However, we have clearly demonstrated the utility of these loci in mediating alterations in the expression of the key V. cholerae antigen and phage receptor.
Pathogenic V. cholerae O1 has evolved to live in very diverse environments including fresh water, salt water and the human small intestine. The O1 polysaccharide antigen is the dominant cholera antigen and can induce protective immune responses in humans and animals [61], [71]–[73], and thus is a critical immunogen guiding cholera vaccine development. The extent to which V. cholerae can vary expression of the O1 antigen is not currently appreciated. We demonstrate for the first time that the O1 antigen is subject to phase variation and show that this is mediated by three homonucleotide tracts in two genes (wbeL and manA), which are critical for O1 antigen biosynthesis. The ubiquitous presence of these phase variable homonucleotide tracts in all V. cholerae O1 strains points to the significant role they play in modulating expression of this surface exposed antigen. Moreover, by identifying manA as critical for O1 antigen biosynthesis, we have extended the genome boundaries previously believed to contain all the necessary genes for O1 antigen biosynthesis in V. cholerae. Phase variation mediated by homonucleotide tracts has not been previously well-documented in V. cholerae. To our knowledge, the only prior report of phase variation in V. cholerae was that by Carroll et al., [74] in which expression of the membrane bound virulence regulator, TcpH, was observed to be subject to phase variation mediated by a poly-G (G9) tract. However, with the growing list of currently available V. cholerae O1 genome sequences, it is clear that this tract is not well-conserved (only three of the available 37 sequenced strains have this tract [data not shown]), and thus this likely does not represent a wide-spread mechanism employed by V. cholerae O1 to alter virulence expression. Examination of the currently available V. cholerae O1 genome sequences may facilitate further exploration of phase variation in this organism; it is interesting to note that there are only twelve homonucleotide tracts of nine or greater nucleotides in length located within coding regions in the V. cholerae O1 N16961 genome, and several of these are located within known virulence factors (data not shown), however the significance this remains to be examined.
The biological role of phase variation in mucosal pathogens is frequently anticipated to facilitate immune evasion in the host [75]. However, in the case of the facultative pathogen V. cholerae, our data point to the primary role for O1 antigen phase variation as a strategy for dealing with the strong opposing selective pressures of phage predation in the environment and the strict requirement of O1 antigen for colonization of the intestinal tract. Phase variation of these genes thus allows for a subset of the population of V. cholerae being disseminated from a patient or being ingested in contaminated water, to be resistant to O1-dependent phages or to be virulent, respectively, thus contributing to the overall fitness of this pathogen. We hypothesize that the host immune response represents yet a second strong selective pressure against the O1 antigen, though the effects of this on circulating strains of V. cholerae within immune populations has not been studied.
The ubiquitous presence and overall success of ICP1-related phages is likely, at least in part, due to their use of a critical virulence factor as a receptor [40]. Our observation that mutational escape facilitated by wbeL and manA predominates ex vivo strongly suggests that ICP1 is particularly adept at predation of V. cholerae O1 within the human host where the requirement for colonization and virulence necessitates the maintenance of the O1 antigen. This may suggest a mechanism whereby this phage and the human host act synergistically to limit V. cholerae during infection, and perhaps how phage contribute to the overall decline of a given cholera epidemic as has been hypothesized [41], [64]. It remains to be seen if there are additional mechanisms employed by V. cholerae O1 to evade phage predation, specifically within the human intestinal tract, and how this arms race between ICP1 and its bacterial host shapes the evolution of the circulating V. cholerae O1 strains within the endemic region of Bangladesh.
Strains were grown on Luria-Bertani (LB) agar or in LB broth at 37°C with 100 µg/ml streptomycin (Sm). When indicated M9 minimal media (supplemented with trace metals, vitamins (Gibco MEM Vitamins, Invitrogen), 0.1% casamino acids) with 0.4% glucose and/or 0.4% mannose was used. Strains containing the pMMB67EH vector were grown in the presence of 100 µg/ml Sm and 50 µg/ml ampicillin (Amp). Expression from the Ptac promoter was induced by the addition of 1 mM isopropyl-β-d-thiogalactopyranoside (IPTG). Phage susceptibility was determined by the soft agar overlay method as described previously [40] and/or by measuring growth of a bacterial isolate in the presence of ICP1 (to an approximate MOI = 1) in LB plus Sm broth culture using a Bio-Tek microplate reader.
A wild type V. cholerae O1 strain (E7946) was used in standard plaque assays with phage ICP1 as previously described [40]. Following overnight incubation, colonies were routinely observed in the center of plaques indicating the presence of phage resistant isolates. Four independent colonies were chosen for further analysis including phage resistance assays and whole genome sequencing using an Illumina genome analyzer II (Tufts University Core facility) as previously described [40]. Assembled genomes were aligned to the V. cholerae O1 N16961 [47] and E7946 (unpublished data) reference genomes. Two of the independently isolated phage resistant strains had a single nucleotide deletion in the poly-A tract of wbeL (designated wbeL*), while one phage resistant derivative had a single nucleotide deletion in the second poly-A tract of manA (designated manA*). The other derivative not chosen for further study had a nonsynonymous substitution in manB.
PCRs for sequencing and cloning were carried out using EasyA polymerase (Agilent). Primer sequences are available upon request. In-frame unmarked deletions were constructed using splicing by overlap extension (SOE) PCR [76] and introduced using pCVD442-lac [77]. Deletion alleles constructed in this study are missing the entire open reading frame, except for the start and stop codons (with the exception of the wbeL deletion allele which also preserved a single codon immediately upstream of the stop codon). Expression plasmids were constructed by cloning the desired open reading frame(s) (including the predicted ribosome binding site) into the multiple cloning site of pMMB67EH. Expression vectors were transferred into V. cholerae by conjugation with E. coli SM10λpir and selection of SmR AmpR colonies. Strains utilized in this study are shown in Table 3.
Slide agglutination tests were performed using V. cholerae O1 Ogawa polyclonal rabbit antiserum (Difco). LPS was extracted from overnight cultures unless otherwise indicated, as described previously [72]. Briefly, cultures were centrifuged and washed twice in TM buffer (50 mM Tris [pH 7.5], 10 mM MgCl2) supplemented with 1 mM DL-Dithiothreitol before being lysed by bead-beating (BioSpec Products, Inc.) with 0.1 mm zirconia beads for a total of three minutes with intermittent incubations on ice. Whole cell lysates were treated with proteinase-K (Sigma) at 37°C for 24–48 h as required. Phenol extraction was performed using phase-lock gel light tubes (Eppendorf). Extracts were centrifuged at 75,000× g for 60 min, the pellet was washed with TM buffer and centrifuged as before. Purified LPS was separated on a 4–12% NuPage Bis-Tris gel (Invitrogen) and visualized by silver-staining (SilverQuest, Invitrogen). The concentration of V. cholerae LPS was determined by comparison to a standard curve of E. coli O26:B6 LPS (Sigma) using a Fujifilm FLA-900 scanner as previously described [72]. The ability of purified LPS to neutralize plaque formation was determined as previously described [40].
In vivo competition experiments were done using 4–5 day old CD-1 mice. The dams and their litters were housed with food and water ad libitum and monitored in accordance with the rules of the Department of Laboratory Animal Medicine at Tufts Medical Center. The inoculum was prepared as a 1∶1 mixture of the strain of interest (lacZ+) and the appropriate control strain (ΔlacZ). Mice were infected intragastrically with approximately ∼105 CFU and sacrificed 24 hours post-infection. Small intestines were homogenized in 1 ml LB+16% glycerol, diluted in LB broth, and plated on LB agar plates containing 100 µg/ml Sm and 40 µg/ml 5-bromo-4-chloro-3-indolyl-β-d-galactopyranoside (X-gal). The competitive index was calculated as the ratio of the mutant compared to the control strain normalized to the input ratio. In vitro controls were included in each of these experiments in which the same inoculum was diluted 1∶100 into at least five independent LB cultures and the output ratios of mutant to the control strain were determined on Sm X-gal agar plates as above.
Polymyxin B killing assays were done as previously described with minor modifications [57]. Briefly, overnight cultures were subcultured 1∶100 into LB and grown at 37°C to the desired OD (OD600 = 0.15 and OD600 = 0.5). 5 µl polymyxin B (Invitrogen) at 500 µg/ml was added to 45 µl of the above culture in a well of a 96-well polypropylene microtiter plate to obtain a final test concentration of 50 µg/ml polymyxin B. After three hours of incubation at 37°C with shaking, serial dilutions of each culture were plated on LB Sm plates. The percent survival was calculated as (CFU(polymyxin treatment)/CFU(untreated))×100. The average percent survival was determined from two biological replicates, each having been done in technical duplicate.
Overnight cultures of wild type V. cholerae E7946 were serially diluted in filter sterilized pond water to approximately 105 CFU/ml. 10 µl of diluted culture (103 CFU) was used to inoculate 1 ml chitin solution (1% chitin from crab shells [Sigma] in filter sterilized pond water). To assess the impact of phage on the appearance of phase variants under these conditions, approximately 10 PFU of ICP1 was immediately added following inoculation of bacteria (MOI = 0.01). The mixture was allowed to incubate for 24 hours at 30°C statically at which time the mixture was vortexed and plated for CFU. ICP1 was enumerated by adding chloroform to a 100 µl aliquot of the above solution, diluted and plated for PFU with E7946 using the soft agar overlay method as described above.
Three cholera stool samples collected at the ICDDR,B in 2001 and stored in the presence of glycerol were assayed for the presence of isolates with altered O1 antigen. Single isolates from each sample were found to be O1 Inaba that were sensitive to ICP1. wbeL* and manA* mutants of this clinical O1 Inaba isolate were recovered after plating with ICP1 and used to assess the applicability of replica plating on polymyxin B as a tool to identify heterogeneity within a stool sample. Both the wbeL* and manA* isolates in this background failed to grow on LB agar plates containing 800 µg/ml polymyxin B, while the parental strain maintained its ability to grow. Each stool sample was plated on LB agar containing 100 µg/ml Sm and incubated overnight at 37°C. Plates were then replica plated onto polymyxin B plates and incubated overnight at 37°C to identify polymyxin B sensitive isolates in the stool sample. Approximately 5000 colonies were analyzed per stool sample.
All animal experiments were done in accordance with NIH guidelines, the Animal Welfare Act and US federal law. The experimental protocol using animals was approved by Tuft University School of Medicine's Institutional Animal Care and Use Committee. All animals were housed in a centralized and AAALAC-accredited research animal facility that is fully staffed with trained husbandry, technical, and veterinary personnel.
|
10.1371/journal.ppat.1007759 | Functional expression and characterization of the envelope glycoprotein E1E2 heterodimer of hepatitis C virus | Hepatitis C virus (HCV) is a member of Hepacivirus and belongs to the family of Flaviviridae. HCV infects millions of people worldwide and may lead to cirrhosis and hepatocellular carcinoma. HCV envelope proteins, E1 and E2, play critical roles in viral cell entry and act as major epitopes for neutralizing antibodies. However, unlike other known flaviviruses, it has been challenging to study HCV envelope proteins E1E2 in the past decades as the in vitro expressed E1E2 heterodimers are usually of poor quality, making the structural and functional characterization difficult. Here we express the ectodomains of HCV E1E2 heterodimer with either an Fc-tag or a de novo designed heterodimeric tag and are able to isolate soluble E1E2 heterodimer suitable for functional and structural studies. Then we characterize the E1E2 heterodimer by electron microscopy and model the structure by the coevolution based modeling strategy with Rosetta, revealing the potential interactions between E1 and E2. Moreover, the E1E2 heterodimer is applied to examine the interactions with the known HCV receptors, neutralizing antibodies as well as the inhibition of HCV infection, confirming the functionality of the E1E2 heterodimer and the binding profiles of E1E2 with the cellular receptors. Therefore, the expressed E1E2 heterodimer would be a valuable target for both viral studies and vaccination against HCV.
| Hepatitis C virus (HCV) is an enveloped virus that infects millions of people worldwide and may lead to cirrhosis and hepatocellular carcinoma. HCV has two envelope proteins, E1 and E2, which form heterodimers on viral surface and are critical for HCV cell entry. However, current studies of HCV E1E2 are often limited by the poor quality of the in vitro expressed E1E2 heterodimers. Here we express the ectodomains of HCV E1E2 with different tags, and are able to isolate soluble E1E2 ectodomains suitable for structural and functional studies. Then we generate the 3D reconstruction of E1E2 heterodimer by electron microscopy and also model the E1E2 structure by the coevolution based strategy with Rosetta, showing the potential interactions between E1 and E2. Moreover, the E1E2 heterodimer is applied to examine the interactions with the HCV cellular receptors, neutralizing antibodies as well as the inhibition of HCV infection. These results suggest that the expressed E1E2 heterodimer would be a promising target for both viral studies and vaccination against HCV.
| Hepatitis C virus (HCV) is an enveloped positive-stranded RNA virus that belongs to the genus Hepacivirus in the family of Flaviviridae [1, 2]. Its genome consists of a single open reading frame encoding a protein product, which is cleaved by cellular and viral proteases into ten smaller proteins, including three structural proteins, namely core protein, E1 and E2, and seven nonstructural proteins [3]. HCV causes both acute and chronic infections, and the chronic infection may lead to liver diseases such as cirrhosis, hepatocellular carcinoma and liver failure [4]. According to the statistics from WHO, approximately 71 million people have chronic HCV infection globally and nearly 400,000 people die each year from hepatitis C, mostly through cirrhosis and hepatocellular carcinoma. Current antiviral medicines against HCV show high cure rates (>95%), but the high cost, side effects, viral resistance and the potential of reinfection [5–7] are limiting the antiviral effects. Up to date, no vaccine is available for HCV, largely due to its high polymorphism in genotypes and morphologies. The lack of structural information also hampers the development of HCV vaccines.
HCV has two envelope glycoproteins, E1 and E2, which mediate the cell entry through the interactions with host cell receptors and are promising targets for vaccine development. A large number of studies have shown that several cell surface receptors are involved in HCV cell entry. Among them, glycoprotein E2 has been reported to interact directly with tetraspanin/CD81 [8–11], scavenger receptor class B member 1 (SR-B1) [12, 13] and very low density lipoprotein receptor (VLDLR) [14]. Glycoprotein E1 is suggested to be responsible for the fusion between viral and cellular endosomal membranes during HCV entry process and it might also interact with the apolipoprotein E (ApoE) [15, 16]. In addition, several other receptors have also been reported to be important for HCV cell entry, for example, claudin-1 (CLDN1) [17], occludin (OCLN) [18], and NCP1L1 [19]. However, the exact roles of these receptors in viral entry are not fully understood [20].
Both E1 and E2 of HCV are type I transmembrane proteins containing an N-terminal ectodomain (160 residues for E1 and 330 residues for E2) and a well-conserved C-terminal transmembrane domain of about 30 amino acids [21]. E1 and E2 form heterodimers on viral surface, which is important for the maturation as well as the infectivity of viral particles, and their transmembrane domains might be involved in the heterodimerization process [20, 22]. E1 and E2 ectodomains have five and eleven potential glycosylation sites, respectively, and these carbohydrates might be important for the stability and antigenicity of HCV particles [23]. Moreover, E1 and E2 ectodomains also contain eight and eighteen cysteines, respectively, which can form both intra- and inter- molecular disulfide bonds that may affect the host receptor interactions [20, 24]. Glycosylation and disulfide bonds might be critical for the folding and maturation of E1 and E2 as overexpression of these proteins often results in mis-folded disulfide bond-linked aggregates [25, 26]. Up to date, only partial structural information of E1E2 heterodimer is available. The crystal structure of an N-terminal fragment of E1 ectodomain shows a covalently linked domain-swapped homodimer [27]. The core of E2 has been solved in complex with antibodies [28–30], where E2 adopts a central immunoglobin-like fold formed by β-sheets surrounded by short α-helices dispersed in loops. However, both E1 and E2 used for structural studies are truncated proteins, part of the ectodomains are missing in the solved structures [31, 32]. Meanwhile, several groups have been trying to co-express E1E2 heterodimer to ensure the correct folding of the intact proteins as the folding and maturation of E1 and E2 may depend on each other [33–36]. Unfortunately, co-expression of E1E2 heterodimer often leads to poor quality samples and the structural characterization of the intact E1E2 heterodimer has not been successful.
Here we expressed the ectodomains of HCV E1E2 heterodimer fused with either an Fc-tag or a de novo designed heterodimeric tag and characterized the structure of E1E2 heterodimer with both electron microscopic reconstruction and the coevolution-guided modeling using Rosetta. Moreover, we examined the interactions of the E1E2 heterodimer with the known HCV receptors and neutralizing antibodies and the inhibition of HCV infection by the heterodimer, suggesting that the expressed E1E2 heterodimer is functional and could be a valuable target for further structural studies and vaccine development against HCV.
In order to obtain E1E2 heterodimer suitable for functional and structural studies, we first tried to co-express E1 and E2 ectodomains (from HCV strain H1b) in insect cells by replacing the transmembrane domains of E1 and E2 with a Flag-tag and a 6xHis-tag, respectively. But almost no expression of E1 and low expression of E2 were detected. This is not surprising as the folding of E1 and E2 may depend on each other, and the transmembrane domains of E1 and E2 have been shown to be important for the formation of E1E2 heterodimer [22]. Therefore, we designed a construct using an IgG Fc fragment to substitute the transmembrane domains of E1 and E2 (Fig 1A), where the IgG Fc fragments would dimerize and may facilitate the formation of E1E2 heterodimer. To isolate the E1E2 heterodimer, a Flag-tag at the C-terminus of E1-Fc and a 6xHis-tag at the C-terminus of E2-Fc were added for affinity chromatography (Fig 1A). The construct was expressed as secreted soluble forms in insect cells and the purified E1E2 heterodimers were obtained after two consecutive affinity purification steps with Ni-NTA resin and anti-Flag M2 resin. The size-exclusion chromatography (SEC) showed that the E1E2 proteins contained both heterodimers and oligomers, which correspond to the two peaks in the chromatogram (Fig 1A). Proteins from both peaks were loaded onto SDS-PAGE under reducing and non-reducing conditions (Fig 1B). The peak that corresponds to the heterodimers showed a single band with the molecular weight equal to E1E2-Fc under non-reducing conditions, while E1 and E2 were separated under reducing conditions, which is not unexpected as there are disulfide bonds in Fc tag. Similarly, the peak that corresponds to the oligomer also showed a smeared band with high molecular weights under non-reducing conditions (Fig 1B), suggesting they might be disulfide bond-linked oligomers, which has been reported before [20, 24, 37–40]. The E1E2 heterodimer was collected and showed mono-dispersed particles by negative stain EM imaging (Fig 1C). The two-dimensional (2D) class averages of the boxed particles showed that E1E2-Fc contained a head and a tail region. The head region should correspond to the E1E2 heterodimer and the tail region is formed by the Fc homodimer (Fig 1C). The E1E2 oligomers were also collected for negative stain EM, and the images showed larger particles with blurry 2D class averages, implying that the oligomers might be heterogeneous (S3A Fig).
During the expression of E1E2 with Fc tag, E2-Fc homodimer was also found as expected, however, no E1-Fc homodimer was detected in supernatant, suggesting that E2 might be required for the secretion of E1. To validate this result, both E1-Fc and E2-Fc were expressed individually using the similar expression system in insect cells, and indeed, E2-Fc can be found in supernatant (S1A Fig), whereas E1-Fc can only be detected in cell pellets (S1C Fig), suggesting that E1-Fc alone may not fold properly and are retained intracellularly. The mono-dispersed particles of E2-Fc homodimer could be seen on the negatively stained EM images and the 2D class averages also showed that E2-Fc had two regions corresponding to the E2 homodimer and the Fc tail, respectively (S1B Fig). And the E2 homodimer revealed different features from the E1E2 heterodimer in the 2D averaged images, which is expected and confirms the formation of E1E2 heterodimer with Fc tag.
To mimic the native folding and maturation of E1E2, we used a pair of de novo designed helical hairpins (DHD15, PDB entry: 6DMA) to replace the transmembrane domains of E1 and E2 (Fig 2A). The designed helical hairpins only form heterodimers specifically [41], therefore could maximize the yield of E1E2 heterodimer. In addition, the N-terminus of each helical hairpin locates close to each other, allowing the direct fusion of E1 and E2. Furthermore, the de novo designed hairpin heterodimer is thermal stable, which may facilitate the folding and maturation of E1E2 heterodimer.
The construct of E1E2-DHD15 was expressed in insect cells and purified using both Flag-tag and His-tag followed by SEC as described above. Similarly, the SEC profile showed two peaks for E1E2-DHD15 (Fig 2A), which corresponded to E1E2 heterodimer and oligomer, respectively. The E1E2-DHD15 heterodimer was loaded onto SDS-PAGE under both reducing and non-reducing conditions (Fig 2B). The non-reducing SDS-PAGE showed a single band with the molecular weight equal to E1E2-DHD15, while the bands of both E1 and E2 are detected under reducing conditions. Since there is no disulfide bond in the DHD15 heterodimeric tag, the results suggest that disulfide bonds might be formed between E1 and E2 in the expressed heterodimers. Meanwhile, the fractions from the oligomer peak were also loaded onto SDS-PAGE under both reducing and non-reducing conditions (Fig 2B), the non-reducing SDS-PAGE showed a smeared band with high molecular weights, while the bands of E1 and E2 are separated under reducing conditions, confirming the formation of disulfide bond-linked oligomers during expression, which is consistent with the data reported before [24]. Moreover, fractions from both peaks were negatively stained and observed under EM. The images of the heterodimer showed uniform particles and the 2D averaged images revealed a doughnut-shaped head and a tail, which correspond to the E1E2 heterodimer and the DHD15 tag, respectively (Fig 2C). By contrast, the EM images of the E1E2-DHD15 oligomer showed larger heterogeneous particles (S3B Fig), consistent with the results of SEC and SDS-PAGE.
Glycosylation has been shown to be important for the function of E1E2 [23]. Considering the different glycosylation patterns generated by insect and mammalian cells, we expressed both DHD15-tagged and Fc-tagged E1E2 heterodimer in mammalian cells. Both E1E2-DHD15 and E1E2-Fc expressed in HEK293 cells showed smeared bands with higher molecular weights on SDS-PAGE (Fig 3A and 3C), which is expected as mammalian cells usually produce larger and nonuniform glycosylation patterns. The SEC peak of E1E2-DHD15 also contained two species, E1E2 heterodimer and oligomer (Fig 3A), and both of them showed single bands under non-reducing conditions, whereas E1 and E2 were separated under reducing conditions, indicating the formation of disulfide-bond linked heterodimer or oligomer, which is similar to the proteins expressed in insect cells described above. The E1E2-Fc expression in HEK293 cells behaved similarly as E1E2-DHD15 (Fig 3C). These results suggest that the disulfide-bond linked E1E2 heterodimer and oligomer are formed independent of the expression systems, which is in agreement with data from other groups [20, 24, 37–40]. In addition, the negatively stained EM images also showed mono-dispersed particles for both samples (Fig 3B and 3D), however, the 2D averaged images did not show clear features as the insect cell expressed proteins, probably due to the heterogeneity resulting from the larger and nonuniform glycosylation of the samples.
Similar to the insect cell expression system, E2-Fc alone could be expressed and secreted properly by mammalian cells (S2A Fig), but no E1-Fc could be detected in supernatant if expressed by itself (S2C Fig). The negatively stained EM images showed that the mammalian cell expressed E2-Fc were also mono-dispersed, and the head and the tail regions can be seen in the 2D averaged images (S2B Fig). In parallel, we also expressed the E1E2 heterodimer from a different HCV strain, genotype 1a H77, and the similar results were obtained (S2D Fig), suggesting that the co-expression system described above could be applied to other HCV strains to obtain soluble E1E2 heterodimers.
In the meantime, we also treated E1E2 heterodimer with Endoglycosidase H (Endo H) as it has been shown that the virion-associated mature HCV glycoproteins are resistant to Endo H treatment [24, 39]. Indeed, the E1E2 proteins expressed in HEK293 cells were resistant to Endo H treatment (S6B Fig), whereas the insect cell expressed E1E2 proteins could be slightly deglycosylated by Endo H (S6A Fig). The negatively stained EM images were also collected for the Endo H treated E1E2-DHD15 expressed in insect cells, and the images showed well-dispersed particles sharing similar features with the untreated proteins (S6C Fig).
To validate the folding of the expressed E1E2 proteins, we tested the interactions of the Fc- and DHD15-tagged E1E2 heterodimers with the known HCV neutralizing antibodies. AR3A is an E2-specific antibody recognizing a discontinuous epitope on E2 and has been shown to be able to block the binding of E1E2 to CD81 [42, 43]. The ELISA data showed that AR3A could bind to the E1E2 heterodimers expressed in both insect and mammalian cells (Fig 3E and 3H). In parallel, the Fab fragments of neutralizing antibody IGH526, which recognizes a conformational epitope on E1 and may also have minimal binding activity to E2 [44], and antibody HCV1, which binds to a β-hairpin motif on E2 [45], were expressed and purified from HEK293 cells (S9 Fig). Similarly, the binding data showed that both IGH526 and HCV1 could bind to the E1E2 heterodimer expressed in both insect and HEK293 cells (Fig 3). In particular, IGH526 showed much higher binding activity to E1E2 heterodimer than E2 homodimer (Fig 3I), consistent with the reported data for this antibody [44]. Therefore, these results suggest that the E1E2 proteins expressed by our strategies have the similar epitopes as the E1E2 on viral surface. Furthermore, we also tested the binding of the oligomeric fractions of E1E2 with the neutralizing antibodies described above, and the data showed that E1E2 oligomers could bind to the neutralizing antibodies as well (S7 Fig), suggesting that the oligomers of E1E2 also contain correct epitopes.
To investigate the structure of E1E2, we applied 3D electron microscopy reconstruction to the E1E2 heterodimers expressed in insect cells as they showed mono-dispersed homogenous particles on negative staining EM images. The 2D averaged images showed that the E1E2 region of the E1E2-Fc heterodimer adopted a doughnut-like conformation (Fig 1C). But the 3D reconstruction based on the 2D images was not successful as the Fc portion, which occupies roughly one third of the total volume, was rather flexible relative to the E1E2 portion, thus making the particle alignment difficult for 3D reconstruction. By contrast, DHD15 tag is smaller with relatively low flexibility, allowing us to reconstruct a 3D model of E1E2-DHD15 heterodimer based on the negatively stained EM images at ~27 Å resolution (Fig 4).
The EM reconstruction of E1E2-DHD15 showed a volume with a doughnut-like head and a tail (Fig 4). The de novo designed helical bundle DHD15 can be fitted into the tail volume reasonably well (Fig 4A). The head that corresponds to the E1E2 heterodimer can be roughly divided into two density blobs at the high-density contour level (Fig 4A), which might correspond to the ectodomains of E2 and E1, respectively. Docking of the known crystal structures of E1 and E2 into the EM volume is difficult as these structures are truncated and only occupy roughly 50% of total EM volume (Fig 4A) [27–29].
It has been reported that during HCV entry, viral particles are internalized and transported into endosomes where E1E2 may undergo conformational changes in response to acidic pH [46–48]. Therefore, we incubated E1E2-DHD15 in acidic buffer (pH 5.5) overnight and imaged under EM. The resulting 2D class averages and the 3D reconstruction reveal no obvious difference (S4 Fig), suggesting that the pH-induced conformational change of E1E2 may not be large enough to be visible at low resolution. Another possibility is that the E1E2 heterodimer requires post-attachment priming steps before it responds to low pH during viral entry [47], as in the case of pestiviruses [49].
To further explore the structure of E1E2 heterodimer, we modeled the E1E2 heterodimer by combining the coevolution analysis with GREMLIN [50] and the molecular modeling with Rosetta [51, 52] in the context of the 3D EM model. According to the coevolution theory, the coevolving residues within or between proteins usually form spatial contacts, and such information would facilitate protein structure prediction by Rosetta [53–55].
The accuracy of coevolution-based contact prediction depends on the availability of a large amount of diverse (< 90% sequence identity) sequences where coevolving residues can be detected unambiguously. There are about 50,000 different sequences of HCV glycoproteins in the database. Although these sequences are not diverse enough to guarantee accurate contact predictions based on the previous studies [54], we hypothesized that the massive amount of somewhat different (about 95% sequence identity) sequences may still contain valuable coevolution signals. We calibrated the prediction accuracy using residue pairs present in the crystal structures of E2 (PDB entry: 4MWF and 4WEB) [28, 29] (Fig 5A). The top 0.5L (L is the sum of the length of E1 and E2) predicted contacts are expected to contain 70% correct predictions, and the contacts between E1 and E2 among these predictions are listed in Fig 5C. The calibration also allowed us to assign a probability of being correct to each predicted contact, and the top 0.5L contacts that are separated by at least two residues (Fig 5B) are used as constraints for modeling with Rosetta.
In addition to the coevolution constraints, we used the partial structures of E1 (excluding the swapped beta hairpin from PDB entry: 4UOI) and E2 (PDB entry: 4MWF and 4WEB) as templates to model the complete ectodomains of E1 and E2 with Rosetta. The top five models ranked by Rosetta energy function (including atom pair constraints) are inspected and selected based on the agreement to the EM density and the satisfaction to the coevolution constraints between E1 and E2. The selected models of E1E2 heterodimer were manually docked into the EM density and further refined using Rosetta with coevolution constraints. The resulting model shows that the ectodomains of both E1 and E2 can be divided into two parts, an N-terminal region and a stem region (Fig 4B), and E1 and E2 have two interfaces: one locates at the membrane distal ends of the N-terminal regions and the other one stays near the membrane proximal ends of the stem regions (Figs 4 and 5D). The modeled N-terminal region of E2 is similar to the solved crystal structures, except that the hyper variable loop regions are rebuilt by the Rosetta loop modeling protocol. Deletion of hyper variable region 2 (HVR2) has been reported to abolish the formation of E1E2 heterodimers, indicating its important role in E1E2 heterodimer formation [56], which is consistent with our model where HVR2 comprises a large portion of the E1E2 interface (Fig 4B). Moreover, the model is also in agreement with the previously results showing that the back sheet region of E2 may interact with E1 [57]. In addition, a broadly neutralizing antibody AR3C has been showed to be able to block the binding of CD81 to HCV [28, 43]. The superimposition of the E2-AR3C complex structure [28] with the E1E2 heterodimer model shows that the CD81 binding site locates on the side of the E1E2 heterodimer, away from the E1E2 interface (S4C Fig).
The stem regions are commonly found in the glycoproteins in the Flaviviridae family [10, 58] and may play important roles in viral entry [59–63]. In the E1E2 structural model, the stems regions are composed of helices and interact with each other, which is in agreement with the strong coevolution signal between Q335 and S706 (Fig 5C). In addition, these helices are hydrophobic and have positive WWIHS scores [64], suggesting they may interact with the lipid membrane.
It is noteworthy that the putative fusion peptide of E1 (residue 272–285: CSAMYVGDLCGSVF), which has been suggested to be important for triggering the fusion process during HCV entry [15, 59, 65–68], forms a helix in our model and interacts directly with E2 (Fig 4B). This interaction is supported by the strong coevolutional signal between residues F284 of E1 and W645 of E2 (Fig 5C and 5D). Moreover, the structural modeling of E1 by Rosetta also gives several alternative conformations of the fusion peptide with comparable Rosetta energies. Among them, the extended helix of the peptide could bend in the middle and form a helical hairpin, suggesting that large conformational changes are allowed for the fusion peptide, which may be relevant to the fusion process. In addition, since the fusion peptide is quite hydrophobic, it could be unstable if exposed in the absence of E2, which may explain why E2 is required for the functional expression of E1.
In order to verify the model, we generated two deletion mutants at the interface between E1 and E2, including a HVR2 deletion mutant on E2 and a putative fusion peptide deletion mutant on E1. Moreover, two double mutants of the coevolving residues that may form hydrophobic contacts with each other at the E1E2 interface, including I307S &L653S and F284S&W645S, were also made. In parallel, another version of the two double mutants, I307R&L653R and F284R&W645R, were constructed to increase the probability of disrupting the interface. All the mutants were generated based on the E1E2-DHD15 construct and expressed in insect cells, which would only produce E1E2 heterodimers. However, none of the mutants were detected in supernatants, and the western blot data showed that all these mutants were expressed but retained intracellularly (S5A and S5B Fig), suggesting that the heterodimers were not formed properly probably due to the disruption of the interface between E1 and E2. By contrast, a single mutant Q466R, which showed strong coevolving signal with residue G198 on E1 (Fig 5C and 5D), could be expressed and secreted into supernatant (S5A and S5B Fig). Since Q466 locates at the peripheral region of the interface in the model (Fig 5D), therefore may not be able to affect the interface as other mutated hydrophobic residues, and indeed, the EM imaging of this mutant did not show any obvious difference with the wild type samples (S5C Fig).
A number of cell surface receptors have been reported for HCV cell entry, but the specific interactions between E1E2 and the receptors are not fully characterized. We first examined the interactions of E1E2 with the receptors using E1E2-Fc heterodimers. CD81 is a known HCV receptor that binds to E2 [8–11]. Indeed, the GST-pull down assays showed that the mammalian cell expressed E1E2-Fc heterodimer could bind to CD81 (Fig 6B), and according to the ELISA assays, the CD81 binding affinities of E1E2-Fc and E2-Fc were similar (Fig 6A), suggesting that CD81 might bind to E2 directly and may not have interactions with E1. This is consistent with the modeled structure of E1E2, where the CD81 binding site locates on the side of E2, far away from the E1E2 interface (Fig 5D and S4C Fig). By contrast, apolipoprotein E (ApoE) has been reported to facilitate the HCV entry through its interaction with E1 [16]. The ELISA results showed that E1E2-Fc could bind to ApoE, whereas E2-Fc only bound to ApoE at background level, confirming the recognition between ApoE and E1 (Fig 6C). Among the HCV receptors, very-low-density lipoprotein receptor (VLDLR) has been shown to mediate HCV entry independent of CD81, and E2 plays an important role in this process [14]. Indeed, the ELISA results showed that both E1E2-Fc and E2-Fc bound to VLDLR similarly (Fig 6D), which is consistent with previously studies. In addition, we also tested the interaction between oligomeric E1E2 and CD81, and the results showed that the E1E2 oligomer could also bind CD81 (S7D Fig), as has been reported previously [24].
In the meantime, we also tested the interactions of E1E2-Fc heterodimer with other HCV receptors, including SR-B1, OCLN and CLDN1. Since these receptors are multi-pass transmembrane proteins and difficult to isolate, we transfected HEK293 cells with the full-length receptors fused with GFP and monitored the binding of E1E2-Fc heterodimer by FACS. Among them, SR-B1 could bind to both E1E2-Fc and E2-Fc similarly (Fig 6E), suggesting that SR-B1 might interact mainly with E2, which is in agreement with the previous results [12]. By contrast, we were not able to detect any binding signals of E1E2-Fc on the OCLN or the CLDN1 transfected cells (Fig 6F and 6G), this is somewhat unexpected as these two molecules have been shown to be indispensable for HCV entry to murine cells [17, 18]. However, it has also been reported that other cofactors might be required for E1E2 to recognize OCLN and CLDN1 [69], therefore they may not interact directly with E1E2 during viral entry. In parallel, we also tested the binding of E1E2-DHD15 with CD81, SR-B1, OCLN and CLDN1, respectively, and similar results were obtained (S8 Fig).
To further confirm the functional activities of the expressed E1E2 heterodimer, we tested the inhibition of HCV infection with the expressed E1E2 heterodimer or E2 homodimer. The results showed that both E1E2 heterodimer and E2 homodimer could block the HCV infection, therefore validating the functionality of the E1E2 heterodimer (Fig 6H). Among them, the E1E2-Fc heterodimer expressed in mammalian cells appeared to block the infection better than other constructs, suggesting that both E1 and E2 as well as the glycosylation pattern of the envelope proteins may all affect the viral entry. Taken together, these results confirmed that the expressed soluble E1E2 heterodimer described above is functional and could be applied to explore the HCV entry mechanism and might also be a valuable target for developing prophylactic vaccines against HCV.
Although HCV was identified nearly thirty years ago [70], the structure and the life cycle of HCV have not been fully understood. Current publications suggest that HCV cell entry is a multistep process involving a number of receptors in a temporally and spatially ordered manner [1, 2], and the two envelope glycoproteins, E1 and E2, are the key players in the viral cell entry process. Probably due to the complex folding and maturation process of E1 and E2, the native form of E1E2 is difficult to isolate [33, 35, 36]. E1 and E2 have been shown to form a heterodimer through their transmembrane domains on viral surface, and the folding and maturation of E1 and E2 may depend on each other [20, 25, 26]. To mimic the native expression of E1E2 glycoproteins, we utilize either an IgG Fc region or a de novo designed heterodimeric tag to substitute the transmembrane domains of E1 and E2, resulting in E1E2 heterodimers similar to the native form of E1E2 on HCV particles, which has been validated by the binding of neutralizing antibodies that recognize conformational epitopes on both E1 and E2. Previous evidence has shown that the intracellular forms of E1 and E2 might be assembled as non-covalent heterodimers, whereas the virion-associated envelope glycoproteins could form covalent dimers or oligomers stabilized by disulfide bonds, and the disulfide bond-linked E1E2 complexes were in a conformation competent for cell entry [24]. Interestingly, the E1E2 proteins expressed by our strategy usually contain two species, E1E2 heterodimers and oligomers, and they all form inter-subunit disulfide bonds and could bind to the neutralizing antibodies and cellular receptors, which is consistent with the findings about the virion-associated E1E2 glycoproteins.
Previously, similar co-expression systems have been used successfully with either GNA enrichment method or fusing E2 with an Fc tag to facilitate purification [33, 71–74]. An advantage of the strategy described here is that the E1E2 heterodimers are expressed as water-soluble forms and secreted into media with reasonable yields, which makes purification and functional characterization much easier than isolating the heterodimers from membranes with detergents. The EM images also show that the purified E1E2 heterodimers are mono-dispersed and suitable for further structural studies.
De novo protein design has recently shown significant success in therapeutic drugs, new enzymes and biocatalysts, drug delivery tools and other applications [75]. Here we use a de novo designed helical bundle DHD15 to induce soluble heterodimer formation by replacing the transmembrane domains of E1 and E2. The DHD15 tag has several advantages: (1) The two N-termini of DHD15 are close to each other, which can be fused to the ectodomains of E1 and E2 without introducing extra geometric hindrance. The helical bundle of DHD15 also mimics the conformation of the transmembrane domains of native E1E2; (2) DHD15 can form stable heterodimer quickly and facilitate the folding and maturation of E1 and E2; (3) Since DHD15 forms a heterodimer, it could maximize the yield of E1E2 heterodimer during expression, because in the case of Fc tag, both E1E1 and E2E2 homodimers are also produced during expression; (4) The helical bundle of DHD15 is quite rigid containing only 75 amino acids with no disulfide bonds or glycosylation sites, making it suitable for crystallographic and EM studies. The application of this de novo designed heterodimeric tag for HCV E1E2 glycoproteins suggest that computational protein design could be a powerful tool to facilitate biological researches.
HCV glycoproteins are challenging targets to study with current structural approaches due to the conformational flexibility, glycosylation and folding requirements, and only partial structural information is available for E1 and E2 [27–29, 76]. Given the high sequence variability and the availability of a vast number of sequences, the E1E2 of HCV is a reasonable target for the in silico modeling using coevolution information derived from sequence alignments. In the context of the low-resolution EM reconstruction, the intact E1E2 ectodomain has been modeled by Rosetta using the coevolution information regarding the residue contacts. The E1E2 heterodimer roughly forms a doughnut-like conformation with two interfaces between E1 and E2 (Fig 4). Both E1 and E2 ectodomains exhibit elongated electron densities in the EM reconstruction and can be divided into an N-terminal region and a stem region. The putative fusion peptide on E1, the HVR2 and the back-sheet region on E2 are involved in forming the membrane distal interface between E1 and E2 in our model. This is in consistent with previous experimental and computational studies, for example, the HVR2 region on E2 has been shown to play an important role in E1E2 heterodimer formation [56], and the coevolution analysis shows the critical role of the back sheet region on E2 in the E1E2 interface. Moreover, a recent high-throughput mutagenesis study also emphasizes the importance of HVR2 and the back-sheet region in the heterodimer formation [77]. The stem regions of E1E2 include hydrophobic helices, which might be involved in forming the membrane proximal interface between E1 and E2, since the coevolution analysis shows strong residue coupling signals between the stem regions of E1 and E2 (Fig 5C).
Rosetta modeling has been used before for generating a computational model of E1E2 heterodimer [78], which shares some structural features with our model. During the modeling process, we combine Rosetta modeling with the coevolution analysis, which has been shown to be able to improve the accuracy of Rosetta predictions [79], and the structural information from the 3D EM model. Several mutants have also been made to test the structural model, especially the residues at the E1E2 interface with coevolution signals. The results show that most of the mutants cannot be secreted into media, suggesting that they might be critical for the formation of E1E2 heterodimer.
Several cell surface receptors have been reported to be involved in HCV cell entry, however, the direct binding profiles between E1E2 heterodimer and the receptors are still incomplete. The binding assays based on the expressed soluble E1E2 heterodimer suggest that CD81 interacts with E2, which is consistent with the published results as well as the modeled E1E2 structure. Similarly, two other receptors, SR-B1 and VLDLR, are also mainly interacting with E2. By contrast, ApoE could bind to the E1E2 heterodimer rather than the E2 ectodomain alone, suggesting that E1 might be involved in viral attachment through ApoE. Interestingly, no binding signals are detected for E1E2 heterodimer with OCLN or CLDN1, which have been shown to be functional at late stages of HCV entry. One possibility is that these two tight junction proteins might be involved in the endocytosis process without having direct interactions with E1E2 or other co-factors are required for the binding to E1E2 [69]. The HCV infection inhibition assays also show that the expressed E1E2 heterodimer could block the viral infection effectively. As exposed proteins on HCV surface, the E1E2 heterodimer is the target of immune system and the soluble E1E2 heterodimer obtained here would be a promising target for generating antibodies and facilitate the development of prophylactic vaccines against HCV.
The cDNA sequences encoding E1 and E2 of HCV genotype 1b, Con1 (Accession number AJ238799) and genotype 1a, H77 (Accession number AF009606) were synthesized. In order to co-express E1 (residues 192–354, for both Con1 and H77) and E2 (residues 384–717, for both Con1 and H77) glycoproteins in insect cells, the cDNA fragments of E1 and E2 excluding the transmembrane domains were sub-cloned into a pFastBac Dual vector (Invitrogen) (E1E2), then mouse IgG Fc homodimeric fragment with a Flag and a 6xHis tag at its C-termini was fused to the C-termini of E1 and E2, respectively (E1E2-Fc). Similarly, a de novo designed heterodimeric tag (DHD15), which contains a 6xHis tag and a Flag tag at its C-termini, was fused to the C-termini of E1 and E2, respectively (E1E2-DHD15). In parallel, both E1 and E2 fused with mouse IgG Fc with a 6xHis tag at the C-termini were also individually cloned to the pFastBac vector (E1-Fc and E2-Fc). Similar cDNA fragments, including E1E2, E1E2-Fc, E1E2-DHD15 as well as E1-Fc and E2-Fc were sub-cloned into pMlink co-expression vector [80] for transient expression in HEK293F cells (Invitrogen).
For antibody expression, sequences of IGH526 and HCV1 Fab fragments were obtained from PDB (4N0Y, 4DGV). The cDNA sequences were sub-cloned into pMlink co-expression vector [80] with 6xHis tag fused at the C-terminus of light chain for transient expression in HEK293F cells (Invitrogen).
For receptor binding assays, the ectodomain of human VLDLR (residues 28–797) (Han lab, Xiamen University) with a C-terminal 6xHis tag was cloned into pFastBac vector for expression in insect cells. The full-length human SR-B1 (Sino Biological), OCLN (Sino Biological), and CLDN1 (Sino Biological) fused with C-terminal GFP were also individually cloned into a pTT5 vector for transient expression in HEK293 cells.
For protein expression in insect cells, baculoviruses of the target proteins were generated following the Bac-to-Bac baculovirus expression protocol (Invitrogen), then High-5 cells (Invitrogen) were used for protein expression in ESF921 medium (Expression Systems). The supernatants were collected after 72~96 hours and buffer-exchanged with 50 mM Tris, 150 mM NaCl at pH 8.0 by dialysis, then applied to Ni-NTA affinity column (Qiagen) and Flag M2 affinity column (GeneScript) before loading onto a HiLoad Superdex 200 prep grade column (GE Healthcare) with Tris-NaCl buffer (50 mM Tris, 150 mM NaCl at pH 8.0) for further purification. The purified proteins were loaded onto SDS-PAGE for detection.
For protein expression in mammalian cells, target protein constructs were transiently expressed in HEK293F cells following the manufacturer’s protocol (Invitrogen) using PEI as transfection reagent. The transfected cells were cultured in Gibco FreeStyle-293 medium (Invitrogen) at 37°C for 6 days, then the supernatants were collected for purification using the similar buffers and conditions described above.
A fragment of human CD81 (Genewiz) (residues 122–202) fused with either the small ubiquitin-like modifier (SUMO) or Glutathione S-transferase (GST) were expressed in E. coli BL21(DE3) cells (Novagen) using expression vector pET28a or pGEX6p-1. The soluble SUMO-CD81 or GST-CD81 were purified from the supernatants of cell lysates by Ni-NTA affinity column (Qiagen) followed by SEC chromatography using a HiLoad Superdex 75 prep grade column (GE Healthcare) with Tris-NaCl buffer (50 mM Tris, 150 mM NaCl at pH 8.0). The human apolipoprotein E was purchased from Novoprotein.
Purified Fc- and DHD15-tagged E1E2 or E2-Fc proteins were separated by SDS-PAGE (6% or 8%) and stained with coomassie brilliant blue R-250 (Aladdin). For western blot detection, both supernatants and cell pellets were run on SDS-PAGE (8%) for separation and transferred onto a polyvinylidene difluoride (PVDF) membrance (Invitrogen). The membrane was probed with mouse anti-His tag antibody (1:1000 dilution; Proteintech) or mouse anti-Flag M2 antibody (1:1000 dilution; Sigma) followed by the HRP-conjugated rabbit anti-mouse IgG secondary antibody (Proteintech). After washing three times with the buffer (25mM Tris, 150mM NaCl, pH 7.4, 0.05% Tween-20), the membrane was incubated with Diaminobenzidine (DAB, Sigma) for detection.
10 μl of purified HCV glycoprotein was apply to the glow-discharged EM carbon grids and stained with 0.75% (wt/vol) uranyl formate. Negatively stained EM grids were imaged on a Tecnai T12 microscope (FEI) operated at 120 kV. Images were recorded at a nominal magnification of 67,000x, using a 4k x 4k Eagle CCD camera, corresponding to a pixel size of 1.74 Å per pixel on the specimen. e2boxer.py program in EMAN2 suite was used to pick particles. e2refine2d.py of EMAN2 was used to generate 2D averaging classifications. The initial model was generated using the program e2initialmodel.py, and e2refine_easy.py of EMNA2 was used for refinement and reconstruction. The final resolution was estimated at 27Å based on the gold standard criterion.
The homologs of HCV E1 and E2 were identified from Refseq [81] and Uniref [82] databases and aligned using BLASTP [83]. A pair of E1 and E2 sequences from the same protein sequence were concatenated and the resulting alignment was filtered using HHfilter (-id 95 -cov 75) [84]. The filtered alignment was analyzed using GREMLIN [55] with two sets of parameters: (1) -e 0, -n 100, -w 0.8; and (2) -e 0, -n 100, -w 0.9. The resulting scores from the two CCMpred runs were averaged to obtain the final coevolution score of each pair of residues. All the predicted contacts were ranked by the strength of coevolution and extracted the top L (L is the length of E1 and E2) contacts. Some of the contacting residues from the top predictions were present in the crystal structures of E2 (PDB ids: 4MWF and 4WEB), therefore could be used to evaluate the accuracy of prediction. The prediction accuracy at each rank iL (L is the length of E1 and E2, and i = 0.1, 0.2, …, 1.0) were calculated as the number of correctly predicted contacts in the experimental structure divided by the total number of predicted contacting pairs that are present in the experimental structure. A predicted contacting residue pair is considered to be correct if the shortest distance between the residues in the structure is below 6 angstroms.
Both E1 and E2 ectodomains were partitioned into two parts, one N-terminal region (the first 125 residues for E1 and the first 280 residues for E2) and a highly hydrophobic region. Then RosettaCM [85] protocol was applied to model the N-terminal regions of E1 and E2, and used the partial crystal structures of E1 (PDB: 4UOI) and E2 (PDB: 4MWF and 4WEB) as templates. In addition, the top 0.5L predicted contacts were also used as constraints to aid the modeling. The constraints were set up as previously described [53], so that satisfying a contact was rewarded while missing a contact is still tolerated. The C-terminal regions were de novo modeled with coevolution-derived constraints.
E1 core region (E1c), E1 stem region (E1s), E2 core region (E2c), and E2 stem region (E2s) were modeled separately. The average pairwise TMscore (roughly means the percent of residues that can be aligned within 5 Å) of top 10 (out of thousands) models ranked by Rosetta energy function is a good estimator for modeling accuracy [53]. The accuracy of E1c, E1s, E2c, and E2s are 0.47, 0.41, 0.70, and 0.44 by TMscore, respectively. The models with the lowest Rosetta energy were selected except E1c. The E1c model has the second lowest energy as this model agrees better with the coevolving residues between E1 and E2. Guided by both the EM reconstruction and the coevolution constraints, we manually arranged E1c, E1s, E2c, and E2s together to generate a model of E1E2 heterodimer. The flexible loops in this model were removed first and then rebuilt and refined in the context of the whole structure using Rosetta hybridize protocol [85].
Glutathione-Sepharose 4B beads (GE Healthcare) were mixed with GST-CD81 or GST protein alone (approximately 50 μg) in 100 μl PBS, then the beads were incubated with E2-Fc or E1E2-Fc (about 20 μg protein) in 800 μl PBS on a rotary shaker for 2 hrs at 4 °C. After washing 3 times with PBS, the beads were boiled and centrifuged before loading onto SDS-PAGE for detection with coomassie brilliant blue R250.
The expressed receptors (VLDLR, SUMO-CD81), Fab fragments (IGH526 and HCV1), antibody AR3A and ApoE (Novoprotein) were coated onto 96-well MaxiSorp plates (Nunc) with ∼2 μg protein per well at 4 °C overnight. The plates were blocked with the TBST buffer (25mM Tris, 150mM NaCl, pH 7.4, 0.05% Tween-20) containing 5% (w/v) BSA for 3 hrs. The purified E1E2 (E1E2-DHD15, E1E2-Fc) or E2-Fc were serially diluted and added to each well in a binding buffer (25mM Tris, 150mM NaCl, pH 7.4, 0.05% Tween-20, 1% BSA). Then the plates were incubated at room temperature for 3 hrs (for VLDLR and SUMO-CD81) or at 4 °C overnight (for ApoE). After incubation, plates were washed with the TBST buffer for five times. For E1E2-DHD15 detection, mouse anti-FLAG M2 antibody (Sigma) were added to each well at 1:1000 dilution and incubated at room temperature for 1 hr, followed by washing with TBST for five times. After washing, HRP-conjugated rabbit anti-mouse IgG antibody (Proteintech) was added to each well at 1:1000 dilution and the plates were incubated at room temperature for 1 hr. After washing five times with the TBST buffer, 100 μl of chromogenic substrate (1 μg/mL tetramethylbenzidine, 0.006% H2O2 in 0.05 M phosphate citrate buffer, pH 5.0) was added to each well and incubated for 30 min at 37°C. Then, 50 μl H2SO4 (2.0 M) was added to each well to stop the reactions. The plates were read at 450 nm on a Synergy Neo machine (BioTek Instruments).
HEK293 cells were transfected with pTT5 vectors containing GFP tagged human SR-B1, OCLN or CLDN1. After 48 hrs of transfection, E1E2-DHD15, E1E2-Fc or E2-Fc (~20 μg) were added to the transfected cells in PBS and incubated on a rotary shaker for 2 hrs at room temperature and followed by washing three times with PBS. For E1E2-DHD15 detection, mouse anti-Flag M2 antibody (1:1000 dilution; Sigma) was added and incubated on a rotary shaker for 1 hr, followed by washing three times with PBS. After washing, anti-mouse IgG, F(ab')2 fragment Alexa Fluor 647 Conjugate antibody (1:1000 dilution; Cell Signaling Technology) was added to the cells and incubated on a rotary shaker for 1 hr at room temperature. After washing three times with PBS, cells were analyzed by a LSR Fortessa flow cytometer (Becton Dickinson). Data analysis was performed using FlowJo software (Tree Star).
Either insect or mammalian cell expressed E1E2-DHD15 (~10 μg) were incubated with endo-β-N-acetylglucosaminidase H (Endo H) (50 U) at 37 °C for 1 hr according to the manufacturer’s instruction (Novoprotein). Then the treated proteins were load onto SDS-PAGE for detection. The treated proteins were also purified by SEC and then loaded onto EM grids for negative staining and imaging.
To perform an HCV infection-blocking assay, Huh-7 cells (Stem Cell Bank, Chinese Academy of Sciences) seeded at 1 x 104 cells/well in a 96-well plate were incubated with serially diluted E1E2-Fc (insect cell expressed), E2-Fc (insect cell expressed), E1E2-Fc (HEK293 expressed), E2-Fc (HEK293 expressed) or bovine serum albumin (BSA) (New England BioLabs). After 1 hr incubation at room temperature, about 100 focus-forming units of JFH1 HCV cell culture (HCVcc) were added to the cells, and the protein-virus mixture was removed after 6 hrs of infection. After 3 days of cell culture in complete DMEM, cells were fixed with 2% paraformaldehyde and blocked with buffer (3% BSA, 0.3% Triton X-100, and 10% FBS in PBS), followed by incubation with anti-HCV NS5A MAb (Abmart), Alexa Fluor 488-conjugated donkey anti-mouse IgG and Hoechst dye. The infection efficiency was determined by counting the number of NS5A-positive fluorescent foci under a fluorescence microscope.
|
10.1371/journal.pcbi.1000689 | Unfolding Simulations Reveal the Mechanism of Extreme Unfolding Cooperativity in the Kinetically Stable α-Lytic Protease | Kinetically stable proteins, those whose stability is derived from their slow unfolding kinetics and not thermodynamics, are examples of evolution's best attempts at suppressing unfolding. Especially in highly proteolytic environments, both partially and fully unfolded proteins face potential inactivation through degradation and/or aggregation, hence, slowing unfolding can greatly extend a protein's functional lifetime. The prokaryotic serine protease α-lytic protease (αLP) has done just that, as its unfolding is both very slow (t1/2 ∼1 year) and so cooperative that partial unfolding is negligible, providing a functional advantage over its thermodynamically stable homologs, such as trypsin. Previous studies have identified regions of the domain interface as critical to αLP unfolding, though a complete description of the unfolding pathway is missing. In order to identify the αLP unfolding pathway and the mechanism for its extreme cooperativity, we performed high temperature molecular dynamics unfolding simulations of both αLP and trypsin. The simulated αLP unfolding pathway produces a robust transition state ensemble consistent with prior biochemical experiments and clearly shows that unfolding proceeds through a preferential disruption of the domain interface. Through a novel method of calculating unfolding cooperativity, we show that αLP unfolds extremely cooperatively while trypsin unfolds gradually. Finally, by examining the behavior of both domain interfaces, we propose a model for the differential unfolding cooperativity of αLP and trypsin involving three key regions that differ between the kinetically stable and thermodynamically stable classes of serine proteases.
| Proteins, synthesized as linear polymers of amino acids, fold up into compact native states, burying their hydrophobic amino acids into their interiors. Protein folding minimizes the non-specific interactions that unfolded protein chains can make, which include aggregation with other proteins and degradation by proteases. Unfortunately, even in the native state, proteins can partially unfold, opening up regions of their structure and making these adverse events possible. Some proteins, particularly those in harsh environments full of proteases, have evolved to virtually eliminate partial unfolding, significantly reducing their rate of degradation. This elimination of partial unfolding is termed “cooperative,” because unfolding is an all-or-none process. One class of proteins has diverged into two families, one bacterial and highly cooperative and the other animal and non-cooperative. We have used detailed simulations of unfolding for members of each family, α-lytic protease (bacterial) and trypsin (animal) to understand the unfolding pathways of each and the mechanism for the differential unfolding cooperativity. Our results explain prior biochemical experiments, reproduce the large difference in unfolding cooperativity between the families, and point to the interface between α-lytic protease's two domains as essential to establishing unfolding cooperativity. As seen in an unrelated protein family, generation of a cooperative domain interface may be a common evolutionary response for ensuring the highest protein stability.
| α-lytic protease (αLP), a prokaryotic serine protease of the chymotrypsin family, has evolved an unusual energetic landscape, providing it a functional advantage over its metazoan homologs. Unlike most proteins, αLP's active state is not stabilized by thermodynamics, but by a large kinetic barrier to unfolding, with an unfolding t1/2 of ∼1 year.[1] While thermodynamically stable homologs like trypsin have similar unfolding rates, they are degraded at rates up to 100x faster than αLP under highly proteolytic conditions.[2],[3] In addition, the rates of αLP unfolding and degradation are nearly identical, indicating that partial unfolding leading to proteolysis is negligible. Therefore, αLP's functional advantage is derived from not only its very slow unfolding, which it shares with trypsin, but also its suppression of local unfolding events that would render it protease-accessible. Thus, it appears that the evolution of αLP has generated such extreme cooperativity in unfolding in order to maximize its functional lifetime under harsh conditions. The cost of maximizing resistance to unfolding comes in the form of extremely slow folding (t1/2 ∼1800 years) and the consequent loss of thermodynamic stability of the active state relative to the unfolded state.[1],[3] However, αLP also evolved a large Pro-region folding catalyst, which speeds folding by nine orders of magnitude and is then degraded by the mature protease, decoupling the folding and unfolding landscapes so that unfolding resistance can be maximized.[1],[2],[4]
Given αLP's unusual energetic landscape and its reliance on kinetic stability, much effort has focused on elucidating its unfolding mechanism in detail. Native-state hydrogen-deuterium exchange showed over half of its 194 backbone amides are well-protected from exchange, and 31 have protection factors greater than 109.[2] This extreme rigidity is spread throughout both domains and is indicative of αLP's high unfolding cooperativity. Thermodynamic decomposition of the unfolding energetics into entropic and enthalpic contributions suggested a prominent role for the extensive domain interface in unfolding, with the critical step involving solvation of the domain interface while the individual domains remain relatively intact.[5] Mutational studies on αLP inspired by the acid-resistant homolog NAPase were consistent with this hypothesis. The distribution of salt-bridges in NAPase and αLP differ markedly; replacement of a salt-bridge at αLP's domain interface with an intra-domain salt-bridge (as in NAPase) resulted in significant increases in αLP's resistance to low pH unfolding.[6] A major component of the domain interface, the Domain Bridge (Figure 1), is the only covalent linkage between the two domains. This structure exists only in prokaryotic proteases and varies considerably among αLP and its homologs. The area buried by the domain bridge is inversely correlated with the high-temperature unfolding rate for four kinetically stable proteases, indicating both its relevance and that it is weakened early in unfolding.[7] Another domain interface component is a β-hairpin in the C-terminal domain (CβH), unique to kinetically stable proteases, that forms part of the active site (Figure 1). Substitution of a more stable β-turn was consistent with an unfolding pathway where CβH loses its domain interface contacts early in unfolding.[8] Despite much progress, we still lack a global picture of αLP unfolding, especially at high resolution.
For higher-resolution views of protein folding/unfolding, researchers have often turned to φ-value analysis.[9]–[12] These studies involve large-scale protein engineering experiments which investigate the molecule's folding and unfolding kinetics after making perturbing mutations, normally hydrophobic deletions. By analyzing sufficiently large numbers of perturbations, structure in the transition state ensemble (TSE) can be inferred and a folding/unfolding mechanism can be proposed. Unfortunately, the extremely slow folding and unfolding rates for αLP make large-scale φ-value analysis on αLP impractical. As an alternative, we decided to investigate the αLP unfolding pathway computationally in order to explain previous experiments and guide new ones.
High-temperature molecular dynamics (MD) unfolding simulations offer the highest structural and temporal resolution for studying protein unfolding, but their results must be validated experimentally. Since unfolding rates for proteins are typically very slow under physiological conditions (ranging from minutes to a year for proteins such as αLP), very high temperatures (450–500 K) are required to accelerate the unfolding into the ns range required for computational analysis. As a consequence, initially there was significant concern as to the relevance of the high temperature TSEs to real proteins under physiological conditions. Daggett and co-workers have been pioneers in this field, using Chymotrypsin Inhibitor 2 (CI2) as a model system and have shown that the simulated unfolding calculations agree remarkably well with experimental φ-values and were even able to predict faster folding mutants.[13]–[16] Further work on other proteins by multiple groups has established MD unfolding simulations as a useful tool in examining protein unfolding at atomic resolution while correlating well with experiments.[17]–[20]
A critical step in analyzing unfolding simulations is accurately pinpointing the TSE from the multitude of conformations generated. Because the TSE is experimentally accessible through a molecule's folding and unfolding kinetics, its identification computationally can be used for both explanatory and predictive purposes. Various methods for identifying the TSE have been used in the past, breaking down into conformational clustering and landscape methods.[13], [15], [17], [19], [21]–[23] Conformational clustering relies on all-versus-all comparisons of conformations, often by Cα RMSD, while landscapes separating native from unfolded structures can be generated using properties of the conformations, such as the fraction of native contacts or secondary structure.
Here, we report the results of multiple MD simulations carried out at high temperature in order to probe the mechanism of αLP's extremely cooperative unfolding. Due to the robustness and cooperativity of αLP unfolding, the same TSE is obtained using either conformational clustering or landscape methods. The simulated unfolding pathway for αLP matches well with previously described experiments and provides atomic resolution to previous models for αLP unfolding which highlight the role of the domain interface. In addition, we have performed similar simulations on trypsin with the goal of understanding the observed experimental differences in unfolding cooperativity. Through a novel method for calculating cooperativity in MD simulations, we show αLP unfolds significantly more cooperatively than trypsin, mirroring the experimental results. Finally, by analyzing the domain interfaces of both proteins during unfolding, we propose a mechanism for how this differential cooperativity is achieved.
Simulations were performed with NAMD[24] using the CHARMM22[25] forcefield and TIP3P explicit water (full details in Methods). To test for proper behavior in our simulations, a 298K MD simulation of αLP was performed for 12.1 ns. αLP was quite stable, averaging 0.84 Å Cα RMSD to the crystal structure[26] over the course of the simulation and 0.87 Å Cα RMSD over the last 1 ns, with a maximum of 1.32 Å (Figure 2A). A previous 1 ns MD simulation of αLP at 300K using a different force field and simulation conditions also found little deviation from the crystal structure (average 0.83 Å Cα RMSD).[27] A long loop comprising residues 218–225 (Figure 1, middle right, orange) and several residues at turns contribute most of the differences and have higher than average B-factors in the crystal structure.[26] At 298K, there is little additional exposure of non-polar solvent accessible surface area (NPSASA), with an average increase of 5.5% in exposure (Figure 2C). It should be noted that the rigidity of αLP as seen by 298K simulation is considerably greater than what is observed for other proteins,[14],[18],[19] consistent with the very low crystallographic B-factors[26] and high hydrogen exchange protection factors[2] seen previously.
Five independent 8.1 ns MD simulations at 500K were conducted to determine the unfolding pathway of αLP, with the Cα RMSD of each plotted in Figure 2A. Visual inspection of the trajectories and the high Cα RMSDs attained indicated that αLP had unfolded in each simulation. By contrast, simulations at 450K showed little unfolding at similar timescales making them impractical for analysis (data not shown). Each trajectory shows a generally increasing Cα RMSD throughout the simulation, though there is significant variation in the rates of increase, periods of no change or decrease in Cα RMSD, and final Cα RMSD, as expected for independent simulations. Because relatively high RMSDs were reached in the first 4 ns of the simulations, we hypothesized that the major unfolding transition occurred in that timeframe (Figure 2B).
To confirm that unfolding had occurred, we examined molecular properties orthogonal to Cα RMSD early in the simulations. These properties, non-polar solvent accessible surface area (NPSASA) and a new metric termed Average Local Fluctuation (ALF), can distinguish native from non-native conformations without directly comparing them to the crystal structure. First, non-polar amino acid side-chains, normally buried in a protein's interior, become exposed upon unfolding, increasing NPSASA. The NPSASA for the first 4 ns of 298K1 (for comparison), 500K1, and 500K3 is plotted in Figure 2C. 500K1 and 500K3 were chosen for clarity due to a large difference in unfolding time. Both exhibit relatively small increases to ∼5000 Å2 within the first 0.3 ns, consistent with thermal equilibration. NPSASA then increases very slowly, unlike Cα RMSD, until it rapidly increases at 1.3 and 1.8 ns for 500K1 and 500K3, respectively. These sharp rises are followed by another slowly increasing phase that is highly variable for the rest of the simulations.
The second property, ALF, relies on the notion, derived from funnel energy landscape models of protein folding/unfolding, that molecules in the unfolded ensemble can explore many more conformations than those in the native ensemble.[28] For αLP, where the unfolding barrier has been shown experimentally to be extremely high, cooperative, and entropic in nature, it is certain that conformational space on the folded side of the TSE is quite restricted relative to the unfolded side.[2],[5] If unfolding simulations capture this ensemble behavior, there would be bottlenecks or barriers in the unfolding landscape. ALF was created to assay for these barriers, as it measures the rate of conformational change throughout a simulation (details in Methods). ALF for the first 4 ns of 298K1 (for comparison), 500K1, and 500K3 is plotted in Figure 2D. In the first 0.3 ns of both simulations, ALF increases slightly from 1.0 to 1.3 Å due to thermal equilibration. It remains relatively flat until rapid increases beginning at 1.3 and 1.8 ns for 500K1 and 500K3, respectively, resulting in a permanently higher ALF. In 500K3, ALF increases less sharply relative to 500K1, rapidly decreasing and then recovering in the middle of its rise ∼2.0 ns, which has implications for identifying its TSE (see below). The large and permanent increases in conformational flexibility measured by ALF and their coincidence with similar increases in NPSASA are indicative of seeing true unfolding transitions.
Structurally, the early stages of αLP's unfolding pathway are quite consistent among the five unfolding simulations, though the simulations tend to diverge once the molecule becomes much less native-like. As we will show below, these early events constitute the major unfolding transition and are the primary focus of this work. First, we will describe the pathway in detail for 500K1, with several important conformations shown in Figure 3, and then note any important differences in other simulations. A movie of the full 500K1 unfolding pathway is shown in Video S1. For the first several hundred picoseconds, αLP thermally equilibrates and reaches ∼2 Å Cα RMSD to the crystal structure, with small surface loops the major source of this small deviation. At 0.7 ns, a large loop comprising residues 218–225 unique to αLP becomes more mobile, though its flexibility is somewhat limited by a disulfide bond between residues C189 and C220A. All residue numbering is based on homology to chymotrypsin, as in the PDB files. Because this loop is not conserved in kinetically stable proteases and is relatively mobile at 298K, we feel its overall impact on the unfolding pathway is small. At 1.0 ns, the Domain Bridge, a β-hairpin connecting the two domains of αLP, becomes more mobile but remains intact (Figures 1 and 3). Between 1.2 and 1.4 ns, αLP begins to unfold much more significantly, though the distortions are confined to four main structural areas: the N-terminal strand β1, the Domain Bridge, a region near the active site comprising the CβH and a cis-proline-containing turn (residues 91–102, CPT), and the 218–225 loop (Figures 1 and 3). β1 pulls away from the body of the protein and becomes highly flexible. The Domain Bridge breaks tertiary contacts with nearby residues and its two strands separate. Contacts between the CPT and the CβH break as the two pull away from each other, and the CβH strands separate. The 218–225 loop remains highly flexible, causing residues 215–217, which form part of the substrate binding groove, to separate from the β-barrel and push the CβH away from the body of the protein. These regions continue to unfold, accelerating the unfolding of nearby structure, though several regions remain relatively well-structured at 1.64 ns, including the β-sheets β4-β7-β6 and β14-β15-β16, and the C-terminal α-helix (Figure 3). The C-terminal β-barrel unfolds and further weakens the domain interface, with very few native-like interactions bridging the two domains at 2.4 ns (Figure 3). By 4.2 ns, little residual structure remains, as the Cα RMSD is 11.4 Å, though the molecule does continue to unfold, reaching a Cα RMSD over 16 Å within 8 ns (Figure 3). The presence of three disulfide bonds most likely prevents more extreme unfolding.
Early on, each of the unfolding simulations follows a similar trajectory to that of 500K1 although with variability in the timing (Figure 2), beyond this, some other differences do exist. In 500K4, β5 unfolds much earlier relative to the other simulations, separating from β2 and β6 and partially exposing the interior of the N-terminal domain to solvent. The turn connecting β5 to the more stable β6 (Figure 1, upper left, light blue) is quite flexible in all five unfolding simulations and has some of the highest B-factors in the crystal structure, which may explain part of this behavior.[7],[26] In 500K3, the Domain Bridge does break some tertiary contacts with surrounding regions early in unfolding, but its two strands separate relatively late. The N-terminal β1 does not completely separate from the body of the protein in 500K2 and 500K3 early on, as it does in the other three simulations, but its contacts are somewhat disrupted in both. Other differences at early time points appear to be relatively minor and are to be expected given five independent high temperature unfolding simulations.
Because computational studies of protein unfolding are severely restricted in the number of molecules that can be simulated, they must use the vast amount of information present in each simulation in order to identify the TSE. As in other types of single-molecule experiments, there will be significant variation within the properties of the ensembles, such as time to unfold. Unlike experimental studies, where there is often a single reporter of the molecule's conformation, such as tryptophan fluorescence, MD simulations provide every conformation sampled, an enormous amount of data. However, there is no a priori way to say whether a particular three-dimensional structure is “folded” or “unfolded.” The challenge then is to derive properties from the conformations, either those directly computable from each structure or those that rely on comparing structures to each other, that can be used to clearly separate the folded from the unfolded conformations.
Previous studies investigating the nature of a protein's TSE by unfolding simulations have often determined TSEs from individual simulations and combined them into an overall TSE.[15],[18],[29] These approaches depend on the assumption that the TSE is a small region of conformational space at the edge of the native basin, hence identifying them requires methods that clearly separate native from non-native conformations. One method that has had considerable success is a conformational clustering procedure pioneered by Li and Daggett.[13],[14] A pairwise Cα RMSD matrix is generated for all trajectory conformations and then projected down into two or three dimensions using multi-dimensional scaling. Visual clustering then separates the native conformations from the non-native, placing the TSE at the exit of the native cluster. While the method does require a significant level of subjective judgment, the Daggett group has had good success correlating results of their unfolding simulations to protein engineering studies of the same proteins. Conformational clustering was performed for each of the unfolding simulations here, with the three-dimensional projection of the 500K1 trajectory shown in Figure 4. Individual conformations extracted every 10 ps are shown as spheres and are connected chronologically by sticks; the color goes from blue to red as the simulation progresses. The first 1.41 ns of 500K1 is tightly clustered around the native state (lower left) and then rapidly moves away from the native state, forming much less dense clusters as it progresses through the simulation. Similar behavior is seen for the other unfolding simulations, allowing them to be effectively clustered (Table 1). However, it is much more difficult to identify a common TSE by conformationally clustering all five unfolding simulations simultaneously; hence we sought a method that would allow a common TSE to be generated, testing the conformationally clustered TSE.
Although the ALF metric captures some of the significant changes during unfolding, it should be possible to gain a better picture of the unfolding process across all of the unfolding simulations, by not looking as a function of time, but rather through changing properties. Many structural properties, such as secondary structure content and fraction of native contacts, have been used to cluster trajectories or create energy landscapes both in unfolding simulations and in equilibrium simulations utilizing umbrella sampling.[15],[19],[22],[23],[30],[31] Using common protein folding/unfolding metrics (here, the number of native contacts and NPSASA) as order parameters, we have computed a single two-dimensional unfolding landscape that integrates data from all the simulations despite their individual differences in timing (Figure 5A). Histograms of the individual metrics are shown at the top and right of the landscape. The landscape shows three well-populated basins (dark blue), one native-like (upper left) and two progressively less native (middle and lower right). There is a bottleneck in the landscape, shown enlarged in the inset and centered around 450 native contacts and 5900 Å2 NPSASA, that separates the native from non-native basins. Also shown in the inset is a trace of the 500K1 simulation, at 10 ps intervals, for clarity (the landscape was constructed using conformations at 1 ps intervals, a total of 40500 conformations). Significantly, all simulations cross this bottleneck only once, implying a shared barrier to unfolding with these order parameters. The actual crossing transition occurs at different times in the different simulations, for example occurring between 1.41 and 1.42 ns for 500K1 (Table 1). We propose that this barrier is the location of the αLP TSE in these simulations and have generated a TSE from the structures making up the barrier (Table 1).
In reality, the αLP unfolding landscape is highly multi-dimensional and is only approximated by NPSASA and native contacts, which are clearly highly correlated. In order to utilize more of those dimensions, ten parameters were measured for each conformation (details and full listing in Methods). Principal components analysis (PCA) was used to eliminate the inherent correlations in the parameters and allow visualization in less than ten dimensions. The first two principal components explain 90% of the variance in the parameters and were used to generate a landscape as above (Figure 5B). Again, the region comprising native-like conformations is well-separated from the non-native region by a sparsely populated barrier centered around −2.7 on PC1 and 0.0 on PC2. Crossing times for all of the simulations are within 30 ps of the crossing times in the NPSASA/native contacts landscape, and, as above, we have generated a TSE from the PCA landscapes (Table 1). The first principal component, which contains relatively equal weightings from all ten parameters, is mostly a function of each conformation's nativeness (Table S1). There is little variation in the second principal component in the native-like region, and the simulation trajectories begin to diverge more significantly upon reaching the unfolding barrier. The second principal component is dominated by the size of the molecule and backbone exposure to solvent, as the three largest components are non-native mainchain hydrogen bonds, polar SASA, and radius of gyration (Table S1).
It is important to note that the landscapes in Figure 5 are not free energy landscapes[23], as the simulations analyzed here are non-equilibrium simulations, but represent the degree of sampling of the relevant structural properties. While interpretation of these landscapes is not as straightforward as that for free energy landscapes, we believe that they accurately identify the TSE. Unfolding should proceed rapidly once the TSE is passed in an individual simulation, as seen by the ALF metric (Figure 2D), which will limit sampling of the TSE region. Here, we have performed five independent simulations, observing a shared region in parameter space that is under-sampled and on pathway to the unfolded state. Importantly, the simulations only cross this region once, as expected given the strongly unfolding conditions. This coincidence of under-sampled parameter space for the combination of five simulations almost exactly coincides with the native exit cluster based on pair-wise structural comparisons for four of the five simulations, with a small error for 500K3. Agreement between such quite different methods is not a given, as has been observed in simulations of spectrin R17.[19] It is likely that the remarkable agreement seen here between conformational clustering and the landscape methods is due to the high cooperativity of αLP unfolding, which is experimentally observed[2]. Finally, we believe the PCA-landscape-derived TSE is the most accurate one, as its clustering is the least subjective, which may be an issue with 500K3 conformational clustering, the crossings of its observed barriers are unambiguous, and its generation from all five simulations adds additional evidence to its relevance.
For the remainder of this work, the αLP TSE is derived from the PCA landscape, generated by taking the conformations spanning the barrier crossing for each of the individual simulations (10 ps, conformations saved at 1 ps intervals) and combining them, yielding a TSE with 50 conformations. Some general properties of the TSE are listed in Table S2. Due to heterogeneity in large portions of the molecule, it is difficult to visualize the entire set of conformations (representative members are shown in Figure S1). As one way of visualizing the TSE, all TSE conformations and the crystal structure were superimposed using the structural superposition program THESEUS and the average deviation from the crystal structure at each Cα over all conformations was computed.[32],[33] These deviations were then mapped onto the crystal structure by color and thickness of the tube used to represent the backbone, as seen in Figure 6. Several observations can be made from this representation. First, significant deviations from the crystal structure are confined to several regions, notably those mentioned above. Much of the molecule is quite native-like, including the sheet β2-β3-β4-β7-β6 in the N-terminal domain and most of the β-barrel in the C-terminal domain. Second, as evident in stereo, the “front” face of αLP as depicted deviates far more from native than the “back” face. The “front” face contains the active site and these deviations would severely disrupt enzymatic activity. In addition, preliminary native state hydrogen exchange experiments found that denaturing agents had a more significant effect on the “front” face of αLP.[34] Third, with the exception of the 218–225 loop, which is not conserved, unfolding of the regions identified in each of the unfolding simulations, β1, the Domain Bridge, and the active site hairpins, would disrupt the domain interface and expose much of it to solvent.
The entropic nature of the αLP unfolding barrier previously led us to hypothesize a solvated domain interface at the TSE, and investigations of the pH-dependence of unfolding has lent credence to that model.[5],[6] The TSE model presented here (Figure 6) is consistent with the individual domains remaining well-folded throughout the unfolding transition, but is seemingly at odds with the hypothesis that the domains open up in the TSE. To better investigate the domain interface's response to unfolding, we calculated the number of residue-residue intra-domain and inter-domain contacts present in each simulated conformation and normalized them by the corresponding number present in the crystal structure (shown for 500K1 in Figure 7A). Note that at the TSE (1.4 ns), the drop in inter-domain contacts is much more steep and continues much longer than the relatively shallow drop in intra-domain contacts. This effect is exaggerated if only contacts present in the crystal structure are considered. Gray and red curves represent these native intra-domain and inter-domain contacts, respectively. As before, native inter-domain contacts are being lost much more quickly at the TSE. At 2.0 ns, just 0.6 ns after the TSE, only 15% of native inter-domain contacts remain while 50% of native intra-domain contacts are present. This general pattern holds for the other unfolding simulations, providing additional evidence that a key step in αLP unfolding is the opening of the domain interface.
The Domain Bridge is an integral part of the αLP domain interface and has been experimentally implicated as a determinant of the unfolding rate.[7] To quantify its role in unfolding, we have calculated the normalized number of native contacts it makes, as above, though using atom-atom contacts due to the relatively small number of residues. Plotted in Figure 7B is are the fraction of native contacts between two residues both in the Domain Bridge (DB-DB, black) and between one residue in the Domain Bridge and any other residue (DB-O, green) for the first 3 ns of 500K1. DB-DB contacts are quite stable until the molecules begins to unfold significantly at 1.2 ns, reaching about 60% of native, and then losing all native contacts right at the TSE at 1.41 ns. DB-O contacts are lost more gradually prior to the TSE than DB-DB contacts, but they experience the same steep loss at the TSE. With the exception of 500K3 as noted above, the other unfolding simulations exhibit similar behavior. The high unfolding cooperativity of the Domain Bridge and its coincidence with the TSE observed here is consistent with the previous experimental studies.
A critical feature for αLP's kinetic stability is its extremely high unfolding cooperativity. Previous work has shown that while αLP and trypsin, a thermodynamically stable homolog, have similar unfolding rates, αLP unfolding is much more cooperative as measured by proteolysis, providing it a functional advantage in highly proteolytic environments.[2],[3] Because determining the origins of this remarkable difference is crucial for understanding the molecular basis for kinetic stability, we sought to compare the behaviors of αLP and trypsin as revealed by unfolding simulations. Four 10.1 ns unfolding simulations at 500K were performed for trypsin. Although a thorough discussion of the details of the trypsin TSE and unfolding pathway will be presented elsewhere, the general behavior of these simulations is reported in the Supplementary material (Figures S2 and S3, Table S3).
To quantitatively compare unfolding cooperativity, we developed a new metric defined by how many conformations were similar (based on a Cα RMSD threshold) to the ith conformation within the n total simulation conformations. The cooperativity graph for a perfectly cooperative unfolding transition would be high and flat for the beginning of the simulation, drop steeply at the TSE, and then be much lower for the duration of the simulation. Specifically, it would have a value of j from 1 to j, where j is the TSE conformation, and drop to a value k≪j after the TSE. Less cooperative transitions would feature gradually increasing and/or decreasing values prior to the TSE and less steep drops after the TSE. Cooperativity for αLP (500K1) and trypsin (500K2T) are shown in Figure 8A and 8B, respectively. The cooperativity profile for αLP is very similar to that of the hypothetical perfectly cooperative unfolding transition. Before the TSE at 1410 ps, the value is near 1400 and relatively flat. It drops sharply right at the TSE, and then is much lower for the duration of the simulation. Trypsin, on the other hand, unfolds much less cooperatively. Its increasing profile from 0 to 900 ps represents gradual unfolding, because structures that have partially unfolded are similar to both the native structure and to more unfolded conformations. It has no clear steep drop from the native state as αLP does, only a gradual and very noisy decline. Its values post-TSE are much higher than those observed for αLP, which suggests a more rugged, gradual unfolding process. Cooperativity plots for the other simulations show the same general trends and the behavior is qualitatively similar with different choices of Cα RMSD thresholds. We believe this work is the first example of both measuring cooperativity in simulated unfolding and comparing it across two proteins where that difference has functional relevance.
A major motivation for this study was providing atomic resolution to previous biochemical experiments on αLP unfolding, but first those lower resolution results must be reproduced. A comprehensive analysis of experimental data on protein unfolding barriers revealed a stark difference between those of αLP and thermodynamically stable proteins: the αLP unfolding barrier is significantly more entropic, suggesting the αLP TSE is considerably more native-like than those for thermodynamically stable proteins.[5] In addition, m-value analysis of unfolding found the fraction of SASA buried at the αLP TSE was computed to be 80%, also highly-native like.[35] The simulation-derived TSE reported here is quite similar to the native structure, with an average Cα RMSD of 4.4±0.4 Å and with 38% of Cα atoms being less than 2.0 Å from native. The average fractional SASA at the TSE is 82±2%, slightly higher but quite consistent with the value derived from experiment. One possibility for the slight deviation is that the elevated temperature in the simulations shifts the TSE somewhat towards the native, a modest Hammond effect that was seen for CI2.[36],[37]
One of the challenges in this study is to identify a robust TSE from the unfolding trajectories at high temperatures. The most widely accepted definition of the TSE is the pfold definition, which defines the TSE as the ensemble of states in which at physiological temperatures, half the states refold and the other half unfolds. In recent studies of two very small domains, CI2 [38] and the engrailed homeodomain [39], it was demonstrated that conformations from the TSE identified by multi-dimensional scaling in high temperature unfolding simulations satisfies the pfold definition at physiological temperatures. Indeed, at temperatures near the transition temperature, these conformations are seen to oscillate between the folded and unfolded states [40]. However the pfold verification required 36000 times longer simulation time than the original unfolding simulation, a calculation that would not be feasible for αLP, a much larger protein. Once practical, it would be useful to carry out such calculations. Nevertheless, based on the studies above, we believe that the TSE of αLP identified here represents a useful estimate of the physiological TSE. One reason is that in the PCA landscape of αLP (Figure 5), there is a particularly tight bottleneck that separates the folded and unfolded states. As significant unfolding trajectories must pass through this bottleneck, the TSE must be found near the bottleneck. Extrapolating this behavior to physiological temperatures, we expect this bottleneck to become even more pronounced, thereby localizing the TSE in conformation space. Moreover, there is strong connection between inferences developed from the calculated TSEs and a wide range of experimental observations with αLP, providing further confidence that the calculated TSEs provide relevant insight into the TSEs at physiological temperatures.
Previous experiments have shown a large role for the domain interface in αLP unfolding. The thermodynamic analysis referenced above suggested a possible model for the TSE: a “cracked egg” where the two β-barrel domains are largely intact but the extensive domain interface between them is disrupted.[5],[6] Relocation of salt bridges spanning the domain interface significantly decreased αLP's sensitivity to low pH unfolding, consistent with the “cracked egg” model,[6] (P. Erciyas, private communication). The simulations presented here confirm the disruption of the domain interface at the TSE, provide atomic detail as to how it happens, and extend these insights to two other critical structural regions: β1 and the CPT and CβH.
The Domain Bridge, the covalent linkage between αLP's two domains, has been shown to modulate the unfolding rate.[7] The simulations support this; they reveal that many of its native contacts are lost at the TSE, including separation of its strands, allowing it to make non-native contacts. The domain bridge makes several contacts with the N-terminal β-strand β1, which is also significantly disrupted at the TSE. Our results indicate a probable coupling of the unfolding of the Domain Bridge and β1, though the coupling is less evident in 500K2 and 500K3. When full-length Pro-αLP is synthesized, the C-terminus of the Pro region is covalently connected to the protease's N-terminus. As the protease domain folds, it gains proteolytic activity, cleaving the Pro-αLP junction that is positioned across the active site.[41],[42] The active site is 20 Å away from the location of the N-terminus in the native state and hence folding requires a significant rearrangement of the N-terminal strand. The flexibility of the N-terminus at the TSE in our simulations is consistent with its requirements during Pro-assisted folding. The last region at the domain interface disrupted at the TSE forms part of the active site.
Previous studies on αLP have also implicated the CβH as important to the folding/unfolding landscape. Mutations in the hairpin affected both the unfolding rate and the Pro-catalyzed folding rate.[8],[43] The Pro-αLP complex structure revealed that this hairpin forms a larger five-stranded β-sheet with Pro; mutants disrupting the interface there significantly weaken Pro's foldase activity.[42] The hairpin forms several side-chain contacts and two main-chain hydrogen bonds with the CPT in the native state; CPT residues F94, which forms the bulk of the contacts with the CβH, and cis-P95 are both completely conserved in kinetically stable proteases. The amides in these hydrogen bonds have relatively weak protection factors compared to the rest of the protein, consistent with them being broken at the TSE.[2] These contacts are also relatively long-range in sequence space, requiring that the molecule must give up significant conformational entropy in bringing them together, again arguing that they are broken early in unfolding. In our simulations, once the contacts between the two structures are broken, CβH pulls away from the body of the protein and its strands separate; the presence of the Pro region would keep the hairpin in a position ready to make contacts with the cis-proline turn, stabilizing the TSE. By understanding the αLP unfolding pathway and TSE in atomic detail, we can begin to explore how the Pro region stabilizes the TSE and accelerates folding 109-fold.
The three regions of the domain interface disrupted at the TSE have something else in common: they are only found in the kinetically stable proteases and not the thermodynamically stable family members, such as trypsin. αLP and trypsin are good structural homologs; 120 (of 198) αLP's Cα's have an equivalent position in trypsin, with 99 of them within 2.0 Å of their trypsin equivalent.[44] It seems an unlikely coincidence that the regions of αLP that unfold at the TSE happen to be in the 1/3 of the protein that is not homologous to trypsin. In fact, for both the αLP and trypsin simulations, the structurally conserved regions are much more native-like at the TSE and beyond than are the non-conserved regions. Significantly, large parts of the αLP domain interface are made up of the non-conserved regions, likely resulting in the dramatic differences observed between folding of individual αLP and trypsin domains. For both chymotrypsin and trypsin, the two domains fold independently and upon mixing will form the active enzyme.[45],[46] By contrast, active αLP cannot be reconstituted from unfolded individual domains even in the presence of the Pro region.[47] αLP's cooperativity in folding echoes that of the unfolding reaction and likely involves the Domain Bridge and other regions of the domain interface which are distinct from its metazoan homologs.
By closely examining the differences between the domain interfaces of αLP and trypsin, we can begin to discover the mechanism of αLP's unfolding cooperativity. The buried residues (less than 5% exposed in the crystal structure) of the αLP and trypsin domain interfaces are shown in space-filling spheres in Figure 9A and 9B, respectively. Residues colored light red are exposed to solvent at the TSE, while residues still buried at the TSE are further subdivided into light green and light blue residues, which at 600 ps post-TSE are either exposed to solvent or still buried, respectively. For both αLP and trypsin, residues near the “top” and “bottom” of the molecules are more likely to be exposed at the TSE, while the “middle,” which contains the conserved active site, has fewer red residues. Clearly, more of trypsin's buried domain interface residues are exposed to solvent at the TSE than for αLP. In addition, the αLP core is much more blue than trypsin, as this core is much more resistant to solvation even post-unfolding than its metazoan counterpart.
As alluded to previously, much of the domain interface is not conserved between the two families of proteases. Figures 9C and 9D focus on the conserved interface, and correspond to 9A and 9B, respectively, after removing all residues that are not common to both domain interfaces (this implies position and sequence conservation). Here, the difference between αLP and trypsin is striking; over half of the positions in trypsin are red (solvent exposed), while residues in αLP are generally exposed to solvent much later, over half of them blue. An area near the active site (see Figure 1) comprising αLP residues D102, L180, and S214 (all green) and their trypsin equivalents D102, M180, and S214 (all light red) is particularly interesting. Again, this region is composed of the CβH unique to the kinetically stable proteases, and though it unfolds at the TSE, only once it completely unfolds does it begin to exposed the conserved core to solvent. This is not the case in trypsin, where the different architecture, composed of loops, allows relatively small unfolding events to expose the buried interface. Moreover, unlike αLP, these unfolding events in trypsin must be uncoupled from one another resulting in much greater structural variability in the TSE or equivalently, multiple, parallel approaches to the TSE. Thus one can think of trypsin as slowly diffusing over a broad unfolding barrier whereas αLP's behavior is much more concerted and the passage over its barrier is more tightly constrained.
Examining the differences between the full-domain interface and the conserved domain interface figures then highlights the non-conserved regions. At the αLP Domain Bridge, its unfolding exposes relatively little of the domain interface at the TSE, while the much larger equivalent area in trypsin is quite solvated. An important difference between the two proteases is that the αLP Domain Bridge is a compact, cooperative substructure, a simple β-hairpin. In trypsin, the domain interface is formed by two long and relatively floppy loops, which are inherently less cooperative than the domain bridge. Many of the non-conserved domain interface residues in αLP are also in secondary structure or tightly constrained turns near the Domain Bridge or active site (i.e. G18 and G19 connect β1 to β2 and interact with the Domain Bridge, V120B and V121 form the base of the Domain Bridge, V177 is in the CβH), while in trypsin these areas are formed with much less constrained loops. The differences seen here in the Domain Bridge and active site regions provide evidence that extreme unfolding cooperativity is generated by using highly cooperative substructures to protect the rest of the domain interface from solvent.
Intriguingly, increased protease resistance mediated through high inter-domain cooperativity has been observed in an unrelated system.[48] A screen of the Escherichia coli proteome for protease resistance found 40 proteins, one of which was the glycolytic enzyme phosphoglycerate kinase (PGK).[49] Young et al. found that while the E. coli and Saccharomyces cerevisiae enzymes had similar stabilities, the yeast PGK unfolded and was degraded much faster than the E. coli PGK.[48] The difference was attributed to the domain interface; the separated domains of yeast PGK fold independently and are quite stable, unlike the E. coli PGK domains, analogous to the difference between prokaryotic αLP and eukaryotic trypsin.
Recent work from Hills and Brooks on flavodoxin fold proteins using Gō models offers relevant insight into dynamics and cooperativity.[31],[50] First, they showed that contact density, the ratio of native contacts to the number of residues within a subset of the protein, was an accurate predictor of the nucleating subdomain in flavodoxin folds. Intriguingly, increased contact density in a specific region in Spo0F was consistent with local rigidification relative to other flavodoxin folds; this extra contact density induced higher topological frustration during Spo0F folding simulations. Using the native contacts definition in the Methods, αLP has a higher overall contact density than trypsin (3.89 and 3.64); this difference is magnified when considering the buried domain interface residues (Figure 9) (6.23 and 5.62 for all, 6.12 and 5.37 for the non-conserved residues). The insights from contact density are again consistent with the previous experimental and computational results presented here. αLP is more rigid than trypsin and unfolds more cooperatively, stemming from the non-conserved regions of its domain interface, but its folding landscape is extremely frustrated, resulting in folding kinetics on the order of millennia instead of seconds.
The costs of evolving extreme unfolding cooperativity are high; for αLP, the bacterium must synthesize a 166 residue protein to catalyze αLP folding after which it is immediately degraded. αLP's extremely slow folding is a consequence of the large energy gap between its unfolded/molten globule states and the TSE.[1] One likely contributor that has been previously noted is its high glycine content, as glycines in formed structures lose much conformational entropy relative to unstructured glycines.[51] These glycines, which make up 18% of kinetically stable proteases, are used to form tight turns and tight packing in areas where even an alanine would be sterically hindered.[1],[26] Like most proteins, the metazoan proteases have much lower glycine content (about 9%) and have many correspondingly longer loops than the prokaryotic proteases. These loops, like those in the domain interface of trypsin, are likely the reason for trypsin's lack of cooperative unfolding.
Finally, the idea that a protein's folding transition state is determined by its native structure, as shown through studies of Contact Order and folding rates, poses interesting questions for this class of proteases.[52],[53] While trypsin fits well in the Contact Order plot, αLP is an extreme outlier, perhaps not surprising given its remarkably slow folding.[5] The two proteins have the same fold and would be expected to have similar TSEs. Here, we have identified the TSEs for both, and remarkably, those TSEs both contain much of the conserved core of the fold. However, the regions where the two proteases differ are critical parts of the TSE structures. While the general structure of the TSE may be mostly determined by the native structure, the details, such as highly cooperative units making up the domain interface in αLP and not trypsin, can provide large functional advantages depending on the environment of the particular protein.
1SSX and 5PTP PDBs were used for αLP and trypsin, respectively. All non-protein and hydrogen atoms were removed and hydrogens were added back with XPLOR.[54] For residues with multiple conformations, the “A” conformation was used. Protein molecules were placed in cubic boxes with a minimum of 12 Å distance to the edge and solvated with TIP3P explicit water and chloride counter-ions using Packmol,[55] where the approximate density was determined by the density of liquid water at the corresponding temperatures.[56] The number of atoms for 298K αLP, 500K αLP, 298K trypsin, and 500K trypsin were 32760, 28005, 33223, 28468, respectively. All simulations were performed using NAMD 2.5 with the CHARMM22 forcefield.[24],[25] Simulations were carried out with periodic boundary conditions, a 12 Å cutoff for non-bonded interactions, and Particle Mesh Ewald for long-range electrostatics. A timestep of 1 fs was used and snapshots were saved every 1 ps. Each system was equilibrated using the following protocol. The protein was fully constrained and the solvent was minimized for 500 steps using a conjugate gradient algorithm. The solvent was equilibrated for 100 ps under NPT conditions (298K and 1.01325 bar or 500K and 27 bar) using Berendsen coupling for both pressure (100 fs relaxation time) and temperature (2.0 ps coupling constant).[57] The solvent was then fully constrained and the protein was minimized for 50 steps. The entire system was then minimized for 50 steps. Finally, the system was equilibrated for 100 ps under the same NPT conditions. Multiple independent simulations were generated by starting the whole-system equilibration using different random number seeds for each. After equilibration, production simulations were carried out in the NVE ensemble, with the box size fixed at its final size from the equilibration. One 298K αLP (12.1 ns), five 500K αLP (8.1 ns each), one 298K trypsin (3.6 ns), and four 500K trypsin (10.1 ns each) simulations were performed for 96.6 ns total simulation time.
In several analyses presented here (Conformational Clustering, ALF, and Cooperativity), Cα RMSDs were calculated with two fits to the target structure in order to lessen the impact of a small number of poorly aligning residues. Structures were aligned using all Cα atoms and the mean and standard deviation of the deviations were calculated. Cα atoms whose deviations were greater than two standard deviations above the mean were discarded for the second fit and calculation of Cα RMSD. This fitting procedure eliminated an average of 5% of the Cα atoms.
For all overlapping 90 ps windows in a simulation, all pairwise two-fit Cα RMSDs were calculated for the 10 snapshots (10 ps intervals), resulting in 45 two-fit RMSDs at 801 windows for an 8.1 ns simulation. These RMSDs were averaged to give the ALF at the midpoint of each window. ALF therefore measures the extent of short-timescale (90 ps) fluctuations throughout the simulation, as it is the mean RMSD between any two snapshots within a short time window.
For each simulation, pairwise two-fit RMSDs were calculated at 10 ps intervals, forming a symmetric N × N matrix, with N = 810 for αLP and N = 1010 for trypsin unfolding simulations. Multi-dimensional scaling, as implemented in the MATLAB Statistical Toolbox,[58] was used to calculate the first three eigenvectors of the RMSD matrix. The resulting three-dimensional graph, where each point represents a single conformation, was visually clustered to identify the native state ensemble and its exit.
Atoms less than 4.6 Å apart or 5.4 Å apart if one of the atoms was C or S and more than two residues separated in the primary sequence were judged to be in contact. A contact was defined as native if the two residues had a contact in the crystal structure. For the purposes of defining inter-domain, intra-domain, and domain bridge contacts in αLP, the N-terminal domain is residues 15A-120E and 231–245, the Domain bridge is residues 120A-121, and the C-terminal domain is residues 120G-230.
For each simulation snapshot, the number of native residue-residue contacts and the NPSASA were calculated. The values were binned into a two-dimensional histogram using bin sizes of 5 native contacts and 50 Å2. The landscape was generated by taking the negative natural logarithm of the bin counts at each position.
Ten conformational properties were used to generate the landscape: Cα RMSD, native intra-domain atom-atom contacts, native inter-domain atom-atom contacts, non-native intra-domain atom-atom contacts, non-native inter-domain atom-atom contacts, radius of gyration, non-polar SASA, polar SASA, non-native main-chain hydrogen bonds, native main-chain hydrogen bonds. Properties were scaled by dividing by subtracting the mean value and dividing by the standard deviation for each. Principal components analysis was performed with the MATLAB Statistics Toolbox. Loadings for each term in the PCA are shown in Supplemental Table 1. A two-dimensional histogram was computed using the first two principal components, with a bin size of 0.1 units. The landscape was generated by taking the negative natural logarithm of the bin counts at each position.
Two-fit Cα RMSDs were calculated for each pair of snapshots (10 ps intervals to reduce the number of pairwise comparisons) in a simulation. Cooperativity was defined as the number of snapshots less than 3 Å of the above Cα RMSD at each time point in the simulation multiplied by the snapshot interval (10 ps). Results were qualitatively similar using thresholds of 3.5 and 4.0 Å.
PyMOL[59] was used to generate Figures 1, 3, 4, 6, and 9.
|
10.1371/journal.pbio.1001157 | LIN-44/Wnt Directs Dendrite Outgrowth through LIN-17/Frizzled in C. elegans Neurons | Nervous system function requires proper development of two functional and morphological domains of neurons, axons and dendrites. Although both these domains are equally important for signal transmission, our understanding of dendrite development remains relatively poor. Here, we show that in C. elegans the Wnt ligand, LIN-44, and its Frizzled receptor, LIN-17, regulate dendrite development of the PQR oxygen sensory neuron. In lin-44 and lin-17 mutants, PQR dendrites fail to form, display stunted growth, or are misrouted. Manipulation of temporal and spatial expression of LIN-44, combined with cell-ablation experiments, indicates that this molecule is patterned during embryogenesis and acts as an attractive cue to define the site from which the dendrite emerges. Genetic interaction between lin-44 and lin-17 suggests that the LIN-44 signal is transmitted through the LIN-17 receptor, which acts cell autonomously in PQR. Furthermore, we provide evidence that LIN-17 interacts with another Wnt molecule, EGL-20, and functions in parallel to MIG-1/Frizzled in this process. Taken together, our results reveal a crucial role for Wnt and Frizzled molecules in regulating dendrite development in vivo.
| Neurons have distinct compartments, which include axons and dendrites. Both of these compartments are essential for communication between neurons, as signals are received by dendrites and transmitted by axons. Although dendrites are vital for neural connectivity, very little is known about how they are formed. Here, we have investigated how dendrites develop in vivo by examining an oxygen sensory neuron (PQR) in the nematode C. elegans. Using a genetic approach, we have discovered that Wnt proteins, a group of highly conserved secreted morphogens, interact with their canonical Frizzled receptors to control the development of the PQR dendrite. We show that Wnt molecules act as attractive signals to determine the initiation and direction of dendrite outgrowth. Interestingly, Wnt proteins act specifically on the dendrite without affecting the axon, suggesting that outgrowth of the dendrite can be regulated by distinct processes that are independent of axon formation. We predict that similar mechanisms may be in place in other species owing to the conserved roles of Wnt and Frizzled molecules in development.
| Correct dendrite development is essential for the establishment of neuronal connectivity and, in sensory neurons, for the detection of external stimuli. However, the complexity and variety in morphology of dendrites has made the study of their development more challenging than that of axons. Previous findings have shown that some axon guidance molecules can also regulate dendrite development, often with opposing effects. For example, the guidance cue Slit can simultaneously repel axons and enhance dendrite branching and outgrowth in cortical neurons [1]. Similarly, Semaphorin 3A, a guidance molecule that acts through the Neuropilin-1 receptor, functions as both a chemorepellent for cortical axons and a chemoattractant for dendrites within the same neurons [2]. The differential response of axons and dendrites to Semaphorin 3A is mediated by asymmetric localization of a soluble guanylate cyclase to the dendrites [2]. In cultured hippocampal neurons, local elevation of cAMP and reduction of cGMP in undifferentiated neurites promotes axon formation and suppresses dendrite formation, whereas the reciprocal levels of these molecules have the opposite effects [3]. Interestingly, local upregulation of cAMP in a single neurite results in long-range inhibition of cAMP levels in all other neurites, suggesting a mechanism for the development of one axon and multiple dendrites and indicating that dendrite formation in this context is secondary to axon formation [3].
More recently, in vivo studies have uncovered molecules that regulate dendrite development independently of the axon. Sensory neurons in the head of C. elegans develop by anchoring their dendritic tips to the nose while the cell body migrates away, extending a dendrite (retrograde extension) [4]. In the C. elegans tail motor neuron, DA9, the extracellular guidance cue UNC-6/Netrin controls the final extension of the dendrite in an axon-independent manner through its interaction with the receptor UNC-40/DCC [5]. In a different highly branched mechanosensory neuron, PVD, the cell-autonomous activity of the EFF-1 fusogen promotes branch retraction to retain a precise patterning of arbors during dendrite development [6]. In a Drosophila sensory neuron (vch'1), correct orientation of the dendrite is regulated by Netrin-A and its receptor Frazzled and is mediated by a migrating cap cell, which drags the tip of the dendrite into position [7]. In all these cases, however, the cell-intrinsic molecules involved in the initial stages of dendrite formation remain elusive.
Wnt morphogens and their Frizzled receptors are highly conserved molecules with diverse functions in nervous system development [8],[9]. In rat and mouse hippocampal neurons, Wnt molecules promote dendritic arborization [10],[11], whereas in Drosophila neuronal activity regulates the remodeling of dendritic branches in a Wnt-dependent manner [12]. In C. elegans there are five Wnt ligands, (LIN-44, EGL-20, CWN-1, CWN-2, and MOM-2) and four Frizzled receptors (LIN-17, MIG-1, CFZ-2, and MOM-5). The posteriorly expressed Wnt ligand, LIN-44, regulates neuronal polarity, axon guidance, axon termination, and synapse formation, acting mainly as a repellent through the LIN-17/Frizzled receptor on neurons in the posterior of the animal [13]–[17]. Another posteriorly expressed Wnt ligand, EGL-20, controls cell migration and axon guidance of different cells along the anterior-posterior axis of the worm [18]–[20]. CWN-1 and CWN-2, which are expressed more broadly along the anterior-posterior axis, affect neurons in the mid-body and the head of C. elegans, regulating neuron migration, axon guidance, nerve-ring placement, as well as the outgrowth and pruning of neurites [21]–[25].
In this study, we show that LIN-44/Wnt initiates and guides the development of the dendrite in the PQR oxygen sensory neuron, through a mechanism that occurs prior to and independently of the formation of the axon. In contrast to its role as a repellent in synapse formation and axon termination, in the context of PQR development LIN-44 acts as an attractant that is specific for the outgrowth of the dendrite. The effect of LIN-44 is mediated through the LIN-17 receptor, which functions in a cell-autonomous manner. We also identify EGL-20/Wnt and MIG-1/Frizzled as crucial molecules in PQR dendrite development. Taken together, these findings show for the first time that Wnt signals and Frizzled receptors can promote dendrite-specific outgrowth in developing neurons in vivo.
PQR is an oxygen-sensory neuron with its cell body positioned in the posterior lumbar ganglion on the left side of the animal [26]. PQR extends a single axon anteriorly along the ventral nerve cord and a single dendrite posteriorly towards the tail (Figures 1D, 2A). The tip of the dendrite, which is part of the left phasmid sensory organ, protrudes with its sensory cilia into the pseudocoelom. PQR is born post-embryonically, facilitating investigation of its development in newly hatched larvae. A gcy-36::GFP reporter was used as a selective marker for PQR, allowing visualization of its dendrite during development, starting at the L1 stage (see Materials and Methods).
PQR arises as a descendant of the QL neuroblast, and subsequently migrates towards the tail. We observed that upon reaching its final destination, at 5.5–6.5 h after hatching, PQR assumed a rounded or elliptical shape, without any neurites (Figure 1A). At 6.5–7 h, dendrite formation began with lamellipodia-like extensions emerging on the dorsal-posterior region of the cell body, which had become elliptical or triangular in shape (Figure 1B). At this stage, no other projections were present, indicating that dendrite outgrowth is initiated before outgrowth of the axon. At 7–7.5 h, the dorsal-posterior protrusion thinned and extended into a developing dendrite with a growth cone at its distal tip, and the cell body became rounder in shape (Figure 1C). At the same time, the axon began to emerge from the ventral-anterior side of the cell, appearing as a small neurite that, unlike the dendrite, did not present a large growth cone at its tip. By 7.5 h, both the dendrite and axon were visible and continued to extend to their final positions until 18 h after hatching (L2/L3) (Figure 1D). PQR subsequently maintained its morphology throughout adulthood (Figure 2A). Overall, our analysis demonstrates that the PQR dendrite forms by growth cone crawling and is initiated prior to axon outgrowth.
We next used a candidate gene approach to discover the molecules regulating dendrite development in PQR. We found that animals mutant for LIN-44/Wnt presented severe defects, with PQR dendrites that were short, absent, or misrouted in the anterior direction (Figure 2B–D, and quantified in 2E). The axon, however, appeared morphologically normal. These defects could arise from a dendrite-specific effect or a change in neuronal polarity whereby the identity of the neurites is compromised. To distinguish between these two possibilities we investigated whether there were any changes in the location of the presynaptic sites of PQR, which are normally on the axon. rab-3 encodes for a vesicle-associated Ras GTPase, which localizes to presynaptic densities [27],[28]. Using a YFP::RAB-3 fusion protein expressed specifically in PQR (Pgcy-36::YFP::RAB-3), we found that the presynaptic sites in lin-44 mutants were largely located on the axon as in wild-type animals (Figure 2F). This suggests that the identity of the neurites is unchanged and that the PQR defect of the lin-44 mutant is dendrite-specific.
Next, we tested if the PQR dendrite defect of lin-44 mutant animals could arise from an abnormal cell division in the precursor cell. However, we found that the asymmetric cell divisions of the PQR precursor occurred normally in the lin-44 mutant animals (Figure S1), precluding this possibility.
Finally, we investigated whether the absent and short dendrite phenotypes we observed were generated either by excessive pruning or by direct outgrowth failure. Examination of early stages of PQR development in lin-44 mutants revealed that the dendrite often failed to form or fully extend (Table S1); we also observed animals with dendritic growth cones developing abnormally on the anterior side of the neuron, which would explain the anteriorly misrouted dendrites observed in adults (Table S1). Thus, our results indicate that LIN-44 acts at very early stages of PQR development by regulating proper formation of the growth cone and its extension.
The Wnt ligand LIN-44 is expressed in close proximity to the PQR neuron from four hypodermal cells (hyp-8, -9, -10, and -11) in the tip of the tail [29], a position posterior to the PQR dendrite (Figure 3A). As the PQR dendrite grows towards the source of LIN-44, we hypothesized that this molecule might act instructively as an attractive cue for the developing dendrite. Alternatively, LIN-44 may act as a permissive cue, whereby its positional information is not essential for correct dendrite development. To distinguish between these two possibilities, we expressed LIN-44 ectopically from regions anterior to the PQR cell body in lin-44 mutant animals, using a version of LIN-44 genomic DNA that had been engineered to contain a secretion signal sequence to ensure proper secretion from cells that do not normally produce LIN-44 [16]. Transgenic lines were generated to express LIN-44 from the myo-2 promoter [30] in the pharynx (Pmyo-2::LIN-44), or from a short fragment of the cwn-1 promoter [21] in the intestine and head neurons (Pcwn-1::LIN-44) (Figure 3A and Figures S2, S3). When compared to lin-44 mutant animals, transgenic animals expressing LIN-44 anterior to PQR displayed a decrease in the proportion of normal dendrites and an increase in the proportion of dendrites that were misrouted in the anterior direction, towards the ectopic source of LIN-44 (Figure 3B and Figures S2, S3). On the contrary, expression of LIN-44 from its endogenous promoter (Plin-44::LIN-44) provided strong rescue of the PQR dendrite defect of lin-44 mutant animals (Figure 3B).
We next examined the ectopic expression of LIN-44 from the myo-2 promoter in the wild-type background and found that it altered the normal development of the PQR dendrite (Figure S4). Thus, the worsening of dendrite defects observed when LIN-44 is ectopically expressed from anterior regions suggests that LIN-44 has an instructive role in PQR dendrite development, whereby it acts as an attractive cue to direct the outgrowth of the dendrite.
In wild-type C. elegans, the four tail hypodermal cells hyp-8, -9, -10, and -11 express LIN-44 throughout embryogenesis and larval stages [29]. In order to define the time period in which LIN-44 is required for normal PQR dendrite development we eliminated larval production of LIN-44 by laser ablation of the hyp-8, -9, -10, and -11 hypodermal cells. Remarkably, in adult animals that were laser-ablated as newly hatched L1 larvae, the PQR dendrite appeared to be largely unaffected (Figure 4A) even though the ablations were performed several hours before PQR is born in the mid-L1 stage. This result indicates that LIN-44 expression from these hypodermal cells during embryogenesis is sufficient for the correct development of the PQR dendrite.
To further define the temporal requirement of LIN-44 we next utilized an inducible heat shock promoter to express LIN-44 (Phsp16-2::LIN-44) in a lin-44 mutant background at specific times during development. Heat shock-induced LIN-44 expression in newly hatched L1 animals partially rescued PQR dendrite defects (Figure 4B and Figure S5). However, when animals were heat shocked later, at the time of dendrite outgrowth, no such rescue effect was observed (Figure 4B), suggesting that LIN-44 expression is required prior to PQR dendrite outgrowth.
The hsp16-2 promoter drives expression broadly throughout the body of the animal, in cells that are both anterior and posterior to PQR [31]. Thus, the dendrite rescue we observed in heat shocked animals could indicate that LIN-44 plays a permissive role, or that the ligand is produced more efficiently from regions posterior to PQR. To further investigate this we expressed Phsp16-2::LIN-44 into a wild-type background and found that the ectopic expression of LIN-44 generated PQR defects similar to those of lin-44 mutants, confirming the instructive role of this molecule (Figure S6). Taken together, these results suggest that a molecular pattern of LIN-44 generated prior to PQR formation, during embryonic development and early L1, is both necessary and sufficient to instruct PQR dendrite outgrowth hours later, at which time the source of LIN-44 expression becomes dispensable.
LIN-17 is a Frizzled molecule known to function as a receptor for LIN-44 in a variety of developmental processes [14],[16],[17],[29],[32]–[35]. We found that lin-17 mutants had defects resembling those of lin-44, with PQR dendrites that were short, absent, and misrouted anteriorly (Figure 5A). lin-17 mutants also presented a strong migration defect [18],[19], with a high percentage (60% to 90%) of PQR neurons mispositioned in anterior regions of the body. Thus, our analysis was performed on those animals in which PQR was correctly positioned in order to eliminate any possible effect that the aberrant location may have had on PQR dendrite development. Importantly, lin-17 mutants, like lin-44 mutants, appeared to have largely normal localization of presynapses to the axon, as visualized using the YFP::RAB-3 fusion protein expressed specifically in PQR (Pgcy-36::YFP::RAB-3), eliminating the possibility of a switch in neurite identity (Figure 2F).
In addition to testing known alleles of lin-17, we also performed a forward genetic screen and isolated a previously uncharacterized allele, vd002, consisting of a G to A transition in position 490 of the lin-17 gene that resulted in a cysteine residue being replaced by a tyrosine residue (Figure 5A). The isolation of this mutant from an unbiased screen further supports the significance of lin-17 in this process.
To investigate whether there might be a genetic interaction between lin-17 and lin-44 with respect to PQR dendrite development, we next examined lin-17 lin-44 double mutants and found that the dendrite defects were qualitatively and quantitatively similar to those of lin-17 mutants (Figure 6A). This indicates that these two molecules function in the same genetic pathway with respect to PQR dendrite development and strongly suggests that LIN-44 acts as a ligand for LIN-17 in this process.
LIN-17 is expressed extensively and dynamically in several cells of the tail region including PQR (Figure S7) [35]. Wnt signaling through the LIN-17 receptor could occur cell-autonomously within PQR or could result from interactions with the surrounding cells. We first tested whether LIN-17 acts cell-autonomously by expressing the wild-type lin-17 cDNA from the gcy-36 promoter, which is transcriptionally active in PQR during the final stages of its migration. This transgene failed to rescue the dendrite defects, despite being tested at a range of different concentrations (see Materials and Methods). We therefore questioned whether LIN-17 might be required in PQR at earlier stages, before the gcy-36 promoter is transcriptionally active. To test this possibility we used the egl-17 promoter that is highly and selectively expressed in the precursors of PQR during the L1 stage [36],[37] to drive LIN-17 expression from the time PQR was born. Wild-type LIN-17 cDNA expressed by the egl-17 promoter (Pegl-17::LIN-17::YFP) strongly rescued the PQR dendrite defects of lin-17 mutants, to levels similar to that of the endogenous promoter (Plin-17::LIN-17::YFP) (Figure 5B). These results suggest that LIN-17 regulates dendrite development in a cell-autonomous fashion and is required very early in development, before or during PQR migration.
The PQR dendrite is ensheathed by PHso2L, a glia cell of the left phasmid sensillum; this sensillum comprises two socket cells (PHso1L, PHso2L), a sheath cell (PHshL), and two sensory neurons (PHAL and PHBL) [26]. Recent results in different systems have demonstrated a role of the support cells in regulating dendrite development [4],[7]. To determine if similar mechanisms were in place for PQR development, we next performed cell-ablation experiments whereby we selectively eliminated the socket cells or the socket cells together with the sheath cells. PQR morphology in ablated animals was largely normal, with only a small number of animals presenting short dendrites when left and right phasmid socket cells were ablated (3/15) or when left phasmid socket and left sheath cells were ablated (2/19). We never observed the penetrance and variety of defects of the lin-17 mutants. These results indicate that glial cells play a minor role in only the final stages of dendrite extension and suggest that LIN-17 does not have an effect on the PQR dendrite through these support cells (Table S2). In addition, ablations of the phasmid neurons PHA and PHB also had no effect on PQR dendrite development (Table S2), thereby providing further evidence that the function of LIN-17 in PQR dendrite development is unlikely to be mediated by the surrounding cells.
To further understand how LIN-17 acts on the PQR dendrite, we then asked at what stage in PQR development LIN-17 was visible on the cell membrane and how LIN-17 was distributed in PQR. Using a LIN-17::YFP functional fusion protein expressed under the control of the egl-17 promoter, we observed faint, relatively uniform localization of LIN-17 on the membrane of the QL.a cell as it was dividing into QL.aa and PQR (unpublished data). Following this division, the membrane-localized LIN-17::YFP in PQR decreased until it was barely visible at the time at which PQR had completed its posterior migration (unpublished data). This reduction in LIN-17::YFP appeared to be independent of down-regulation by the egl-17 promoter and is consistent with our previous results suggesting an early role for LIN-17 in regulating PQR dendrite outgrowth. We suggest that ubiquitous membrane-localization of LIN-17 may be required to detect the posterior source of Wnt ligand, which acts as the directional signal for the PQR dendrite.
Multiple Wnt ligands and Frizzled receptors are known to function in basic developmental processes in C. elegans and have frequently been shown to have redundant or synergistic roles. Although lin-44 mutants present striking PQR dendrite defects, 32% of these animals still have the ability to sprout a normal PQR dendrite, suggesting the involvement of other molecules in this process. We therefore tested three other Wnt molecules–EGL-20, CWN-1, and CWN-2–for possible roles in PQR dendrite formation. EGL-20 is expressed around the PQR cell body, in a group of epidermal and muscle cells near the anus [13],[20], and CWN-1 and CWN-2 are expressed to a greater extent anteriorly in the intestine, body wall muscle, and neurons in the midbody and head regions, anterior to the PQR cell body [13],[22],[38]. No significant dendrite defects were observed in cwn-1 or cwn-2 single mutants. The cwn-1 cwn-2 double mutant presented a higher percentage of ectopic processes from the cell body, and dendrite branching, compared to the single mutants, but no absent-dendrite or dendrite-misrouting defects were observed (Table S3). This suggests that these molecules are less directly involved in development of the PQR dendrite, but are important to prevent the formation of ectopic processes. Although the loss of cwn-1 alone caused no significant dendrite defects on PQR, when combined with the lin-44 mutation it was able to enhance the dendrite misrouting defects of lin-44 mutants (Table S3). Thus, CWN-1 might have a minor and redundant role in PQR dendrite development.
As previously described, egl-20 mutants have a very strong Q cell migration defect [18]–[20] resulting in 97%–98% of animals having anteriorly positioned PQR neurons. Restricting our analysis to those animals with PQR correctly positioned, we found that only 7% of egl-20 animals developed a normal, full-length dendrite, whereas the rest presented qualitatively similar defects to those of lin-44 and lin-17 animals, with absent, short, and anteriorly misrouted PQR dendrites (Figure 6B). egl-20 mutants presented a higher proportion of anterior dendrites, as compared to lin-44 mutant animals (Figure 6B), but the PQR dendrite phenotype of the egl-20 lin-44 double mutant did not display a significant worsening of defects when compared to the egl-20 single mutant. This suggests that egl-20 and lin-44 may interact to regulate PQR dendrite formation (Figure 6B). Furthermore, the egl-20 lin-17 double mutant was no worse than either of the single mutants (Figure 6C), suggesting that LIN-17 may act as a receptor for both EGL-20 and LIN-44.
Taken together, the above results indicate that egl-20 and lin-44 are the major regulators of PQR dendrite outgrowth, and appear to genetically interact, whereas cwn-1 plays only a minor role in the process. To determine the possible roles of other Frizzled receptors, we also studied PQR dendrite formation in cfz-2 and mig-1 mutants. cfz-2 mutants showed no significant defects, whereas mig-1 mutants presented 50% normal PQR dendrite (Figure 6D, Table S3). Thus, LIN-17 appears to be the main Frizzled receptor regulating PQR dendrite formation. To analyze functional redundancy among the Frizzleds, we tested whether mig-1 could enhance the lin-17 defect. In the mig-1 lin-17 double mutant, there was almost a 2-fold increase in the absent-dendrite phenotype (Figure 6D), indicating a possible parallel role of mig-1 in PQR dendrite formation.
Dendrites, the specialized structures that allow neurons to receive sensory information from the environment and to relay signals to one another, must develop properly in order to build a functioning nervous system. Recent reports of dendrite morphogenesis have advanced our understanding of dendrite sculpting and arborization [1],[6],[11],[12],[39], neuronal polarity [3],[11],[40], dendrite extension [4],[5], and dendrite orientation [2],[7]. To our knowledge, our study is the first to demonstrate that the initial outgrowth of a dendrite in vivo is controlled by Wnts and Frizzleds. Mutations in the Wnt ligands LIN-44 and EGL-20 and in the Frizzled receptors LIN-17 and MIG-1 cause a failure in dendrite development, resulting in dendrites that are absent, short, or misrouted. Our findings demonstrate that the Wnt ligand LIN-44 instructs the development of the dendrite through an attractive mechanism and is required prior to the initiation of dendrite outgrowth. This effect is likely to be mediated through the LIN-17 receptor, which acts cell-autonomously in PQR.
Several studies across different model systems have shown that Wnts can act instructively as both attractants and repellents in neurodevelopmental processes such as axon guidance, synapse formation, and neurite outgrowth [13],[16],[17],[23],[41]–[43]. Conversely, Wnt molecules can also act in a permissive manner, as non-spatial cues [14],[15],[20],[22].
Our results suggest that posteriorly expressed LIN-44 acts as an attractive cue for the PQR dendrite. Ectopic expression of LIN-44 from the anterior side of PQR increases the tendency for dendrites to emerge and grow anteriorly, towards the source of LIN-44. This role of LIN-44 as an attractant in PQR dendrite development differs from its role as a repellent signal for synaptic clustering in the dorsal section of the DA9 motor neuron [16], highlighting the distinct effect of LIN-44 on these neighbouring neurons.
The partial rescue of PQR dendrite defects by ubiquitous expression of LIN-44 from the heat shock promoter could suggest a permissive role for LIN-44. However, a possible alternative interpretation is that local asymmetry of the ligand is generated, providing rescue when the concentration is higher on the posterior side of PQR. This conclusion is supported by the observations that a higher concentration of ligand (increased length of heat shock) is unable to increase the rescue, and that in the wild-type background heat shock-directed expression causes dendrite defects. To be fully functional, Wnts must undergo post-translational modifications, sorting in the endoplasmic reticulum, and secretion from the cells where they are expressed [44]. It is possible that cells that do not normally express LIN-44 have lower efficiency in regulating the proper maturation and secretion of this Wnt molecule. Hence LIN-44 expression from the heat shock promoter may provide functional, secreted LIN-44 with variable efficiency depending on the tissue of expression.
Wnt patterning occurs during embryogenesis, at a time when many neurons are born. Our observation that PQR forms a normal dendrite following ablation of the tail hypodermal cells at the time of hatching suggests that embryonically expressed LIN-44 provides spatial information needed by the developing PQR several hours later. However, PQR remains receptive to heat shock misexpression of LIN-44 up until the dendrite begins developing. It is not known how stable Wnts are in C. elegans; however, in Drosophila the Wnt Wingless (Wg) and the morphogen Decapentaplegic (Dpp) are stable for about 3 h [45],[46]. Wnts can also function at long distances. In C. elegans, for example, EGL-20 has been shown to direct cell migration across half the animal's body length [20],[46]. Similarly in Drosophila, Wg can cover 10–20 cell diameters away from its source in the developing wing [47],[48] spreading over a distance of about 50 µm in 30 min [46]. Our results showing an effect of LIN-44 when expressed in the pharynx from the promoter myo-2 in a region far from PQR also suggest a potential long range effect for this ligand.
Emerging evidence suggests that dendrites of sensory neurons are shaped in a variety of ways. In contrast to dendrite development by retrograde extension, or towing by associated cells [4],[7], we and others [49] have observed that the dendrite of PQR forms by growth cone crawling, a mode of development more commonly seen in axons. In LIN-44 mutants, this growth cone often fails to form, preventing the outgrowth of a dendrite. Our results demonstrate that LIN-17, a receptor for LIN-44, cell-autonomously regulates the initiation and outgrowth of the PQR dendrite. To our knowledge, a ligand-receptor pair that can specifically affect the development of a dendrite in this manner has not previously been described.
Interestingly, phasmid glia associated with the PQR dendrite do not have a major effect on its development. It has previously been shown that lin-44 and lin-17 mutants have defects in phasmid socket glia that arise due to disrupted polarity of the T cell precursor [29],[35],[50]. However, the aberrant structure of the phasmid in these mutants does not seem to be the main cause of dendrite defects, as ablation of these cells did not reproduce the mutant phenotypes. Notably, glia appear to have some involvement in the final extension of the dendrite, as some ablated animals had short dendrites. This is reminiscent of a previous study in which it was demonstrated that ablation of the sheath glia associated with the CEP sensory neuron in the head of C. elegans resulted in a failure of the sensory dendrite of this neuron to fully extend [51].
Different lines of evidence suggest that LIN-17, like LIN-44, may be required early in development to promote normal dendrite outgrowth. Cell-specific LIN-17 expression can rescue lin-17 dendrite defects if induced very early, before PQR is born, but has no such effect when induced later, once the cell has almost completed its migration. Furthermore, LIN-17::YFP expression from the rescuing egl-17 promoter appeared to become extremely faint or absent by the time the dendrite began to develop. This raises the interesting possibility that levels of LIN-17 receptor on the PQR cell surface are temporally regulated to elicit the appropriate response to Wnt ligands. We propose a model in which the LIN-17 receptor, present at low levels on the membrane of the PQR cell from the moment it is born, detects a posterior source of LIN-44 that signals the dendrite to emerge from the posterior side of the cell (Figure 7A,B). This initial specification of the site of dendrite outgrowth appears to be an important determinant of the subsequent direction of dendrite outgrowth. The tendency for lin-44 and lin-17 mutant dendrites to grow anteriorly from the PQR cell, rather than from random orientations (including dorsal or ventral), may imply the presence of an intrinsic anterior-posterior bias of the site and direction of PQR dendrite outgrowth controlled by Wnts and Frizzleds, or the existence of a dorso-ventral dendrite outgrowth controlled by other guidance molecules still unknown.
In C. elegans, Wnts are expressed in different regions along the anterior-posterior axis. These different Wnts have often been shown to have distinct effects on cells that are located in proximity to the respective source of Wnt expression. Our genetic studies suggest that, similar to LIN-44, the posteriorly expressed Wnt ligand EGL-20 also acts through the LIN-17 receptor to regulate PQR dendrite development (Figure 7C), which could explain why lin-17 defects are more severe than those of lin-44. However, whether EGL-20 plays an instructive role in this process remains unclear. Previous studies have also shown that both LIN-44/Wnt and EGL-20/Wnt can function through LIN-17/Frizzled; however, whether Frizzled receptors can simultaneously bind multiple Wnts, or whether Wnts can form homo- or hetero-dimers, remains unknown.
The Wnt molecules CWN-1 and CWN-2 are both expressed more broadly in the body wall muscle, intestine, ventral cord neurons, and some head neurons [13],[21],[22],[38]. Although these Wnts do not appear to directly regulate PQR dendrite development, our observation that a significant proportion of cwn-1 and cwn-2 mutants present ectopic processes on PQR suggests an indirect role in neurite pruning. This is consistent with recent findings that identify CWN-1 and CWN-2 as key regulators of developmental pruning of the head neuron AIM [21].
The MIG-1 receptor appears to act synergistically in a parallel pathway to LIN-17 (Figure 7C). Notably, the increase in the percentage of the absent dendrite phenotype of the lin-17 mig-1 double mutant compared with the lin-17 mutant suggests a role for MIG-1 in regulating the ability of the neuron to send out a dendrite, regardless of its direction.
Wnt morphogens have diverse functions in developmental processes across species, yet how they act with such precision on a single cell within a closely wired nervous system remains enigmatic. As we and others have shown, spatio-temporal organization of Wnts and their Frizzled receptors must be tightly orchestrated. The challenge now will be to gain insight into how these molecules are patterned and how they can be interpreted differently by individual cells.
Nematodes were cultured using standard methods [52]. All experiments were performed at 18°C except where otherwise noted. The following mutations were used: LGI, lin-17(n677), lin-17(n671), lin-17(n3091), lin-17(vd002), lin-44(n1792), mig-1(e1787); LGII, cwn-1(ok546); LGIV, egl-20(n585), cwn-2(ok895); LGIV, cfz-2(ok1201). Transgenes used were: kyIs417[Pgcy-36::GFP, Podr-1::dsRed], kyIs403[Podr-1::dsRed2, Pflp-18::UNC-43g::dsRed2, Pgcy-36::YFP::RAB-3, Pgcy-36::mCFP], vdEx127[Phsp16-2:LIN-44 (10 ng/µl), Pcoelomocyte::GFP (25 ng/µl)], wyEx806[Plin-44::signal sequence:: flag::GFP::lin-44 genomic coding::lin-44 3′UTR, odr-1::GFP], vdEx224[Pcwn-1::signal sequence::flag::GFP::lin-44 genomic coding (20 ng/µl), Pcoelomocyte::GFP (30 ng/µl)], vdEx235(Pmyo-2::signal sequence::flag::GFP::lin-44 genomic coding (20 ng/µl), Pcoelomocyte::GFP (30 ng/µl)], vdEx251[Podr-1::dsRed (30 ng/µl), Pegl-17::LIN-17::YFP (20 ng/µl), Pgcy-36::mCherry (0.5 ng/µl)], vdEx133[Plin-17::LIN-17::YFP (10 ng/µl), Pchs-2::dsRed (2 ng/µl), pSM (10 ng/µl)], vdEx265[ Plin-17::mCherry (20 ng/µl), Pegl-17::GFP (50 ng/µl)]. The kyIs417 strain was generated in Cori Bargmann's lab, the kyIs403 strain was provided by Manuel Zimmer and Cori Bargmann, and the wyEx806 strain was provided by Kang Shen.
Standard molecular biology methods were used. All constructs were cloned into pSM (a kind gift from Steve McCarroll and Cori Bargmann), a derivative of pPD49.26 (Andrew Fire). The Pmyo-2::GFP::LIN-44 and Pcwn-1::GFP::LIN-44 constructs were generated by cloning a myo-2 promoter and a 170 bp fragment of the cwn-1 promoter [21] into FseI/AscI sites of pSM. A sequence encoding signal sequence::flag::GFP::LIN-44 genomic DNA (modified from the wyEx806 transgene [16]) was cloned downstream of each promoter into BamHI/NheI sites.
The Plin-17::LIN-17::YFP rescue plasmid was generated by inserting a HindIII/NheI digested 6.5 kb fragment of the lin-17 promoter upstream of a LIN-17::YFP clone (Pitr-1 pB::LIN-17::YFP [a gift from Kang Shen]).
The Pegl-17::LIN-17::YFP rescue plasmid was made by inserting a 5.4 kb NotI/FseI digested fragment of the egl-17 promoter upstream of LIN-17::YFP (digested from Plin-17::LIN-17::YFP plasmid and cloned into NheI/PspOM1 sites).
Pegl-17::mCherry was generated by digesting mCherry from a pSM mCherry clone and inserting into KpnI/PspOMI sites behind the egl-17 promoter. Pgcy-36::mCherry was created by inserting 1.1 kb gcy-36 promoter into pSM mCherry. The Plin-17::mCherry clone was generated by cloning a BamHI/NheI digested 6.5 kb fragment of the lin-17 promoter into pSM mCherry.
A Pgcy-36::LIN-17::YFP expression construct was unable to rescue dendrite defects in lin-17(n677) mutants when injected at concentrations of 0.2, 0.5, 1, 2, and 10 ng/µl.
We analyzed PQR development in synchronized populations of anesthetized larvae (L1 stage) in a kyIs417(Pgcy-36::GFP) background. Animals were synchronized by collecting newly hatched animals, from a plate containing only eggs, every 10 min using M9 buffer. Synchronized animals were transferred to fresh plates and grown for 5–9 h at 22°C before imaging. Developmental stages were characterized in synchronized populations, with little variation among animals.
PQR morphology was scored at L4 or adult stages. Mutations in mig-1, lin-17, and egl-20 caused PQR migration defects, resulting in anterior (and in some cases posterior) mis-positioning of PQR. Given that this would cause PQR to be in a different position in relation to its normal surroundings, and importantly the source of LIN-44, we chose to score dendrite defects only in those animals where PQR had migrated to its normal position. The PQR dendrite was scored as short if it was less than three cell bodies in length.
Wild-type and lin-44(n1792) mutant animals carrying the Phsp16-2::LIN-44 transgene were maintained at 18°C. As development at this temperature is slower than at 22°C as in Figure 1, dendrite outgrowth occurs at ∼8 h rather than ∼6.5 h. L1 animals were heat shock-induced at 33°C in a water bath for 30 min (or longer, where indicated) at different stages of development as indicated, following which they continued to grow at 18°C. Transgenic animals (lin-44; Phsp16-2::LIN-44) and non-transgenic controls (lin-44) were scored at the L4 stage or as young adults.
Animals were mounted on 4% agar pads and immobilized using tetramisole hydrochloride (0.01%–0.03%). Epifluorescence was used to visualize animals with a Zeiss Axioimager Z1 and a Zeiss Axioimager A1 microscope. A Photometrics camera, Cool snap HQ2, was used for imaging. Metamorph software was used to analyze the collected Z stacks. Developing stages of PQR were imaged using a Zeiss LSM510META confocal microscope and Zen 2008 software. An antifading agent, Dayco, was used in addition to tetramisole hydrochloride.
Laser ablations were performed in L1 animals carrying the kyIs417 transgene using a MicroPoint Laser System Basic Unit attached to a Zeiss Axio Imager A1 (Objective EC Plan-Neofluar 100×/1.30 Oil M27). Animals were ablated 0–1 h after hatching and were scored at the L4 stage. For ablations of phasmid glia and phasmid neurons, ablation success was determined at the L4 stage by soaking animals in DiI on slides for 2 h prior to scoring (DiI stains the phasmid neurons when these cells and the phasmid structure are unaltered [53]–[55]).
Statistical analyses were performed using Primer of Biostatistics 3.01. Error of proportions was used to estimate variation within a single population. The Student's t test was used in all cases, except in those with multiple comparisons, for which the Bonferroni t test was used.
|
10.1371/journal.pgen.1004644 | The Vesicle Protein SAM-4 Regulates the Processivity of Synaptic Vesicle Transport | Axonal transport of synaptic vesicles (SVs) is a KIF1A/UNC-104 mediated process critical for synapse development and maintenance yet little is known of how SV transport is regulated. Using C. elegans as an in vivo model, we identified SAM-4 as a novel conserved vesicular component regulating SV transport. Processivity, but not velocity, of SV transport was reduced in sam-4 mutants. sam-4 displayed strong genetic interactions with mutations in the cargo binding but not the motor domain of unc-104. Gain-of-function mutations in the unc-104 motor domain, identified in this study, suppress the sam-4 defects by increasing processivity of the SV transport. Genetic analyses suggest that SAM-4, SYD-2/liprin-α and the KIF1A/UNC-104 motor function in the same pathway to regulate SV transport. Our data support a model in which the SV protein SAM-4 regulates the processivity of SV transport.
| Most cellular components of neurons are synthesized in the cell body and must be transported great distances to form synapses at the ends of axons and dendrites. Neurons use a specialized axonal transport system consisting of microtubule cytoskeletal tracks and numerous molecular motors to shuttle specific cargo to specific destinations in the cell. Disruption of this transport system has severe consequences to human health. Disruption of specific neuronal motors are linked to hereditary neurodegenerative conditions including forms of Charcot Marie Tooth disease, several types of hereditary spastic paraplegia, and certain forms of amyotrophic lateral sclerosis motor neuron disease. Despite recent progress in defining the cargo of many of kinesin family motors in neurons, little is known about how the activity of these transport systems is regulated. Here, using a simple invertebrate model we identify and characterize a novel protein that regulates the efficacy of the KIF1A motor that mediates transport of synaptic vesicles. These studies define a new pathway regulating SV transport with potential links to human neurological disease.
| Neurons innervate their targets at synapses distant from the soma. Most components of these synaptic specializations, including synaptic vesicles (SVs), active zone proteins and mitochondria, are synthesized in the soma and then transported along axons on the microtubule cytoskeleton [1]. Transport along the axon is bidirectional with anterograde transport driven largely by kinesins and retrograde transport carried out by cytoplasmic dynein [2]. Efficient axonal transport is important in many facets of neuronal development and function. Trophic factors, membrane components, guidance receptors as well as synaptic components are all transported down the axon anterogradely, and maintenance of trophic support requires retrograde transport of signaling endosomes containing activated receptors [2]. Abnormal axonal trafficking has been observed in brain disorders including Parkinson's disease, amyotrophic lateral sclerosis, Charcot-Marie-Tooth disease and hereditary spastic paraplegia [3], [4], [5], [6].
The majority of anterograde transport is performed by a large family of plus-end directed motors of the kinesin superfamily (KIFs) consisting of 21 genes in C. elegans [7] and 45 genes in mouse [8]. KIFs are composed of three domains: a motor “head” domain, a stalk domain and a cargo-binding domain. In plus end directed kinesins, the globular ATPase motor domain is positioned in the N-terminal region of the protein and provides the force to walk processively on microtubules at mean velocities of around 0.5–1.5 µm/second [9]. The C-terminal cargo-binding domain is typically separated from the motor by a long coiled coil stalk or “neck” domain, though the size of this domain varies considerably within the family. By contrast with the highly conserved motor domain, the cargo binding domains are variable and determine the cargo specificity of KIFs. Accessory light chain subunits and distinct adaptor proteins provide additional diversity of cargo binding to KIFs [9]. For example, KIF5 binds APP containing vesicles via its light chain [10], mitochondria via the adaptor Milton [11] and GlrR2 containing vesicles via the adaptor GRIP [12]. Although the cargo binding specificity of numerous kinesins has been defined to some extent, the mechanisms regulating many aspects of kinesin-mediated cargo transport remain largely uncharacterized.
One general theme in the mechanisms controlling axonal transport is the regulation of KIF motor activity. The activity of the motor domain of several different KIFs, including KIF1A/UNC-104, is negatively regulated by their cargo-binding domain in the absence of cargo [13], [14], [15], [16], [17], [18]. In addition, activation of the motor in several cases has been documented to require the binding of other factors. For example, the cargo adaptor JIP1 is not sufficient to activate Kinesin-1, but rather requires the additional cooperative binding of the protein FEZ1 [19]. A RAN-GTPase binding protein has been shown to activate Kinesin-1 ATP activity in vitro [20]. Phosphorylation has also been demonstrated in several cases to regulate cargo binding. For example, CaMKII regulates KIF17 binding to cargo by phosphorylation and GSK3β phosphorylation regulates KIF5 [21], [22]. In addition, the microtubule associate protein (MAP) doublecortin was recently demonstrated to regulate SV transport by enhancing KIF1A motor domain binding to MTs [23]. In summary, regulation of KIF motor activity is complex.
One of the identified KIF1A/UNC-104 regulators is Liprin-α/SYD-2. Liprin-α/SYD-2 belongs to a family of proteins that interact with the cytosolic domain of LAR receptor protein tyrosine phosphatases [24], [25]. In addition to interacting with LAR, Liprin-α interacts with several presynaptic active zone proteins to regulate active zone development [26], [27], . Interestingly, biochemical studies also identified interactions of Liprin-α with KIF1A [32]. In vivo, Liprin-α/SYD-2 is required for SV trafficking in Drosophila [33] and regulates UNC-104 motility in C. elegans [34]. These observations demonstrate that, in addition to the intra-molecular regulation of KIF1A/UNC-104, its activities are also regulated by other factors.
Here, using the C. elegans mechanosensory system as an in vivo model, we identify SAM-4 (Synaptic vesicle tag Abnormal in Mechanosensory neurons) as a novel regulator of KIF1A/UNC-104 directed SV trafficking. sam-4, encodes a conserved SV-associated protein orthologous to human LOH12CR1 [35] that is broadly expressed in neuronal tissue. SAM-4 acts in a cell autonomous manner by binding to SVs to regulate the processivity of anterograde SV transport. sam-4 null mutants show SV trafficking defects in different neuronal cell types. Genetic analyses revealed that SAM-4 acts synergistically with the KIF1A/UNC-104 PH cargo binding domain, but not the motor domain, to regulate SV trafficking and locomotory behavior. Gain-of-function mutations in the unc-104 motor domain suppress sam-4 defects indicating that SAM-4 functions upstream of the motor in regulating SV transport. SYD-2, which regulates SV trafficking to a lesser extent than SAM-4, exhibits similar genetic interactions with UNC-104 but no obvious interactions with SAM-4, consistent with SYD-2 and SAM-4 acting in the same pathway. Imaging of SV cargo movements in vivo demonstrated that SAM-4 is required to maintain cargo processivity rather than motor velocity, while gain-of-function UNC-104 proteins increase cargo processivity. We propose a model in which SV-bound SAM-4 acts in parallel to the UNC-104/KIF1A cargo binding domain to regulate activity of the motor domain.
The response to gentle touch to the body in C. elegans is mediated by a set of six touch receptor neurons (TRNs: ALML/R, AVM, PLML/R, and PVM) (Figure S1A and [36]). We use PLM neurons as a simple in vivo system to examine axonal transport of synaptic components. The two PLM soma are located on each side of the body in the tail ganglia (Figure S1A). Each PLM extends a short posterior-directed and a long anterior-directed neurite, which are easy to image because they are in close apposition to the cuticle. PLMs innervate partners via gap junctions and chemical synapses [37]. The chemical synapses are formed in a large varicosity (∼5 µm long), located at the end of single collateral synaptic branch that extends ventrally from the anterior directed process into the ventral nerve cord, usually just posterior to the vulva (Figure S1A). We examined PLM neurons in vivo by expressing markers using the mec-7 promoter which drives gene transcription selectively in TRNs [38]. SVs preferentially accumulate in the PLM synaptic varicosities as observed using transgenic SV markers SNB-1-GFP [39] and GFP-RAB-3 (jsIs821, Figure 1A and 1C), similar to SV accumulations revealed at the ultrastructural level [37], [40]. When anterograde SV trafficking machinery is disrupted by lesioning the UNC-104/KIF1A motor, SV markers accumulate in the soma and proximal portions of PLM neurites rather than being transported to the synapse (Figure S1B–D′). By contrast, when the retrograde cytoplasmic dynein motor is disrupted, SV markers accumulate abnormally at the distal portions of the anterior process [41]. These observations indicate that homeostatic regulation of SV levels in mechanosensory neuron synaptic varicosities is mediated by the balance of the anterograde and retrograde transport systems.
The sam-4(js415) mutant was isolated in a forward genetic screen for mutations disrupting SV accumulation in PLM synapses, using a SNB-1-GFP transgenic marker [42]. Similar defects were observed when SV localization was analyzed using GFP-RAB-3 (Figure 1A–F). In sam-4(js415), GFP-RAB-3 fluorescence was greatly reduced in PLM synaptic varicosities (Figure 1B and D) and increased both in the soma and the process proximal to the soma where the accumulations were largely punctate (Figure 1B and F). We also found that the accumulation phenotype in PLM neurons is temperature sensitive: mutants raised at 25°C exhibit more severe defects than those raised at 15°C (Figure S2). In addition to the abnormal SV marker accumulations in PLM neurons, similar defects were also observed in other neurons including SAB neurons and ventral nerve cord neurons when using either SNB-1-GFP or GFP-RAB-3 SV markers (Figure S3A–G) [39], [43]. Thus, SAM-4 appears to be essential for efficient transport of SVs in different types of neurons.
The altered distribution of GFP-RAB3 that we observe in sam-4 mutants could be explained by the disruption of neuronal morphology and/or the microtubule cytoskeleton. To test if the reduced levels of SV markers in sam-4 PLM synaptic varicosities are caused by PLM anatomical defects, we examined PLM neurites using a cytosolic fluorescent marker mRFP (jsIs973) and found no obvious morphological changes: PLM neurites extend normally, terminate properly in the mid-body, form the synaptic branches at the appropriate location and form synaptic varicosities in the ventral nerve cord (Figure 1G and H). Since microtubules function as a common track for the anterograde transport of many synaptic components including SVs, mitochondria and active zone proteins [1], we asked if sam-4 mutations cause microtubule cytoskeleton disruptions by examining the localization of synaptic components other than SVs. We found that the distribution of active zone proteins (mec-7p::tagRFP-ELKS-1, jsIs1075; Figure 1I–L) and mitochondria (mec-7p::tagRFP-mito, jsIs1073) (Figure 1M–P) is grossly normal in sam-4 mutants, indicating that transport of other synaptic components is largely intact. Thus, the microtubule cytoskeleton remains competent for axonal transport.
To evaluate systemic effects of sam-4 mutations, we next examined locomotion behavior which has been associated with SV trafficking defects [44]. Surprisingly, sam-4 null (see below) mutants exhibit only mild defects in the velocity of stimulated locomotion and show little, if any, defects in posture or the trace of sinusoidal locomotion tracks (Figure S4). We also examined other behaviors of sam-4 mutants and detected no defects in mechanosensation, egg-laying, or growth rates. Furthermore, sam-4 males remain competent to mate. These observations suggest that sam-4 may encode a specialized neuronal component that promotes efficient SV transport, without being essential for the process.
Positional cloning and transgenic rescue identified sam-4 as the C. elegans gene F59E12.11 (Figure S5 and Materials and Methods for details), which encodes an evolutionarily conserved 240 amino acid protein (Figure S5) with no identifiable domain structure. An additional open reading frame (ORF) was identified in the 5′ UTR of the sam-4 mRNA, but these sequences are not required for functions we describe for sam-4 herein (see discussion for details). SAM-4 is the C. elegans ortholog of human LOH12CR1 that was identified as a candidate tumor suppressor based on frequent deletion of this region of human chromosome 12 in acute lymphoblastic leukemia [35]. To confirm the sam-4 gene identification, we expressed a 3X-FLAG-tagged derivative of the SAM-4 protein (Figure S5C) under its native promoter using a MosSCI strategy [45], [46]. The single copy sam-4-3XFlag transgene completely rescued the Sam phenotypes of sam-4 mutants (Figure 2E). Immunohistochemical analysis of the transgene revealed that SAM-4 is localized primarily to the nerve ring region of the head (Figure S6A), indicating that sam-4 is broadly expressed in the nervous system.
The sam-4 mutations we characterized are recessive and likely represent null alleles of sam-4. The js415 allele isolated in our screen introduces a CAA>TAA nonsense lesion at Gln104 (Figure S5A and S5B). tm3828, another sam-4 allele isolated by the Japanese National Bioresource Project, deletes 149 bp of sam-4. This deletion removes exon sequences coding for amino acids from Leu66 to Ala100 and results in a frame-shift (Figure S5A and S5B). tm3828 and js415 exhibit indistinguishable GFP-RAB-3 mis-accumulation phenotypes (Figure S2). Since both mutations result in severe disruption of coding potential of sam-4 and have similar phenotypes, we conclude that both alleles represent null mutations.
To characterize the role of SAM-4 protein in regulating SV transport, we assayed its function when expressed in different cell types (Figure S5C). We found that sam-4 expression in PLM neurons driven by the mec-7 promoter rescued the SV accumulation defects in PLMs (Figure S5D) while its expression in PLM postsynaptic partners driven by the glr-1 promoter did not. These data suggest that SAM-4 functions cell-autonomously to regulate SV transport.
We next used a functional sam-4-TagRFP transgene expressed in PLMs to further examine the sub-cellular localization of SAM-4. We observed that SAM-4 preferentially accumulates in the synaptic varicosities of PLMs and small quantities of SAM-4 accumulate as puncta in the neurites (Figure 2A and 2A′), a pattern similar to the GFP-RAB-3 marker (Figure 2B and 2B′). Further examination demonstrated that these SAM-4 particles co-localize well with the RAB-3 labeled SV particles (Figure 2C and 2C′) and furthermore the RAB-3 and SAM-4 particles move together (Figure S6B, Movie 1–3). In addition, SAM-4-TagRFP is retained in the cell body in unc-104 mutants as previously demonstrated for many other SV proteins including RAB-3 [39], [47]. These observations suggest that SAM-4 may function as a component of the SV trafficking machinery.
To determine if SAM-4 is a SV component, we examined SAM-4 subcellular localization using cell fractionation analysis. We lysed sam-4-3XFlag transgenic animals under detergent free conditions, cleared the lysate of large membrane organelles, cytoskeleton, and cell debris using a 15K g spin, then fractionated the extract into a membrane containing 150K g pellet and a cytosolic fraction. We observed that, like the SV protein synaptobrevin SNB-1, SAM-4 was present in the SV membrane-containing pellet but was absent from the cytosolic fraction (Figure 2D), indicating that SAM-4 is likely associated with SVs. Bioinformatic analysis predicts that SAM-4 contains a conserved myristoylation site at its amino terminus (Figure S5B). We then tested if SAM-4 localizes to membranes through the myristoylation signal. Myristoylated proteins typically migrate faster than their non-myristoylated counterparts [48]. We observed a decrease in mobility of SAM-4(G2S)-3XFLAG tagged protein compared to the SAM-4-3XFLAG control expressed at endogenous levels consistent with the hypothesis that this mutation disrupts SAM-4 myristoylation in vivo (Figure 2E). However, fractionation of SAM-4 to the membrane compartment was not altered by the SAM-4 (G2S) lesion suggesting that SAM-4 associates with membranes independently of myristoylation (Figure S6C). In fractionation experiments when EDTA and EGTA were omitted from the buffer, we also observed FLAG immunoreactive band 4 kD smaller than the full length SAM-4 which fractionated partially into the cytosol (Figure S6D) suggesting the SAM-4 N-terminus contains a site mediating interactions with an unidentified SV membrane component.
We further examined functional activity of the sam-4(G2S) myristoylation mutant and found that the endogenous expression level of sam-4(G2S) (jsIs1265) only partially rescue sam-4(null) mutants (Figure 2F). In addition, we observed that the SAM-4 (G2S) protein is not efficiently delivered to synapses and is largely retained in the soma (Figure 2G–H′). Taken together, these results argue that SAM-4 functions as a SV component to regulate SV trafficking.
Axonal transport of SVs in synapses is mediated by anterograde transport (largely the KIF1A motor system) and retrograde transport (the dynein motor system). To understand mechanisms by which SAM-4 regulates SV trafficking, we examined genetic interactions of sam-4 with mutations in both unc-104 and the dynein heavy chain gene dhc-1. Hypomorphic mutations were used because null mutations in both genes are lethal and because point mutations in different domains of UNC-104 are available for analysis. We first tested if SAM-4 is involved in regulating the UNC-104 transport machinery by examining sam-4 unc-104 interactions. We previously isolated a hypomorphic unc-104 loss-of-function (lf) mutant, js901, with a G1466V lesion in the cargo binding PH domain of UNC-104 that displays very similar phenotypes to sam-4 (materials and methods for details). These mutants show decreased GFP-RAB-3 levels in PLM synaptic varicosities and increased accumulations in the proximal portion of PLM neurites (Figure 3A–3C′). Furthermore, they displayed mild locomotion defects (Figure 4A, 4C and 4I) while remaining grossly normal in PLM neurite morphology, growth rate and egg-laying behavior. js901 males remained competent to mate. Overall, the phenotypic defects of unc-104(js901) are mild compared to other unc-104 alleles such as the e1265 PH domain and the rh43 motor domain lesions which have severe locomotory defects and slow growth rates. If SAM-4 interacts with UNC-104 to regulate SV transport, we reasoned that sam-4 mutations would exaggerate the mild js901 defects. Indeed, we observed that SV soma accumulations are further increased in the sam-4 unc-104(js901) double mutant relative to either single mutant (Figure 3A–3D′). Additionally, we found that sam-4 unc-104(js901) double mutants exhibit very severe locomotion defects relative to either single mutant, exhibiting defects comparable to severe unc-104 mutants (Figure 4A–4D, 4I). These results suggest that SAM-4 functions in concert with the UNC-104 protein to regulate the SV trafficking.
It has been previously demonstrated that the UNC-104 PH domain functions independently from the motor domain [49], [50]. The motor domain can walk on microtubules independently of the PH domain, and the PH domain can interact with vesicles independently of the motor domain. To assess the mechanistic implications of the genetic interactions between SAM-4 and UNC-104, we examined the allele specificity of these interactions. Specifically, we first examined sam-4 interactions with a SV binding defective unc-104 allele, e1265, which introduces a missense mutation (D1498N) in the PH domain and causes severe defects in SV binding [51]. Since e1265 mutants show virtually no detectable GFP-RAB-3 signal in neurites and severe locomotion defects with essentially no sinusoidal movements within the time-frame of our measurements (Figure 4H), we analyzed the sam-4 and unc-104(e1265) interactions by scoring animals for pharyngeal pumping, a behavior which is also controlled by neuronal activity [52]. We found that pharyngeal pumping rates of the double mutants were significantly lower than those of e1265 animals (Figure 4I). sam-4 unc-104(e1265) double mutants also had lower brood sizes and slow growth relative to either single mutants. These results suggest that SAM-4 acts synergistically with the PH domain to regulate SV trafficking.
We then tested how sam-4 mutations interact with unc-104(lf) mutations in the motor domain. unc-104(rh43) introduces two missense mutations in the motor domain and results in its motility defect [51]. These mutants exhibit severe locomotion defects again limiting our assay of animal movements. We applied pharyngeal pumping tests to evaluate their interaction. Surprisingly, we found that pharyngeal pumping defects introduced by the rh43 mutations are not exacerbated by the sam-4 mutation (Figure 4J). Furthermore, we noticed that while the pumping defects of e1265 mutants are less severe than those of rh43 mutants, these defects of sam-4 unc-104 (e1265) are more severe than those of sam-4 unc-104 (rh43) (Figure 4J). Thus, sam-4 exhibits allele specific synthetic interactions with PH domain lesions, but not motor domain lesions of unc-104. Taken together, these results suggest that SAM-4 functions by improving the UNC-104 motility, and acts in parallel to the UNC-104 PH domain to regulate SV trafficking.
To further explore the notion that SAM-4 enhances UNC-104 movement, we examined UNC-104 motor activity indirectly in sam-4 mutants by determining the localization of native protein. While UNC-104 protein is barely detectable in soma of wild type animals, we observed a dramatic increase of somatic UNC-104 accumulation in sam-4 mutants (Figure 4K–4L′). Since UNC-104 expression level is not affected by the sam-4 mutations (Figure S9B), these data indicate that UNC-104 motility is disrupted.
By contrast with unc-104, dynein dhc-1 mutants exhibit accumulations of GFP-RAB-3 in the distal portion of the anterior PLM neurite presumably due to disruption of retrograde transport, but have largely wild type levels of GFP-RAB-3 in both the PLM soma and synaptic varicosities. We found that dhc-1(js319); sam-4 double mutants show vestiges of both mutant phenotypes: while GFP-RAB-3 levels are modestly increased in the distal portion of PLM neurites resembling dhc-1 phenotypes, GFP-RAB-3 levels are greatly reduced in the PLM synaptic varicosities and increased in the proximal portion and soma resembling sam-4 phenotypes (Figure S7). This combination of phenotypes is similar to that of dhc-1(js319); unc-104(js901) animals (Figure S7). We interpret these phenotypes of dhc-1; sam-4 double mutants as a combination of the sam-4 and dhc-1 induced defects in SV transport. Therefore, sam-4 shows no obvious genetic interactions with dhc-1. Taken together, these genetic interactions support the model that SAM-4 regulates SV anterograde transport through UNC-104 motor domain.
To directly address how SAM-4 regulates SV transport, we examined GFP-RAB-3 puncta dynamics in PLM neurites of sam-4 mutants using time-lapse imaging (Figure 5). We found that SV anterograde transport is significantly reduced in sam-4 mutants as revealed by a reduced number of moving particles, reduced run-length of particles and increased frequency of pauses (Figure 5E–5H). However, the velocity of the GFP-RAB-3 transport was similar to wild type (Figure 5F). Retrograde trafficking is similarly affected by sam-4 mutations (Figure 5), in agreement with previous observations that retrograde trafficking is linked to anterograde trafficking of SVs [51], [53]. sam-4 defects were similar in severity to those of unc-104(js901) mutants. However, sam-4 unc-104 double mutants show more severe defects in GFP-RAB-3 trafficking (Figure 5), consistent with our behavioral and cell biological observations. These findings argue that SAM-4 regulates the anterograde trafficking of SVs by modulating the processivity of SV transport.
In C. elegans, SYD-2 liprin-α has been shown to regulate SV transport by binding the FHA domain and stalk domain of UNC-104 [34]. With our observations on sam-4 unc-104 interactions in regulating SV transport, we next examined the relationship between syd-2 and sam-4. We first confirmed that syd-2(ok217) null mutants show increased GFP-RAB-3 levels in the PLM soma and decreased levels in PLM synaptic varicosities (Figure 3), suggesting that anterograde SV trafficking is reduced. The GFP-RAB-3 accumulation defects in syd-2 mutants are less severe than that in sam-4 mutants (Figure 3). Nevertheless, similar to that observed in sam-4 unc-104(js901) mutants, abnormal soma and proximal neurite GFP-RAB-3 accumulations become much severe in unc-104(js901); syd-2 mutants relative to either single mutant (Figure 3). Furthermore, unc-104(js901); syd-2 mutants show more severe defects in locomotion than either single mutant (Figure 4I). Similar genetic interactions to those observed in sam-4 unc-104(e1265) and sam-4 unc-104(rh43) animals were observed in unc-104(e1265); syd-2 and unc-104(rh43); syd-2 (Figure 4J). Taken together, these data suggest that SYD-2 acts synergistically with UNC-104 PH domains to regulate SV trafficking in a similar manner as SAM-4.
We next examined sam-4 syd-2 interactions and observed no detectable genetic interactions between the two mutants. Double mutants display sam-4-like GFP-RAB-3 accumulation defects (Figure 3), and similar stimulated locomotion behaviors as either single mutants (Figure 4I). Over-expression of sam-4 does not suppress syd-2(ok217) mutants, and syd-2(ju487), a gain-of-function allele, has no effects on sam-4 defects (Figure S8). These results are consistent with the hypothesis that SYD-2 and SAM-4 function in the same pathway to regulate SV trafficking.
To further understand how SAM-4 activity regulates SV trafficking, we conducted a genetic screen for sam-4 suppressors. Using ENU induced mutagenesis, we screened mutated progeny of sam-4(js415); jsIs821 for animals with increased GFP-RAB-3 signal in PLM synaptic varicosities (Figure 6A–6D) and isolated two suppressors from roughly 100,000 genomes screened. Combining traditional genetic mapping and whole genome sequencing strategies, we identified both mutations as novel unc-104 alleles (see Materials and Methods for details). Interestingly, we found that the alleles introduce missense mutations in the UNC-104/KIF1A motor domain: S211A (js1288) and D177A (js1289), both of which are conserved in mammalian molecular motor proteins (Figure S9). Further genetic tests showed that both alleles are semi-dominant in suppressing sam-4 defects. Additionally, we found that over-expression of wild type unc-104 (unc-104(+)) in PLM neurons suppresses sam-4 defects (Figure 7A–7D and 7K), but over-expression of sam-4 does not suppress unc-104 defects (Figure S7). Similar suppression analysis using syd-2 mutants by these unc-104(gf) mutations also revealed suppression by unc-104(gf) alleles (Figure 6). Taken together, these data argue that both js1288 and js1289 are gain-of-function alleles of unc-104, and unc-104 is epistatic to sam-4 and syd-2.
To address how the unc-104(gf) suppresses the SV trafficking defects of sam-4, we characterized the two unc-104 alleles in the absence of sam-4. In isolation, js1288 and js1289 show grossly normal mechanosensory neuron anatomy (Figure 7F, 7G). We analyzed their effects on transport by examining GFP-RAB-3 distribution in vivo. We found that GFP-RAB-3 accumulations are significantly increased in the distal part of PLM neurites (Figure 7E–7H) in each of these unc-104 mutants but decreased in the soma (Figure 7E″–7G″, 7J), indicating that SV transport is enhanced by these two mutations. However, we did not observe GFP-RAB-3 increase in PLM synaptic varicosities (Figure 7E′–7G′, 7I). This is probably because either SV levels in PLM varicosities are already saturated in the wild type background or other mechanisms exist at pre-synapses to maintain SV homeostasis. To further understand how these mutations affect SV dynamics, we examined GFP-RAB-3 trafficking using live imaging (Figure 8). We found that both mutations result in increased run length of GFP-RAB-3 transport (Figure 8E). We also noticed that js1289 results in greater flux of GFP-RAB-3 (Figure 8F), while jsIs1288 reduces SV transport velocity (Figure 8G). Thus, processivity of vesicle transport is increased in both gain-of-function mutants, though perhaps by distinct mechanisms. Western blot analysis of protein levels showed that neither of these two lesions alter UNC-104 protein levels in vivo (Figure S9B). Hence, increasing processivity of the SV transport through the UNC-104 motor domain can partially bypass the need for SAM-4. This is consistent with our hypothesis that SAM-4 functions through the UNC-104 motor domain to regulate SV transport.
In this study we have identified the conserved protein SAM-4 as a novel vesicular component regulating SV transport in C. elegans. SAM-4 behaves as a SV associated protein and modulates transport probably by regulating the motor domain activity of UNC-104. This possibility is supported by our identification of two unc-104(gf) motor domain mutations, which suppress sam-4 SV transport defects. Although our genetic evidence is consistent with a SAM-4 UNC-104 interaction, we have been unable to detect any evidence for physical interactions between SAM-4 and UNC-104 either in vitro by yeast two-hybrid analysis or in vivo by co-immunoprecipitation. Therefore, SAM-4 UNC-104 interactions may be mediated by other components. Our genetic data also indicate that SAM-4 acts in the same pathway as SYD-2 in regulating SV transport. We propose a model in which SV-bound SAM-4 regulates SV transport together with SYD-2 through UNC-104, likely via its motor domain (Figure S10).
It is known that SV transport is regulated, but little is known of the molecular mechanisms involved. The identification of SAM-4 as a SV-bound regulator of KIF1A/UNC-104-mediated transport defines a new pathway for modulation of axonal transport. Although SAM-4 is conserved, analysis of the protein sequence revealed only a N-terminal myristoylation motif, which appears to contribute to SAM-4 activity. The lack of identifiable protein domains in the protein make it difficult to speculate on a specific mechanism of action. We have proposed that SAM-4 modulates SV transport processivity by modulating UNC-104 motor activity because we observed strong genetic interaction between sam-4 and unc-104 cargo binding mutants but not with motor domain mutants. Furthermore, motor domain gf mutations suppress sam-4 defects arguing that increase of motor processivity can partially bypass SAM-4 activity.
In addition, the genetic interactions between syd-2 and sam-4 support a processivity based mechanism of action for SAM-4. Both worm and mammalian Liprin-α/SYD-2 interact with KIF1A/UNC-104 [32], [34] and Liprin-α/SYD-2 is required for efficient SV trafficking in both C. elegans and Drosophila [33]. Our data argue syd-2 functions in the same pathway as sam-4 in regulating SV transport since each null mutant shows very similar interactions with both unc-104(lf) and unc-104(gf) lesions, but do not display obvious interactions with each other. However, SAM-4 may play a more central role in this process since the SV trafficking phenotypes in syd-2(null) are less severe than those of sam-4(null).
In this study, we recovered two gain-of-function mutations in the motor domain of unc-104 that increase the processivity of the motor in cargo movement assays. Kinesin mediated SV transport is an ATP driven process, which depends on motor-microtubule binding. The ATP hydrolysis catalytic core lays in the switch I region of the KIF1A/UNC-104 motor domain (Figure S9). Lesions (for example H215Y in unc-104(y211), see Figure S1) in this domain cause severe SV trafficking defects. The js1288 mutation occurs in S211A adjacent to S212 (S215 in mammalian KIF1A) which coordinates the gamma-phosphate of ATP in the ATP-bound crystal structure [54]. Consistent with the hypothesis that this lesion alters rates of ATP hydrolysis, we observed a lowered velocity of transport in js1288 mutants. However, the biochemical mechanism underlying the increase in processivity is unclear. The other mutation, js1289, is a D177A substitution (mammalian KIF1A D180) in loop 8 of the motor domain. Previous studies [54] showed that loop 8 is one of three microtubule binding regions in the motor domain and thus processivity in this mutant could be increased due to changes in the affinity for microtubules. In addition to suppressing SV trafficking defects in sam-4(null), both of these lesions result in increased accumulations of SVs in the distal portion of PLM neurites where no synapses have been seen at the ultrastructural level [40]. Therefore, these unc-104 (gf) mutations disturb the normal homeostasis of SV trafficking and thus may not necessarily represent beneficial biochemical modifications. However, the lesions argue strongly that processivity is not optimized in KIF1A and suggest the possibility that KIF1A activity could be modified, for example by pharmacological compounds, in diseases where axonal transport is compromised.
It is worth noting that several lines of evidence imply sam-4 also regulates other processes in non neuronal cells. First, sam-4 neuronal phenotypes are partially maternally rescued. Some sam-4 animals segregating from the sam-4/+ mother even display wild type levels of GFP-RAB-3 at PLM synapses. Second, sam-4 is likely post-transcriptionally regulated. The sam-4 locus is highly unusual (for nematodes) in that it has two 5′ “non-coding” exons (Figure S5). These exons contain a small 79 amino acid ORF. A similar ORF is found in the 5′-end of sam-4 in highly divergent nematodes and the synonymous codon usage in these nematodes indicates it is being selected as coding sequence (Figure S11). The ORF is homologous to the APC13, a small subunit of the Anaphase Promoting Complex (APC) which was previously described for plant parasitic nematodes [55], but recognized in model system databases. We, and other investigators, observed lethality when performing RNAi against sam-4 even though both nonsense and deletion alleles of sam-4 are fully viable. These sam-4 RNAi lethal phenotypes are similar to those of other APC complex component genes. These include defects in meiosis in the early embryo [56], oocyte deformation and sterility [57] and failure to segregate germline P-granules [58]. We posit the lethality phenotype associated with sam-4 RNAi is likely due to reduced expression of this upstream ORF encoding an APC13 homolog.
The APC complex plays critical roles in regulating progression through the cell cycle. However, recent work has also highlighted several critical roles for APC complexes in neuronal development [59]. In particular, disruption of the APC complex alters axon growth, post-synaptic glutamate receptor levels [60] as well as the size and number of presynaptic boutons [61]. Interestingly, in regulating bouton number in Drosophila, the APC complex works in conjunction with liprin-α. Thus, the APC complex, SAM-4 and liprin-α appear linked at multiple different regulatory levels. Further investigations of the non-neuronal roles of SAM-4, the role of the APC13 encoding upstream ORF in regulating SAM-4 expression, and the potential role of the APC complex in regulating axonal transport are clearly warranted.
Although SAM-4 is evolutionarily conserved, no human disease conditions have been specifically associated with lesions in human sam4 (LOH12CR1), a gene within a region often deleted in acute lymphoblastic leukemia. Notably, worm sam-4 mutants display virtually indistinguishable phenotypes from mild unc-104/KIF1A mutants and human diseases are associated with KIF1A. Specifically, motor domain lesions (A255V and R350G) in human KIF1A underlie the molecular basis of the rare recessive late onset spastic paraplegia SPG30 [6] and a frameshift mutation in the PH domain underlies a form of hereditary sensory and autonomic neuropathy [5]. Further genetic and biochemical studies of SAM-4 in both invertebrates and vertebrates will be required to define the underlying biochemical mechanisms as well as physiological inputs that modulate SAM-4 action in regulating axonal transport.
C. elegans animals were maintained using standard methods [62]. All strains used except for those used for SNP mapping were derivatives of the Bristol N2 wild type background. Animals were grown at the room temperature (22.5°C), unless specified. Strains used are listed in Table S1. The genotype of strains was confirmed by PCR using oligonucleotides listed in Table S2.
jsIs1238 II, jsIs1156 IV, jsIs1263 IV, jsIs1188 IV and jsIs1189 IV transgenes were integrated using MosSCI with EG4322 for integration on chromosome II and EG5003 for integration on chromosome IV [45], [46] and confirmed to be single copy by long range PCR amplification. jsIs1073 and jsIs1075 were generated using a bombardment protocol with Cbunc-119 as the integration marker [63].
Genetic three-factor mapping narrowed the sam-4 mutation to an interval between dpy-25 and rol-6 on chromosome II. Single nucleotide polymorphism (SNP) mapping was used to position sam-4 with CB4856 as a reference strain and narrowed the mutation down to a 163 kb region on Chromosome II between the SNPs CE2-141 and pkP2147. This region is covered by 5 fosmids. Using germline transformation rescue tests, we further mapped sam-4 down to the fosmids WRM0610dH02 and WRM0632aA08. The sam-4(js415) lesion was identified by candidate gene sequencing in this region. A C>T nucleotide change was detected in the second exon of the predicted gene F59E12.11. sam-4 defects are fully rescued by a transgene expressing the hypothetic F59E12.11 gene, which is predicted to encode a 240 amino acid protein. These data identify F59E12.11 as sam-4.
unc-104(js901) was isolated in a non-clonal forward screen for mutations that mislocalized RBF-1-GFP (jsIs423). L4 jsIs423 animals were mutagenized using 50 mM ethyl methanesulfonate (EMS) for 4 hrs and placed on E. coli seeded agar plates. F2 animals derived from these animals were screened for mislocalization of GFP from the nerve ring to cell bodies surrounding the nerve ring. js901 was mapped to chromosome II by classical genetic mapping strategy, and tested for non-complementation with unc-104(e1265). The entire coding sequence of unc-104 was sequenced in js901 revealing a GGA to GTA that changes Gly1465 to Val. This lesion resides within the PH domain of UNC-104.
unc-104(js1288) and unc-104(js1289) were isolated in a screen for suppressors of the PLM synaptic varicosity phenotype of sam-4. N-ethyl-N-nitrosourea (ENU) mutagenesis was performed using standard methodology [64]. Briefly, sam-4(js415) animals were treated with 0.6 mM ENU for 4 hours at the room temperature. Treated animals (P0) were transferred to fresh food (10 L4s or young adults per 100 mm plate). P0 animals were removed from the plates 24–48 hours later. F1 animals were counted 2–3 days later to estimate number of mutagenized chromosomes screened. The F2 animals were screened for increased GFP-RAB-3 signal in PLM varicosities using a fluorescent dissecting microscope. The suppressors displayed tight linkage to sam-4. Phenotypic analysis of sam-4(js415) js1288/sam-4(js415) and sam-4(js415) js1289/sam-4(js415) revealed both were semi-dominant suppressors of the Sam phenotype. Sequencing of the sam-4 coding region revealed no mutation in sam-4 gene of these isolates. js1289 was mapped to between sam-4 and rol-6 by three factor mapping, crossing lin-31(n301) sam-4(js415) js1289 rol-6(187); jsIs821 to CB4856 and screening for Lin Sam non-Rol and Lin Sam Sup (suppressor) non-Rol recombinants 18 of 21 recombinants had recombination events between sam-4 and js1289 and 3 between js1289 and rol-6. 100 bp paired-end whole genome sequencing of homozygous strains was conducted at Oklahoma Medical Research Foundation. The data were analyzed using Whole Genomes, a web-based alignment and analysis program, and revealed lesions in unc-104: An Asp177 to Ala (GAC to GCC) lesion in js1288 and a Ser211 to Ala (TCA to GCA) lesion in js1289.
Plasmid DNA clones were constructed using standard molecular biology techniques.
Transgenic animals were imaged using epi-fluorescence on an Olympus BX60 equipped with an X-CITE120 mercury lamp (EXFO) using standard GFP and RFP filter sets. Images were taken with a Retiga EXi CCD camera using OpenLab software and processed using Adobe Photoshop.
jsIs1238 II, jsIs1156 IV, jsIs1263 IV, jsIs1188 IV and jsIs1189 IV transgenes were integrated using MosSCI with EG4322 for integration on chromosome II and EG5003 for integration on chromosome IV [45] with modifications. These transgenic lines were confirmed to be single-copy integration events by long range PCR (Details available online: http://thalamus.wustl.edu/nonetlab/ResourcesF/Resources.html). jsIs1073 and jsIs1075 lines were generated by integrating NM2057 and NM2066 using a bombardment protocol with Cbunc-119 as the integration marker [63].
Animals were assayed at the room temperature on NGM agar. L4 animals (or as indicated) were transferred to a bacteria-free plate to allow them clear off bacteria (2–3 min). Subsequently, these animals were transferred to another bacteria-free plate and imaged immediately for 10–20 sec. Animal movements were recorded using LG-3 frame grabber run by ScionImage software at 1 frame/sec for 40 images total. These recordings were then analyzed using wormtracker plus [66]. Only animals in the imaging field>14 consecutive frames recorded were used in the velocity analyses.
L4 animals on bacterial lawns of OP50 on NGM agar were scored at the room temperature. Pharyngeal pumping rates was determined by counting contractions of the terminal bulb for 1 minute per animal.
Immunohistochemistry and western blots were performed as previously described [67], [68]. For FLAG immunohistochemistry staining, animals were grown at room temperature and fixed in methanol/acetone. For SAM-4-3XFlag fractionation, 0.5 mM EGTA and 0.5 mMEDTA were added in the fractionation buffer as indicated in Figure S6. Mouse anti-Flag (1∶200, Sigma, Cat. A8592) primary antibody incubations were performed overnight at 4°C. Alexa conjugated secondary antibodies (Invitrogen) were incubated for 2 hrs at room temperature at 1∶500. Antibody used for western blots: anti-FLAG (1∶1000); anti-β-tubulin (1∶1000, E7, Developmental Studies Hybridoma Bank, Iowa city), anti-UNC-104 (1∶40) [51].
For young adult animals (used in Figure 5), hermaphrodites were immobilized with 3–5 mM levamisole (Sigma-Aldrich) in M9 buffer and mounted on a 2% agarose pad. Time-lapse images of GFP-RAB-3 were acquired and analyzed as described before [51]. The numbers of moving particles in a 15–20 µm region at a distance of 15–25 µm away from the PLM soma were used for flux calculations.
For mid-L1 staged animals (20–24 hrs after hatch, used in Figure 8), we used an anesthetic-free protocol to image GFP-RAB-3 [69]. Specifically, animals were immobilized in 0.5 µl of 0.10 microspheres (Cat# 00876, Polysciences, Inc.) on 10% agarose pads. Time-lapse imaging was acquired using 100×/1.30 oil objective on a Axioskop (Zeiss) equipped with ASI piezo XYZ-motorized stage, Sutter instruments high speed electronic filter wheels and shutters, and a Hamamatsu Orca-R2 cooled CCD camera all controlled by Volocity software (PerkinElmer Inc.). Time lapse images were acquired for 40 seconds at a speed of 5 frames per second with an exposure time of 200 ms. Particle dynamics were analyzed with Volocity software. Total moving particles were counted in the 35 µm region at a distance of 20 µm away from the PLM soma.
To record GFP-RAB-3 co-movements with SAM-4-TagRFP, confocal images were collected with a Hamamatsu Flash 4.0 CMOS camera attached to a Yokogawa Spinning Disc Confocal apparatus on an Olympus IX73 inverted microscope. 0.33 sec Green and Red channels exposure were taken consecutively, and captured at 1 sec intervals, and image series were assembled into movies using Micro-Manager software (available at micro-manager.org).
P values were determined using GraphPad Prism. Multi-group data sets were analyzed by a one-way ANOVA with post-hoc Holm-Sidak's test for multiple comparisons. A t-test was used for paired data sets.
|
10.1371/journal.ppat.1002215 | Discovery of the First Insect Nidovirus, a Missing Evolutionary Link in the Emergence of the Largest RNA Virus Genomes | Nidoviruses with large genomes (26.3–31.7 kb; ‘large nidoviruses’), including Coronaviridae and Roniviridae, are the most complex positive-sense single-stranded RNA (ssRNA+) viruses. Based on genome size, they are far separated from all other ssRNA+ viruses (below 19.6 kb), including the distantly related Arteriviridae (12.7–15.7 kb; ‘small nidoviruses’). Exceptionally for ssRNA+ viruses, large nidoviruses encode a 3′-5′exoribonuclease (ExoN) that was implicated in controlling RNA replication fidelity. Its acquisition may have given rise to the ancestor of large nidoviruses, a hypothesis for which we here provide evolutionary support using comparative genomics involving the newly discovered first insect-borne nidovirus. This Nam Dinh virus (NDiV), named after a Vietnamese province, was isolated from mosquitoes and is yet to be linked to any pathology. The genome of this enveloped 60–80 nm virus is 20,192 nt and has a nidovirus-like polycistronic organization including two large, partially overlapping open reading frames (ORF) 1a and 1b followed by several smaller 3′-proximal ORFs. Peptide sequencing assigned three virion proteins to ORFs 2a, 2b, and 3, which are expressed from two 3′-coterminal subgenomic RNAs. The NDiV ORF1a/ORF1b frameshifting signal and various replicative proteins were tentatively mapped to canonical positions in the nidovirus genome. They include six nidovirus-wide conserved replicase domains, as well as the ExoN and 2′-O-methyltransferase that are specific to large nidoviruses. NDiV ORF1b also encodes a putative N7-methyltransferase, identified in a subset of large nidoviruses, but not the uridylate-specific endonuclease that – in deviation from the current paradigm - is present exclusively in the currently known vertebrate nidoviruses. Rooted phylogenetic inference by Bayesian and Maximum Likelihood methods indicates that NDiV clusters with roniviruses and that its branch diverged from large nidoviruses early after they split from small nidoviruses. Together these characteristics identify NDiV as the prototype of a new nidovirus family and a missing link in the transition from small to large nidoviruses.
| Research in virology is driven towards the characterization of a limited number of socioeconomically important pathogens, mostly those infecting humans. Yet, characterization of other viruses may advance our understanding of these topical pathogens and the fundamentals of virology. Here we describe the discovery of a virus of unknown clinical relevance that has many remarkable features. The virus was coined Nam Dinh virus (NDiV) after a Vietnamese province. It is a mosquito-borne virus with a 20.2 kilobase genome, the largest among non-segmented single-stranded RNA viruses of insects. Employing bioinformatics tools, we show that NDiV prototypes a new family and is a missing evolutionary link connecting the distantly related nidoviruses with small and large genomes, including important and diverse pathogens such as porcine respiratory and reproductive syndrome virus (∼15-kilobase genome) and SARS coronavirus (∼30 kilobases), respectively. NDiV and large nidoviruses form a phylogenetic cluster and share a set of core replicative enzymes. They exclusively encode an exoribonuclease that presumably controls replication fidelity. Its acquisition may have promoted the emergence of viruses with single-stranded RNA genomes larger than ∼20 kilobases. This study highlights the benefits of broad virus discovery efforts for fundamental and applied research.
| Viruses employing positive-sense, single-stranded RNA genomes (ssRNA+) form the most abundant class and its members are known to infect all types of hosts except Archaea. They have evolved genome sizes in the range of ∼3.0 to 31.6 kb (Fig. 1). This size range is the largest among those of the different classes of RNA viruses, although it is small compared to those of DNA viruses and cellular organisms. These profound genome size differences between RNA and DNA life forms are inversely correlated with mutation rates, which are highest in RNA viruses, thought due to the lack of proofreading during replication [1]–[3].
Recently, the molecular basis of the relation between RNA virus genome sizes and mutation rates has been revisited in studies of nidoviruses with large genomes (“large nidoviruses”). These viruses, with genomes of 26.3 to 31.6 kb, include the Coronaviridae and Roniviridae families and are at the upper end of the RNA virus genome size range [4]. They are uniquely separated from other ssRNA+ viruses (3.0–19.6 kb genomes), including the distantly related Arteriviridae family (12.7–15.7 kb genomes; “small nidoviruses”) with which they form the order Nidovirales [4]–[6]. The order includes five major lineages of viruses that infect vertebrate and invertebrate hosts. Their complex genetic architecture includes multiple open reading frames (ORFs) that are expressed by region-specific mechanisms. The first two regions are formed by the two 5′-most and partially overlapping ORFs, ORF1a and ORF1b, which are translated from the genomic RNA to produce polyproteins 1a (pp1a) and pp1ab. The expression level of ORF1b is downregulated relative to that of ORF1a by the use of the ORF1a/1b ribosomal frameshifting signal [7], [8]. Both pp1a and pp1ab are autoproteolytically processed by ORF1a-encoded proteases to yield numerous products that control genome expression and replication [9]. The third, 3′-located region of the nidovirus genome includes multiple smaller ORFs (3′ORFs), although the number of these ORFs varies considerably among nidoviruses. These genes are expressed from 3′-coterminal subgenomic mRNAs to produce the structural proteins incorporated into the enveloped nidovirus particles and, optionally, other proteins modulating virus-host interactions [10]–[13]. With the exception of a few nidoviruses, the subgenomic and genomic mRNAs are also 5′-coterminal. A mechanism of discontinuous negative-stranded RNA synthesis, yielding the templates for subgenomic mRNA production, is thought to control this mosaic structure of nidovirus mRNAs. The synthesis of subgenome-length negative stranded RNAs is guided by short transcription-regulating sequences (TRSs) – located in the common “leader sequence” (near the genomic 5′ end) and in each “mRNA body” (upstream of the expressed ORFs) - that share a conserved core sequence and flank the genome region that is not present in the respective subgenomic mRNAs.
The nidovirus ORF1b encodes key replicative enzymes whose number and type vary between the major nidovirus lineages. They invariably include an RNA-dependent RNA polymerase (RdRp) and a superfamily 1 helicase (HEL1) [14], which are most common in other RNA viruses, and several other RNA-processing enzymes that are either unique to nidoviruses (uridylate-specific endonuclease (NendoU) and 3′-to-5′exoribonuclease (ExoN)) or rarely found outside nidoviruses (2′-O-methyltransferase (OMT); [4]). Among these enzymes, the ExoN domain has properties that are most relevant for understanding the relation between genome size and mutation rate in RNA viruses.
Bioinformatics-based analysis originally identified the ExoN domain only in the genomes of large nidoviruses and mapped it in the vicinity of HEL1, a key replicative enzyme [15]. It also revealed a distant relationship between ExoN and a cellular DNA-proofreading enzyme. Based on these observations, nidoviruses were proposed to have acquired ExoN to control the replication fidelity of their expanding genome [15]. The enzymatic activities of ExoN were subsequently verified and detailed in biochemical studies [16], [17]. Likewise, and in line with the expectations, ExoN-inactivating mutations were shown to decrease RNA replication fidelity by ∼15–20 fold in two coronaviruses, mouse hepatitis virus (MHV) and SARS coronavirus (SARS-CoV), while only modestly affecting virus viability [18], [19]. These results strongly support a critical role of ExoN in the control of replication fidelity of large nidoviruses, although more mechanistic insight is clearly required before the current paradigm connecting RNA virus mutation rates and genome size control could be definitively revised to include proof-reading during the replication of large RNA genomes [20].
Major advancements toward this goal are expected to come from studies of the structure and function of ExoN, which aim to elucidate the molecular mechanism of its action. In addition, genomics studies could contribute to this quest by providing insights into the role of ExoN in RNA virus evolution. Accordingly, if ExoN was acquired to ensure the expansion of RNA genomes beyond a certain size, we may expect (i) a genome size threshold that separates RNA viruses with and without ExoN; (ii) all nidoviruses with genome sizes above this threshold to encode ExoN; and (iii) no other domain than ExoN to correlate, functionally and phyletically, with genome size control in large nidoviruses.
In this respect, the characterization of nidoviruses with a genome size in the gap that currently separates small and large nidoviruses should, in theory, be particularly insightful. However, whether these viruses actually exist has thus far remained an open question. Three considerations suggest that if nidoviruses with intermediate-sized genomes ever evolved they may already have gone extinct. First, it is recognized that the evolution of RNA viruses is characterized by a high birth-death rate and the extinction of numerous virus lineages, resulting in the fast turnover of species [21]. Secondly, the genome size gap between large nidoviruses and all other known ssRNA+ viruses has existed without exception since genome sequencing began in the 1980s. As of the late 1980s, this gap has been bordered by closteroviruses (from the bottom) and nidoviruses (from the top) (Fig. 1). Likewise and thirdly, all nidovirus genomes sequenced to date have sizes that are similar to either IBV (27,600 nt) [22] or EAV (12,700 nt) [23], which were the first fully sequenced coronavirus and arterivirus genomes, respectively. The evident under-representation of RNA viruses with relatively large genomes is even more striking in the light of the continuous flow of newly identified ssRNA+ viruses with smaller genome sizes [24] (Fig. 1).
In sharp contrast to these considerations and prior observations, we here report the discovery of a nidovirus with a genome size that is intermediate between those of small and large nidoviruses. This elusive and precious evolutionary link is an insect-borne virus with the largest ssRNA+ genome for any insect virus known to date. Comparative genome analyses involving this newly identified virus provide evolutionary evidence for the acquisition of the ExoN domain by a nidovirus (ancestor) with a genome size in the range of ∼16–20 kb. This range appears to define the size limit for the expansion of ssRNA+ virus genomes, which may be achieved in evolution without the recruitment of a specialized enzyme that controls replication fidelity. Furthermore, we found that two other replicative enzymes, N7-methyltransferase (NMT) and NendoU, are not encoded by toroviruses and invertebrate nidoviruses, respectively, indicating that they may contribute “optional activities” for the nidovirus replication machinery. Together our results highlight the broad benefits of virus discovery efforts applied to mosquitoes.
In Vietnam, between 2,000 and 3,000 cases of acute encephalitis syndrome (AES) are reported annually, of which about 40% are confirmed to be associated with Japanese encephalitis virus (JEV). The etiological agent(s) in the other 60% of cases remains unknown [25], but they share demographic characteristics and seasonality with the JEV cases. Hence, the involvement of other arboviruses in non-JE AES was postulated and the virus described in this paper was identified in search of such pathogens, which may infect both humans and mosquitoes.
During continued JEV surveillance between September 2001 and December 2003, 359 pools containing one of six mosquito species (see Materials and Methods) were collected indoors in Northern and Central Vietnam at one- to three-month intervals. The study areas included Hanoi and other cities located in the provinces of Ha Nam (Chuyenngoan, Mocbac), Ha Tay (Catque, Phuman and Chuongmy), Nam Dinh, and Quang Binh (Fig. 2). The majority of Catque inhabitants are farmers who cultivate rice in watered paddy fields and raise pigs. Phuman and Quangbinh, however, are highlands.
Mosquito pools were tested for the presence of viruses using infection of different cell lines as a read-out assay. Homogenates that were prepared from some pools containing Culex tritaeniorhynchus and Culex gelidus induced cytopathic effects in the C6/36 mosquito cell line. Most of these were attributed to JEV (24 different strains; data not shown), but for 10 specimens a routine laboratory screening for JEV and other circulating flaviviruses (such as Dengue and West Nile viruses) by RT-PCR and/or serology yielded negative results.
Subsequently, infected culture fluid (ICF) from cells infected with unknown agents were analyzed by electron microscopy, which revealed an enveloped virus with a diameter of 60–80 nm (Fig. 3A). This virus was named Nam Dinh virus (NDiV), after the geographic locality of its first apparent isolation, although this origin could not be confirmed later on. However, for historical reasons, this name was retained for all subsequent isolates, and the analysis of one of those (02VN178) is described here. NDiV was identified in four mosquito pools, two from Culex vishnui and two from Culex tritaeniorhynchus, collected in two other provinces of Vietnam (Table 1). PCR amplification using virus-specific primers to an ORF1b region (see below) was employed to verify the presence of NDiV in the mosquito samples, but to date no other insects have been probed for the presence of the virus. It also remains to be investigated whether NDiV causes disease in susceptible hosts and whether it may infect humans.
Purified NDiV was used for virion protein analysis (Fig. 3B) and genome sequencing (Fig. S1; Materials and Methods). In silico translation of the unsegmented, 20,192 nt-long NDiV genome (GenBank accession number DQ458789) indicated that it contained at least six ORFs: ORF1a (nt 361–7869), ORF1b (7830–15635), ORF2a (15660–18356), ORF2b (15674–16309), ORF3 (18402–18875) and ORF4 (18754–19101) (Fig. 3D). The region encompassing ORFs 3 and 4 also contains a few smaller potential ORFs. The coding region of the genome is flanked by a 5′-untranslated region (UTR) (1–360) and a 3′-UTR (19102–20192), with the latter being followed by a poly(A) tail. The 5′-UTR includes two AUG codons indicating that translation initiation for ORF1a/ORF1b is likely mediated by another mechanism than ribosomal scanning. Three pairs of ORFs (1a–1b, 2a–2b, and 3–4) overlap to variable degrees; particularly, ORF1b overlaps ORF1a in the −1 frame (Fig. 3D; see also below). Overall, these results showed that NDiV is an insect-borne ssRNA+ virus with the largest genome known so far - twice the size of the next largest one, which is the genome of the Iflavirus Brevicoryne brassicae picorna-like virus [26] (Fig. 1).
The NDiV genome organization most closely resembles that of nidoviruses, the only group of ssRNA+ viruses that includes representatives with genomes larger than that of NDiV. This putative relationship was subsequently verified in experimental and bioinformatics analyses of the function and expression of the 3′-ORFs region and in bioinformatics analyses of ORF1a and ORF1b, as described below. The latter studies also provided insights into the evolution and molecular biology of other nidoviruses.
Three virion proteins, p2a, p2b, and p3, were assigned to ORFs 2a, 2b, and 3, respectively, by peptide sequencing analysis (Fig. 3D). No significant similarity was found between these ORFs of NDiV and proteins of other origin in BLAST-mediated searches [27]. The p2b protein is highly hydrophilic and enriched with proline (7.5%) and acidic residues (17.8%), and – relative to other virion proteins – with basic residues (7.9%) making it a potential nucleocapsid (N) protein. The p2a and p3 proteins, and the putative protein encoded in ORF4 (p4) contain, respectively, six, two, and two stretches of hydrophobic residues indicative of transmembrane helices (Fig. S2). These proteins also include, respectively, twelve, two, and three potential N-linked glycosylation signals (NXS/T), and fifteen, six, and four cysteine residues that might form disulfide bridges at locations flanked by hydrophobic regions. These characteristics are typical for glycoproteins of other RNA viruses. Based on size considerations, the largest protein, p2a, might be an equivalent of the spike (S) protein, while p3 and/or p4 might be a smaller glycoprotein and an equivalent of the membrane (M) protein of nidoviruses.
We also asked whether NDiV resembles other nidoviruses in using subgenomic mRNAs for expressing the 3′-end ORFs located downstream of ORF1b. First, we attempted to identify potential TRS motifs in the viral genome sequence, which were expected to reside in the 5′-UTR as well as in the regions immediately upstream of ORF2a, 3, and 4. Although no common repeats larger than six nucleotides were identified in these four areas, we noticed the presence of two pairs of near-perfect repeats: the first pair located in the 5′-UTR (nt 26–40 of the genome) and the region upstream of ORF3 (14 out of 15 residues are identical), and the second pair encompassing nt 125–137 of the 5′-UTR and a sequence immediately upstream of ORF2a/2b (12 out of 13 residues are identical) (Fig. 3D). The two pairs share from ∼43 to 52% pair-wise sequence identity in an alignment containing a single gap (Fig. 3D), and no other repeats of comparable or larger size were found in the analyzed areas. The locations and sizes of these repeats suggest they are TRS signals, although no candidate TRS was identified immediately upstream of ORF4; to our knowledge, the use of two alternative leader TRSs has not been observed in other nidoviruses thus far. These observations suggested that NDiV uses at least two subgenomic mRNAs for the expression of the 3′-located ORFs and that these mRNAs have 5′-terminal sequences of different size in common with the viral genome.
To verify this model, we used a P32-labelled probe complementary to the 3′-end of the NDiV genome in a Northern blot hybridization with total RNA isolated from NDiV–infected C6/36 cells (see Materials and Methods; Table S2, and Fig. 3C). This analysis revealed three prominent RNA species with apparent sizes of about 20, 4.5, and 1.8 kb, which match those expected for the genomic RNA and two subgenomic RNAs, mRNA2 (to express ORF2a and ORF2b) and mRNA3 (for ORF3 and possibly ORF4), respectively. We also observed a set of less abundant bands in the 0.9–1.1-kb size range, whose origin(s) and relevance remain to be established.
Nidoviral ORF1a/ORF1b −1 ribosomal frameshifting (RFS) is controlled by a “slippery sequence” and a stem-loop or pseudoknot RNA structure immediately downstream [7]. RFS is conserved in nidoviruses and this property is widely used for computational mapping of its determinants in newly sequenced genomes. We followed this approach to map potential RFS signals in the NDiV genome (Fig. 4). The 40-nt NDiV ORF1a/ORF1b overlap region was found to have the best match (GGAUUUU) with the slippery sequence (AAAUUUU) of invertebrate roniviruses [28], which deviates considerably from the pattern (XXXYYYZ) conserved in vertebrate nidoviruses (Fig. 4A). No appreciated similarity with the latter motif was found in the NDiV ORF1a/ORF1b overlap region. The distances separating the NDiV putative RFS from the termination codons flanking the ORF1a/ORF1b overlap are within the range found in large nidoviruses, while being out of the distance range to the ORF1a stop codon of small nidoviruses (Fig. 4B). According to the analysis of a 190-nt sequence - which starts within the NDiV ORF1a/ORF1b overlap - with Mfold [29] and pknotsRG [30], the predicted slippery sequence is followed by a complex stem-loop structure; no pseudoknots, unless forced, are predicted in this region (Fig. 4C). The slippery sequence, distance to the downstream RNA secondary structure, and predicted fold resemble those of Red clover necrotic mosaic virus (RCNMV), a ssRNA+ plant virus of the family Tombusviridae [31], [32] (Fig. 4C–D). These results identified the critical elements of the putative NDiV RFS as being most unique among those described for members of the order Nidovirales.
Nidoviruses are distinguished from other RNA viruses by a constellation of 7 conserved domains having the order TM2-3CLpro-TM3-RdRp-Zm-HEL1-NendoU, with the first three being encoded in ORF1a and the remaining four in ORF1b. TM2 and TM3 are transmembrane domains, Zm is a Zn-cluster binding domain fused with HEL1, and 3CLpro is a 3C-like protease [4] (however see below). Since NDiV was found to be very distantly related to the other nidoviruses known to date, sequence-based functional characterization presented a considerable technical challenge. In comparative sequence analysis, profile-based methods that employ multiple sequence alignments are known to achieve the best signal-to-noise ratios [27], [33], [34]. They have been the methods of choice for establishing remote relations in biology, also in our prior studies of nidoviruses [15], [35]–[37]. In this study we used profile vs. sequence and profile vs. profile searches as implemented in HMMer and HHsearch, respectively, for general comparisons. To prepare profiles, we selected representatives of small and large nidoviruses, and also three subsets of large nidoviruses (coronaviruses, toro/bafiniviruses, and roniviruses). Using profile-based searches we identified counterparts (orthologs) of nidovirus-wide conserved enzymatic domains in the NDiV pp1ab. For the identification of TM2 and TM3, predictions of transmembrane helices by TMpred were used.
Six out of the seven nidovirus-wide conserved protein domains, TM2-3CLpro-TM3-RdRp-Zm-HEL1, were mapped in the canonical position and order in the NDiV ORF1a/1b sequence (Table 2). Three of these putative NDiV domains, 3CLpro [9], [38], RdRp [39], and HEL1 [40] are enzymes conserved in all nidoviruses [14]. They have counterparts of all invariant and highly conserved residues implicated in catalysis in other nidoviruses, a finding indicative of the functionality of these proteins in NDiV.
Like its orthologs in corona- and roniviruses, the NDiV 3CLpro is predicted to employ a catalytic His-Cys dyad. Its substrate-binding site is predicted to include a conserved His residue which was implicated in controlling the P1 specificity for Glu/Gln residues in other viruses, a hallmark of 3C/3CLpros [41]. Surprisingly, despite this finding, no candidate cleavage sites with the characteristic 3CLpro-specific signatures could be identified in the NDiV pp1a/1ab. Consequently, the sizes of all NDiV replicative domains described in this paper (Table 2) are based on the hit sizes in profile searches and are subject to future refinement. Collectively, these results strongly indicate that NDiV encodes all nidovirus-wide conserved replicase domains except for NendoU (Figure 3D; see also below), thus supporting the classification of NDiV as a nidovirus.
All large nidoviruses express an ExoN [16] of the DEDD superfamily, which is not found in other ssRNA+ viruses, and an OMT [42], [43] of the RrmJ family, that is not present in arteriviruses [15]. The presence of these domains therefore discriminates large from small nidoviruses. Using profile searches in the ORF1b-encoded part of pp1ab, homologs of these two enzymes were identified in the NDiV genome (Table 2). Using an ExoN multiple sequence alignment of NDiV and large nidoviruses, the conserved motifs I, II, and III, including the catalytic residues (two Asp and one Glu), as well as the ExoN-specific Zn-finger module were identified in the NDiV ortholog (Fig. 5A). Furthermore, the NDiV ExoN shows an insertion whose size and position correspond to those of the second Zn-finger-like module that is exclusively found in roniviruses. However, unlike the ronivirus domain, NDiV appears to lack His/Cys residues potentially involved in Zn-binding. According to a multiple sequence alignment of nidovirus OMTs (Fig. 5B), the putative NDiV OMT contains motifs X, IV, VI and VIII, encompassing residues of the catalytic KDKE tetrad, as well as motif I involved in binding of the methyl donor [42]. These data imply that NDiV ORF1b encodes functional ExoN and OMT domains (Fig. 3D), which are both typical of large nidoviruses.
NDiV ORF1b includes a ∼750-nt region that is flanked by the upstream ExoN and downstream OMT domains and was expected to encode a NendoU domain [44]–[47], given its presence at this locus in all nidoviruses known so far [15], [48]. Surprisingly, however, profile searches of nidovirus NendoUs revealed no significant hits in the corresponding region of the NDiV sequence (E-values>9.5). This observation prompted us to re-examine the NendoU assignment in other nidoviruses, including the invertebrate roniviruses [15]. Using profile-sequence and profile-profile comparisons mediated by HMMer and HHsearch, respectively, NendoU counterparts were readily identified in all corona-, toro/bafini-, and arteriviruses (E-values<10−4), but not in roniviruses (E-values>4.5). We therefore conclude that, unlike other (vertebrate) nidoviruses, the invertebrate NDiV and roniviruses do not encode a NendoU domain (Fig. 3D).
We proceeded to analyze this genomic region flanked by ExoN and OMT in invertebrate nidoviruses in more detail. First, using a ronivirus profile vs. NDiV pp1ab sequence comparison, we found that these domains are moderately similar to each other (E-value = 0.18), suggesting a weak conservation of a common function in these newly recognized orthologous domains of NDiV and roniviruses. Their alignment was converted into a profile with which we screened all domains of our in-house nidovirus profile database (see Materials and Methods). Remarkably, the only significant hit (E-value<10−4) was recorded against the coronavirus NMT profile (Table 2). For comparison, its similarities with NendoU profiles of corona-, toro/bafini- or arteriviruses were not significant (E-value>1.5). These data indicate that NDiV and roniviruses may encode an NMT domain that is flanked by ExoN and OMT.
The coronavirus NMT domain was originally mapped to the C-terminal half of nsp14 [43], [49]. The corresponding domain in toro/bafiniviruses has a much smaller size (80 aa vs. 200 aa). According to our analysis, it has no significant similarity with the NMT of coronaviruses, or the newly recognized putative NMT of roniviruses and NDiV. Based on these observations, we generated an alignment of the NMT domains of corona- and roniviruses and NDiV (Fig. 5C) in order to search for remote cellular homologs. The N-terminal part of the nidovirus NMT includes a conserved methyl donor binding site (motif I), according to the prior assignment for coronavirus NMTs. In line with this observation, a weak hit between nidovirus NMTs and a cellular guanine N7-methyltransferase involving the motif I region was detected in this study. In their C-terminal part, nidovirus NMTs uniquely include four conserved Cys/His residues indicative of a Zn-binding site that may be part of a separate domain (Fig. 5C).
Collectively these results established a mosaic domain relationship in the pp1ab area flanked by ExoN and OMT domains for large nidoviruses and NDiV. In this genomic region coronaviruses encode both NMT and NendoU domains, while other viruses encode either NendoU (toro/bafiniviruses) or NMT (roniviruses and NDiV).
Next, we proceeded to determine the phylogenetic position of NDiV among nidoviruses. The phylogeny was inferred using Bayesian posterior probability trees for a concatenated alignment of three enzymes, 3CLpro, RdRp, and HEL1, that are conserved in all nidoviruses (see Materials and Methods). In line with the current nidovirus taxonomy and genomic data [28], [48], [50], [51], this analysis consistently identified the four known major lineages (arteri-, roni-, corona-, and toro/bafiniviruses), as well as a new one represented by NDiV, as the most deeply rooted branches. Our initial attempts to resolve the relationship among the five lineages produced uncertain results. To address this challenge, we adopted a step-wise approach starting from the analysis of close intra-group relationships in the most abundantly sampled subfamily, Coronavirinae, and the family Arteriviridae, and finishing with an analysis of the most distant inter-(sub)family relationships between the five major lineages. Prior to the nidovirus-wide phylogenetic analysis, the affinity of arteri-, roni-, and toro/bafiniviruses to the subfamily Coronavirinae was evaluated through a profile-based analysis involving conserved domains (see Supplementary Text S1 and Table S1). The obtained results confirmed that the strongest sequence affinity exists between corona- and toro/bafiniviruses, which was evident for the 6 out 8 domains that are conserved between coronaviruses and one or more of the other lineages. The HEL1 was the only domain for which an alternative strongest affinity – between corona- and roniviruses – was documented.
The affinity established above was incorporated as prior knowledge in the nidovirus-wide phylogenetic analysis in order to improve the resolution of the most distant relationships. Accordingly, two alternative reconstructions were conducted with the clustering of toro/bafiniviruses and coronaviruses being either fixed or not. When the clustering was not fixed, roniviruses were found to be closest to coronaviruses (Fig. 6A). This topology indicated that the HEL1 sequence affinity dominated over that of the RdRp (Table S1) in the concatenated 3CLpro-RdRp-HEL1 alignment. An alternative nidovirus phylogeny was inferred when the clustering of coronaviruses and toro/bafiniviruses was fixed prior to the inference (Fig. 6B). Importantly, in both trees, NDiV was consistently albeit relatively distantly clustered with roniviruses, indicating that this grouping does not depend on the choice of tree-building parameters and is likely genuine.
To infer the direction of nidovirus evolution, we sought to root the nidovirus phylogeny using an outgroup approach. Neither other viruses nor cellular organisms encode the domain constellation that is conserved in nidoviruses, precluding an expansion of the original nidovirus dataset with outgroup sequences to root the tree. This prompted us to split the domain constellation and perform separate analyses of the evolution of the two most conserved nidovirus protein domains, RdRp and HEL1, which are also among the most conserved in ssRNA+ viruses (Fig. 6C–D). Prior to the analysis, major clades comprising coronaviruses, toro-/bafiniviruses, roniviruses, and arteriviruses, and an outgroup were each fixed to be monophyletic.
For the HEL1 tree (Fig. 6C), the part of the alignment covering the most conserved region from motif I to motif VI (see [52]) was used. Representatives of rubiviruses, betatetraviruses, omegatetraviruses, and hepeviruses were used as an outgroup. The resulting topology closely resembles that of the relaxed nidovirus phylogeny (Fig. 6A), in which vertebrate coronaviruses and invertebrate nidoviruses are sister clades, thus confirming that it is dominated by the HEL1-related component.
For the RdRp tree (Fig. 6D), an alignment of the most conserved RdRp region delimited by motifs G and E (see [53]) was used. Representatives of three divergent picornaviruses (an enterovirus, a parechovirus, and a hepatovirus) were used as an outgroup. The resulting topology matches that of the constrained nidovirus phylogeny (Fig. 6B), in which the grouping of corona- and toro-/bafiniviruses was forced, and could thus be considered RdRp-like.
Despite somewhat incongruent topologies in the two protein-specific phylogenies, in both cases the outgroups are consistently placed at the branch leading to arteriviruses, thus separating small- from large- and intermediate-size viruses in nidovirus evolution. The support for the positioning of the outgroups in the RdRp and HEL1 trees by Bayesian/ML estimates (0.69/522 and 0.48/990, respectively) is relatively low and/or varied in analyses by two methods, possibly due to the very large evolutionary distances separating the major virus groups, including the outgroups. We used the rooting on the arterivirus branch to root the nidovirus tree that was inferred using a concatenated alignment of three domains (Fig. 6A–B).
According to this analysis, small nidoviruses are separated from other nidoviruses, and NDiV is monophyletic with roniviruses in a separate clade of invertebrate nidoviruses, which clusters with the group formed by corona- and toro/bafiniviruses. NDiV and roniviruses are separated by a large evolutionary distance indicating that NDiV likely is the prototype of a separate family. The topology of the tree in Fig. 6B is compatible with a scenario in which genome size change during nidovirus evolution was dominated by expansion, with contemporary nidoviruses representing different stages in the transition from small to large ssRNA+ genomes.
We describe the discovery of an insect-borne ssRNA+ virus, called NDiV, possessing a genome organization, virion properties, mRNAs, and putative proteome characteristics that place it in the order Nidovirales. In phylogenetic and protein domain analyses NDiV consistently, albeit relatively distantly, clustered with viruses of the family Roniviridae, which seems to make sense biologically given that both infect invertebrate hosts. Although the NDiV classification as the first insect nidovirus is beyond doubt, its characterization was only just initiated in this study. NDiV is likely to possess unique properties concerning, for example, the leader-body junctions of its sg mRNAs and the cleavage sites recognized by its 3CLpro, which both require further characterization.
The principal biological significance of the discovery of NDiV is in the intermediate position this virus occupies between small and large nidoviruses in the genome size distribution observed for ssRNA+ viruses. Prior to this study, the existence of currently circulating nidoviruses with genome sizes within this gap was even highly uncertain (see Introduction). Together small and large nidoviruses cover the upper ∼19 kb (∼66%) of the entire ssRNA+ genome size range and are separated by ∼10 kb (32%). The very existence of NDiV validates the previously established evolutionary relationship between the remotely related arteriviruses and coronaviruses that have very different genome sizes [23]. Characterization of arteri- and coronaviruses by comparative genomics has been instrumental in defining the common and unique features of members of the order Nidovirales [14], and has guided the delineation of potential targets for antiviral drug design [54].
The inclusion of NiDV in this analysis yields additional and novel insights with implications for nidoviruses and other RNA viruses at large. It allowed us to revise and expand the assignment for two replicative enzymes of nidoviruses – NendoU and NMT. Prior to this study, the former was considered to be a genetic marker of nidoviruses [15]. Still, its (universal) function in the replication cycle of (vertebrate) nidoviruses has remained enigmatic, despite steady progress in the biochemical, structural, and genetic characterization of this enzyme in arteri- and coronaviruses [44]–[47], [55]–[60]. Our analysis showed that invertebrate roniviruses and NDiV do not encode a NendoU domain implying that, contrary to the current paradigm, the utilization of this enzyme in replication may be restricted by the host organism. Surprisingly, and in contrast to the case of NendoU, invertebrate nidoviruses were found to encode a putative NMT, whose ortholog was previously identified in SARS-CoV and shown to be conserved in the subfamily Coronavirinae [43], [49]. Our observation indicates that certain aspect(s) of the nidovirus replicative cycle that are controlled by the NMT domain could be similar in coronaviruses and invertebrate nidoviruses, but not toro/bafiniviruses which are otherwise closer to coronaviruses. Collectively, our insights into the phyletic distribution of NendoU and NMT reveal a modularity of some of the major subunits of the replication apparatus in large nidoviruses, which must be rationalized in future mechanistic studies and taken into account in drug development efforts.
Although the NDiV genome size is intermediate between those of small and large nidoviruses, NDiV most closely resembles large nidoviruses in properties that are not universally conserved in the order. Particularly, NDiV does not encode a homolog of the replicative protein of unknown function (nsp12) that is exclusively conserved in arteriviruses [14] and it has a set of three replicative enzymes, OMT, NMT, and ExoN, encoded in large but not in small nidoviruses. These three enzymes are encoded in ORF1b, downstream of the RFS (Fig. 3D and Fig. 4) and in the vicinity of the two key enzymes for RNA synthesis, RdRp and HEL1, with their expression level being downregulated relative to that of the ORF1a-encoded subunits.
Despite these common properties, the two methyltranferases (OMT and NMT) differ from ExoN in their relation to genome size. Particularly, OMTs are known to be also encoded by flaviviruses [61] whose genome size of ∼10 kb is average for RNA viruses, while the NMT domain was found to be lacking in a subset of large nidoviruses represented by toro-/bafiniviruses (this study). Furthermore, an N-methyltransferase function, albeit associated with a domain seemingly unrelated to the NMT domain of nidoviruses, was identified in the large Alphavirus-like supergroup of ssRNA+ viruses, whose members have genome sizes from ∼7,000 to 19.500 nt [62]–[64]. ssRNA+ viruses use methyltransferases to modify the 5′-end of their mRNAs (cap structure), which was recently found to be essential in the control of translation and innate immunity [65], [66]. It is not clear whether the use of methyltransferases may provide particular benefits for genome size control and/or promote genome expansion, although the involvement of OMT in other modifications than 5′-end capping was previously proposed for large nidoviruses [15].
In contrast to the case of the methyltransferases, the link between ExoN and genome size control in nidoviruses is supported by accumulating evidence obtained from different hypothesis-driven genetic studies [4], [20]. First, ExoN is exclusively found in a phylogenetically compact cluster of ssRNA+ viruses with large genome sizes. Second, cellular homologs of ExoN control the fidelity of replication in DNA-based life forms and are essential to maintain these large genomes. Third, ExoN active site mutants in MHV and SARS-CoV showed a stable phenotype characterized by a clearly enhanced mutation rate and nearly wild-type progeny yields.
The identification of the ExoN-encoding NDiV further strengthens the case for the direct involvement of ExoN acquisition in genome size expansion. First, because of its distant relation to any known virus and its insect host range that is a novelty for nidoviruses, NDiV provides an essentially independent verification for the association of ExoN with RNA viruses employing large genomes.
Second, it increases our confidence that no other domain is associated with large genome sizes in nidoviruses as strongly as ExoN is. The existence of such a domain is unlikely but it cannot be formally excluded because the entire proteomes of nidoviruses are yet to be fully described. However, our confidence about the lack of this alternative domain grows with the decrease of difference between genome sizes of nidoviruses containing and lacking ExoN: the smaller this difference the less capacity remains to encode an additional domain. With the identification of NDiV, this genome size gap decreased from ∼10.6 kb to ∼4.5 kb, the largest drop since this gap could have been recognized (∼14.9 kb in 1991) (Fig. S3).
Third, following the discovery of NDiV, only ∼0.8 kb remains of the other genome size gap of ∼7 kb that previously separated the ExoN-containing nidoviruses from all other ssRNA+ viruses (Fig. 1). Thus, a major step has been made towards a more precise definition of the RNA genome size limit above which the recruitment of a specialized enzyme for replication fidelity control may be a prerequisite. According to a custom binomial test (see Materials and Methods), the probability to observe the association of ExoN and large ssRNA+ genome size by chance may be 10−6 or lower. The genome size threshold of ∼20 kb, as defined by NDiV and a closterovirus [67], which has the largest genome size among ssRNA+ viruses other than nidoviruses, is also valid for unsegmented RNA viruses of other classes, all of which do not employ an ExoN in their replicative machinery [21].
The fixation of the ExoN domain in nidovirus genomes may be rationalized in the framework of a unidirectional triangular relationship that includes complexity, replication fidelity (mutation rate), and genome size [68] (Fig. 7). In RNA viruses, the low fidelity of replication severely restricts the size of their genomes, which can encode only relatively simple replication complexes that, hence, suffice to support low-fidelity replication [21], [69]. This low-state trap is known as the “Eigen paradox”. Accordingly, a transition from the “low” to the “high” state may not be accomplished by changing only one element of the triangle, e.g. improving replication fidelity, since such a change would not be compatible with the “low” state of the other two elements [68] [70]. The exclusive presence of ExoN in ssRNA+ viruses above 20 kb supports the logic of the Eigen paradox [68]. It also shows how the paradox could be solved with a single evolutionary advancement, the acquisition of ExoN, which may have relieved the constraints on all three elements of the triangular relationship (Fig. 7), providing a lasting benefit to the virus lineage that acquired ExoN. This advancement may have been accompanied by an immediate fitness gain. Accordingly, the ExoN acquisition could have provided the ancestral virus with improved control over the fidelity of its replication and the mutation spectrum (quasispecies structure) of its progeny [71], [72], which may have facilitated virus adaptation to the environment [20], [73]. Alternatively, ExoN could have been acquired in an evolutionarily neutral event. Through subsequent mutation this enzyme might have gained beneficial properties for the ancestral virus and its progeny. The functional and structural characterization of known nidoviruses and yet-to-be identified viruses in the genome size range around that of NDiV will be required to clarify this key aspect in the transition from small to large nidoviruses.
The acquisition of ExoN by an ancestral nidovirus must have produced viable progeny but it remains unknown whether, besides ExoN, any additional properties of the ancestral nidovirus were critical for genome expansion, as was speculated elsewhere [15]. Recently an exoribonuclease was identified in the ssRNA- arenaviruses, which have genome sizes below 10 kb [74], [75]. Unlike nidoviruses, arenaviruses employ the exoribonuclease as a domain of their nucleocapsid protein that, accordingly, mediates a non-replicative function. In line with these differences, the nidovirus ExoN and the arenavirus exoribonuclease do not share specific sequence affinity (CL and AEG, unpublished data), indicating that both are likely to have been acquired from independent sources and were integrated into different genetic settings to perform different functions.
NDiV may be the first but likely not the last nidovirus identified in mosquitoes [76]. Systematic probing of these and other insects could lead to the discovery of new nidoviruses, and characterization of those with genomes in the size range between small and large nidoviruses could be particularly insightful. As presented in this study, benefits of these advancements could be multifold and provide a foundation for both fundamental and applied research on newly discovered and already known viruses.
During continued surveillance for JEV in Vietnam between September 2001 and December 2003, 24,097 female mosquitoes belonging to six different Culex species (Culex tritaeniorhynchus, Culex gelidus, Culex vishnui, Culex fusco, Culex pseudo, and Culex quinquefaciatus) were collected. They were divided into 359 pools, each containing a single mosquito species and handled with utmost care following the appropriate biosafety measures. For the digestion of blood meals, the samples were kept in 5% glucose for two weeks at room temperature and a humidity of ∼90%. The most abundant species was Culex tritaeniorhynchus (10,194 mosquitoes accounting for a 42.3% share), followed by Culex gelidus (6,199, 25.7%), Culex vishnui (3,780, 15.7%), Culex quinquefaciatus (2868, 11.9%), with the remaining species ranging from 0.3%–4.1%. Mosquito pools were stored at −70 C prior to processing for virus isolation.
Four cell lines were used to isolate viruses, but NDiV was evident only in samples from Aedes albopictus C6/36 cells grown at 28 C in Eagle's Minimum Essential Medium (EMEM) containing 10% fetal calf serum (FCS) and 0.2 mM non-essential amino acids [77]. Pooled mosquitoes were washed three times in sterile phosphate-buffered saline (PBS, pH 7.2) containing 1000 g/ml each of penicillin and streptomycin, followed by rinsing with antibiotics-free PBS. The homogenates were prepared by triturating the mosquitoes in 2%-FCS-EMEM with subsequent centrifugation at 2,000 g for 10 min. The suspensions were filtered (0.22 nm Millipore, USA) and applied to C6/36 cells, which were monitored daily for cytopathic effects, also after three blind passages. The cell death, probably due to apoptosis, was indeed observed upon NDiV infection. The ICF were clarified by centrifugation at 2,000 g for 10 min.
The nucleic acid was extracted from the purified NDiV virus particles using phenol-chloroform extraction. It migrated as a single band in agarose gel electrophoresis, which was sensitive to RNase but not DNAse treatment, indicative of an RNA virus genome. Accordingly, reverse transcriptase (RT) was used to amplify parts of the NDiV genome by Random Arbitrary Primers-PCR (RAP-PCR) in order to initiate sequence analysis. Cassette primers (C1 and C2) coupled to random hexamers (Hx) were employed. Following synthesis of first and second cDNA strands with C1Hx and C2Hx primers, respectively, PCR amplification was performed using the cassette primers C1 and C2 as per the standard protocol [78]. Three amplicons of different sizes, which were specific for the virus-containing samples, were then cloned in the pCR2.1-TOPO vector (TOPO TA Cloning Kit, Invitrogen) according to the manufacturer's instructions. The sequence of the first cloned fragment (referred to as “index clone”) was determined by Big Dye Terminator Cycle Sequencing using M13 forward and reverse primers in an ABI 310 or 3100 automated DNA sequencer (Applied Biosystems). The cloned region of the genome was extended by ‘gene walking’ using primers based on previously obtained sequence information (Table S2).
To sequence the genomic region upstream of the index clone, the following amplification strategy was used, involving two DNA fragments called double-stranded (ds) cDNA and anchor DNA. To produce ds cDNA, viral genomic RNA was mixed with 10 mM dNTP mix and 2 pmol of 15-mer gene-specific primers (NDiV-RACE492-477RP, NDiV-RACE302-288RPB and NDiV-RACE435-420RPC) (Fig. S1A, Table S2).
An anchor DNA was synthesized by PCR that amplified a specific fragment of pUC19, including its multiple cloning site (Fig. S1B). Both, the ds viral cDNA and PCR product obtained from pUC19 (anchor) were digested by several restriction enzymes whose sites are present in the pUC19 multiple cloning site (BamHI, EcoRI, KpnI, HindIII, ScaI, and PstI). The digested pUC19 PCR products were then purified using the QIAXII gel purification kit (Qiagen) in order to collect the longer DNA fragments. The digested viral cDNAs were also purified by filtration using Micropure-EZ (Millipore) and Microcon YM-100 (Millipore) to remove enzymes and buffers. In a next step, the purified cDNAs and anchor DNAs were mixed and ligated using T4 DNA Ligase (TaKaRa). The unknown region of viral cDNA was then amplified by semi-nested PCR using LA-taq (TaKaRa), two viral gene specific primers and one pUC19 primer (Table S2) as shown in Fig. S1C. The reaction process included an initial denaturation at 96°C for 5 min, 35 cycles at 96°C for 30 sec, 53°C for 30 sec, and 72°C for 7 min, and a final extension at 72°C for 10 min.
The known viral genome sequence was further extended by long RT-PCR which resulted in an 8 kb fragment with a 68-nucleotide polyA tail representing the 3′-end of the NDiV genome. The GeneRacer™ Kit (Invitrogen) was used to sequence the 5′-end of the NDiV's genome.
The NDiV origin of newly obtained sequences was further validated by probing different samples with a primer pair designed against the index clone. This pair of primers recognized NDiV isolates, but not JE and dengue viruses (flaviviruses) or SARS-coronavirus (Coronavirus). These results indicated that NDiV is a novel mosquito virus.
Specific primers encompassing NDiV nts 19,733 and 20,126 (including 2 Adenines of the poly (A) tail), respectively, were designed (Table S2). The generated PCR product was purified using the Qiaex II gel extraction kit (500) (Qiagen) following the manufacturer's instructions. The purified PCR product was then ligated to a 3.5 kb plasmid (PCR-XL-TOPO) using the TOPO XL PCR cloning kit (Invitrogen, applying the TA rule based on the Taq polymerase's capacity of adding an extra A at the 3′ end of each DNA chain of a PCR product) as per the manufacturer's indications. Heat shock transformation into One Shot Top 10 chemically competent cells (Invitrogen) was carried out and the transformed cells were incubated in SOC medium at 37 C for 2 hrs. After that, the E. coli cells were cultured in 50 µg/ml containing LB plates overnight and the positive clones were subsequently cultured in LB broth at 37 C overnight. The plasmid alkaline extraction was done using the QIAprep spin Miniprep kit (Qiagen) as the manufacturer indicated. As a next step, verification of the probe orientation was carried out by nucleotide sequencing. Finally, transcription of the cloned DNA sequences was done to generate the RNA probe (in both sense and reverse orientations). The RNA probe was then labeled with 32P by using the AmpliScribe T7 High Yield Transcription Kit (EPICENTRE Biotechnologies) following the company's instructions.
To investigate the possibility that NDiV generates set of 3′-coterminal sub-genomic mRNA's during its replication, Aedes albopictus C6/36 cells were infected with NDiV. Three to four days after infection intracellular poly (A)-containing RNA from mock-infected and NDiV-infected cells was prepared using Dynabeads oligo(dT)25 (Dynal Biotech) as per the manufacturer's instructions. RNA was separated on a glyoxal-based agarose gel system and blotted on a positively charged nylon membrane (BrightStar-Plus membrane). The mRNA bands were then hybridized with an α-32P-multiprime-labeled RNA probe specific for NDiV at 65°C overnight (see above RNA probe generation). The membrane was then washed with low and high stringency wash solutions and the RNAs were analyzed by autoradiography. All reagents for mRNA separation, transfer and hybridization (with the exception of the RNA probe) were provided with the NorthernMax-Gly Kit (Ambion). The manufacturer's instructions were followed. A 0.5–10 Kb RNA Ladder (Invitrogen) was used as a marker set to calculate apparent molecular mass of the analyzed bands.
For electron microscopy, virus was concentrated from ICF by centrifugation at 12,000 g for 30 min at 4 C, after which 6.6% polyethylene glycol 6000 and 2.2% NaCl were added to the supernatant. After stirring for 1 h at 4 C and centrifugation at 12,000 g for 1 h, the supernatant was discarded. The virus-containing pellet was dissolved in saline-Tris-EDTA buffer, sedimented at 250,000 g for 1 h and resuspended a second time. The concentrated virus was negatively stained with 1% sodium phosphotungstic acid, pH 6.0, and examined at 100 KV using a transmission electron microscope (JEM-100CX, JEOL, Japan) [79].
Virions were purified in a 15–50% sucrose density gradient using an SW32Ti rotor (Beckman Coulter, Inc., Fullerton,CA) at 20,000 rpm for 12–16 h at 4°C. Gradient fractions were analyzed by 16% SDS-polyacrylamide gel electrophoresis and Coomassie Brilliant Blue G staining (Fig. 2B). Protein bands were excised and either directly sequenced by automated Edman degradation (Applied Biosystems model 491cLC) or digested with lysylendopeptidase prior to HPLC purification and sequencing.
Genome sizes of ssRNA+ viruses were retrieved from the NCBI Viral Genome Resource [80]. GenBank, version 178.0 [81], Pfam database, version 24.0 [34], SCOP70, version 1.75 [82], and an in-house nidovirus domain profile database [15], [54] updated in this study were used to identify putative functional domains encoded by the NDiV genome. Representatives of the nidovirus species defined according to (http://www.ictvonline.org/virusTaxonomy.asp?version=2009) plus NDiV, whose taxonomical status remains provisional, were used as detailed in Table S3. Species names of coronaviruses were taken from ICTV proposal 2008.085-122V.U that was approved by ICTV in 2009. Fields after the “_” sign in virus abbreviations represents sampling year or period.
The NDiV ORFs were compared with sequence databases using psi-BLAST [27], HMMer 2.3.2 [83], TMpred [84], or HHsearch [85]. Protein secondary structure predicted by Psipred [86] was included in the HHsearch-mediated profile searches. RNA secondary structure analysis was conducted using Mfold [29] and pknotsRG [30]. MUSCLE [87] was used to produce alignments of nidovirus proteins that were manually refined in poorly conserved regions. Alignment derivatives, with the least conserved columns removed [88], were prepared using BAGG [89] and were used for profile searches and phylogenetic analyses. Alignments were prepared for publication using JalView [90]. To compile and plot most graphs and conduct statistical analyses we used the R package [91].
Using the de novo repeat detection program RepeatScout [92] a library of perfect repeats with unit sizes ranging from four to the maximum observed size of 16 was compiled for the NDiV genome sequence. The library was filtered to retain repeats of different types according to the following constraints applied to each type separately: (i) one repeat copy must be located upstream of ORF1a, and (ii) another one must reside within the 300 nt region immediately upstream of either ORF2a, ORF3, or ORF4. Each set of the retrieved repeats was subsequently analyzed for conservation by alignment that included flanking regions of 20 nt at each side. The longest repeats with highest similarity were considered TRS candidates.
To map major nidovirus replicative proteins to pp1ab of NDiV we applied alignment-based methods. Multiple sequence alignments represent a general tool to infer both common ancestry (orthology) of residues for several related sequences (these residues form a fully occupied alignment column) and identify insertion/deletion events (corresponding to alignment columns containing gaps in selected sequences). Multiple alignments can be converted into profiles, which are statistical models that capture the degree of conservation and the likelihood to observe a certain residue or gap in each alignment column. One type of profiles are profile Hidden Markov Models (HMMs) [93] that are particularly suitable for searching for remotely related sequences (like NDiV which presumably represents a new virus family) in a probabilistic framework. They are implemented, for example, in the programs HMMer and HHsearch which were utilized in this study. A profile HMM can be compared to other HMMs or used to search for motifs in a single sequence. Due to the high degree of divergence of nidovirus sequences, we used alignments of amino acid sequences and profiles derived from these alignments to probe relation between proteins in this study.
Phylogenetic analyses were performed as described previously [94]. Bayesian posterior probability trees were compiled utilizing BEAST [95] under the WAG amino acid substitution matrix [96] using Tracer [97] to verify convergence. For the nidovirus-wide analysis, whose sampling is detailed Table S3, we used a concatenated alignment of 3CLpro, RdRp, and HEL1 including 910 aa positions and its derivative of 604 aa positions, from which least conserved columns were removed. In this analysis, the uncorrelated relaxed molecular clock approach (lognormal distribution) [98] was used as it was favored [99] over the strict molecular clock (log10 Bayes factor of 13.6) and equal to the relaxed molecular clock approach with exponential distribution (log10 Bayes Factor of 0.0). Selected internal nodes were fixed using results of separate analyses of subsets of nidoviruses. For phylogenetic analysis of the subfamily Coronavirinae and the family Arteriviridae, we used respective datasets incorporating between one and three sequences per species and including concatenated alignments of ORF1ab domains that are conserved in each of these groups. The datasets included 35 and 10 sequences for corona- and arteriviruses and consisted of 2302- and 2882-aa alignment positions, respectively. The topologies of these trees closely follow those published [51]. They were used to fix internal nodes in corona- and arterivirus clusters in the subsequent nidovirus-wide phylogenetic analysis. The exception was the basal nodes corresponding to the grouping of the Alpha-, Beta-, and Gammacoronavirus genera and the root of arteriviruses (EAV or SHFV), which were left unfixed. Maximum Likelihood trees were compiled utilizing the PhyML software [100]. The WAG amino acid substitution matrix and rate heterogeneity among sites (8 categories) were applied and support values for internal nodes were obtained using the non-parametric bootstrap method with 1000 replicates. Trees were rooted using domain-specific outgroups: for RdRp, three picornavirus representatives (accession numbers: NC_001489, NC_001897, NC_002058); for HEL1, four rubi-/ tetra-/ hepevirus representatives (NC_001545, NC_001990, NC_005898, NC_001434).
We sought to statistically define a genome size threshold that separates ExoN-containing from ExoN-lacking ssRNA+ viruses. To this end, we developed a custom test employing the binomial probability function and including all 43 virus groups displayed in Fig. 1. These groups consist of thousands of viruses that are believed to have emerged from a common ancestor, implying that they are not independent. Their dependence varies in virus pairs but, generally, for each virus pair is inversely proportional to the pair-wise evolutionary distance. To account for the dependence of these sequences in our test is technically challenging. To circumvent this problem, we have created a derivative of the virus dataset in which each virus family/group is represented by a single virus, in total 43 viruses. We considered the sequences of these representatives to be essentially independent due to the (extremely) large divergence that is observed, even in the most conserved genes (e.g. see Fig. 6), the lack of recognizable similarity in other genes, and the accompanied gene loss and gain.
For a given genome size threshold, ssRNA+ viruses were partitioned into two groups (below and above that threshold) and the value of the binomial density function was calculated for both groups using information on the presence or absence of ExoN. The final probability of the test is the product of the binomial probabilities for the two groups. We used a binomial success probability of 4/43 since four out of the 43 ssRNA+ virus lineages (NDiV, toro-/bafiniviruses, coronaviruses, and roniviruses) employ ExoN. The test was applied to each possible threshold separating two unique ssRNA+ genome sizes, in total – 42 thresholds. The threshold of ∼20 kb, between the genome sizes of NDiV and closteroviruses, gave the lowest probability to observe the ExoN association by chance. We consider the obtained value (10−6) as an underestimate of the true probability that should be calculated by taking into account the sequence dependence and all viruses in the 43 groups, which without exception conform to the ExoN distribution observed in the selected virus representatives used now.
RefSeq accession numbers of proteins referred to in the text for a selection of prototype nidoviruses are: 3C-like proteinase (EAV: NP_705584, SARS-CoV: NP_828863, WBV: YP_803213, GAV: YP_001661453), RNA-dependent RNA polymerase (EAV: NP_705590, SARS-CoV: NP_828869, WBV: YP_803213, GAV: YP_001661452), superfamily 1 helicase (EAV: NP_705591, SARS-CoV: NP_828870, WBV: YP_803213, GAV: YP_001661452), exoribonuclease (SARS-CoV: NP_828871, WBV: YP_803213), N7-methyltransferase (SARS-CoV: NP_828871), uridylate-specific endonuclease (EAV: NP_705592, SARS-CoV: NP_828872, WBV: YP_803213) and 2′-O-methyltransferase (SARS-CoV: NP_828873, WBV: YP_803213).
|
10.1371/journal.pgen.1003469 | Genetic and Biochemical Assays Reveal a Key Role for Replication Restart Proteins in Group II Intron Retrohoming | Mobile group II introns retrohome by an RNP-based mechanism in which the intron RNA reverse splices into a DNA site and is reverse transcribed by the associated intron-encoded protein. The resulting intron cDNA is then integrated into the genome by cellular mechanisms that have remained unclear. Here, we used an Escherichia coli genetic screen and Taqman qPCR assay that mitigate indirect effects to identify host factors that function in retrohoming. We then analyzed mutants identified in these and previous genetic screens by using a new biochemical assay that combines group II intron RNPs with cellular extracts to reconstitute the complete retrohoming reaction in vitro. The genetic and biochemical analyses indicate a retrohoming pathway involving degradation of the intron RNA template by a host RNase H and second-strand DNA synthesis by the host replicative DNA polymerase. Our results reveal ATP-dependent steps in both cDNA and second-strand synthesis and a surprising role for replication restart proteins in initiating second-strand synthesis in the absence of DNA replication. We also find an unsuspected requirement for host factors in initiating reverse transcription and a new RNA degradation pathway that suppresses retrohoming. Key features of the retrohoming mechanism may be used by human LINEs and other non-LTR-retrotransposons, which are related evolutionarily to mobile group II introns. Our findings highlight a new role for replication restart proteins, which function not only to repair DNA damage caused by mobile element insertion, but have also been co-opted to become an integral part of the group II intron retrohoming mechanism.
| Mobile group II introns are bacterial retrotransposons that are evolutionarily related to introns and retroelements in higher organisms. They spread within and between genomes by a mechanism termed “retrohoming” in which the intron RNA inserts directly into a DNA site and is reverse transcribed by an intron-encoded reverse transcriptase. The resulting intron cDNA is integrated into the genome by host factors, but how it occurs has remained unclear. Here, we investigated the function of host factors in retrohoming by genetic and biochemical approaches, including a new biochemical assay that reconstitutes the complete retrohoming reaction in vitro. Our results lead to a comprehensive model for retrohoming, which includes a surprising role for replication restart proteins in recruiting the host replicative DNA polymerase to copy the intron cDNA into the genome in the absence of DNA replication. We also find an unexpected contribution of host factors to initiating reverse transcription and a new RNA degradation pathway that suppresses retrohoming. We suggest that key features of the group II intron retrohoming mechanism may be used by human LINE elements and other non-LTR-retrotransposons. Additionally, our results provide new insights into the function of replication restart proteins, which are critical for surviving DNA damage in all organisms.
| Mobile group II introns are non-long-terminal-repeat (non-LTR) retroelements that are commonly found in prokaryotes and in organellar genomes of eukaryotes and are thought to be evolutionary ancestors of splicesomal introns and retrotransposons in higher organisms [1]. They consist of an autocatalytic intron RNA (“ribozyme”) and an intron-encoded protein (IEP), which has reverse transcriptase (RT) activity. These two components function together in a ribonucleoprotein complex (RNP) to promote intron mobility by a mechanism in which the excised intron lariat RNA uses its ribozyme activity to reverse splice directly into a DNA site and is then reverse transcribed by the IEP, yielding an intron cDNA that is integrated into the genome by host enzymes [2]–[5]. By using this mechanism, group II introns insert at high frequency into specific DNA target sites in a process called “retrohoming” and at low frequency into ectopic sites that resemble the normal homing site in a process called “retrotransposition” or “ectopic retrohoming” [6]. These processes enabled the dispersal of group II introns to a wide variety of bacteria and some archaea and likely into eukaryotic nuclear genomes, where ancestral group II introns are thought to have evolved into both spliceosomal introns and non-LTR-retrotransposons [3], [7], [8].
Although the early reverse splicing and reverse transcription steps catalyzed by group II intron RNPs are common to retrohoming pathways in all organisms, the late host-mediated steps of second-strand DNA synthesis and cDNA integration can occur in different ways. In Saccharomyces cerevisiae mitochondria, where retrohoming was studied initially, cDNA integration occurs largely by a recombination mechanism in which the nascent intron cDNA initiated at the recipient allele invades an intron-containing allele for completion of intron DNA synthesis before switching back to the recipient DNA in the upstream exon [9], [10]. In bacteria, however, the fully reverse spliced intron RNA is reverse transcribed to yield a full-length intron cDNA that is integrated directly into the recipient DNA by a RecA-independent mechanism hypothesized to involve host DNA repair enzymes [5], [11], [12]. Recently, non-lariat, linear forms of the Lactococcus lactis Ll.LtrB intron RNA were found to retrohome in Drosophila melanogaster by using host non-homologous end-joining enzymes for cDNA integration [13]. However, host factors that function in late steps in the retrohoming of group II intron lariat RNAs, the major retrohoming pathway used in nature, have not been identified conclusively in any organism, and consequently, the mechanisms used for these steps have remained poorly understood.
Figure 1 diagrams the major steps elucidated thus far in the retrohoming pathway of the L. lactis Ll.LtrB intron, which has been studied extensively as a model system for group II intron lariat RNA retrohoming in bacteria. The Ll.LtrB intron was discovered in a relaxase gene (ltrB) in a conjugative element, where its splicing is required to produce functional relaxase for conjugation [14], [15]. Its IEP, denoted LtrA protein, is multifunctional, with RT, RNA splicing (“maturase”), DNA binding, and DNA endonuclease activities [16], [17]. Transcription of the ltrB gene yields a precursor RNA, which contains the intron flanked by the 5′ and 3′ ltrB exons (E1 and E2, respectively). The IEP, which is translated from within the intron, binds to the intron in the unspliced precursor RNA and promotes its splicing by stabilizing the catalytically active RNA structure [18]. Splicing occurs via two sequential RNA-catalyzed transesterification reactions that yield ligated ltrB exons and an excised intron lariat RNA to which the IEP remains tightly bound in an RNP. RNPs then initiate retrohoming by recognizing a DNA target sequence (corresponding to the ligated ltrB E1–E2 DNA sequence), using both the IEP and base pairing of the intron RNA [19]–[21]. After DNA target site recognition, the intron RNA fully reverse splices into the top strand of the DNA, leading to insertion of the intron RNA between the two DNA exons, while the IEP cleaves the bottom strand 9 nts downstream of the intron-insertion site and uses the 3′ end of the cleaved DNA strand for target DNA-primed reverse transcription (TPRT) of the inserted intron RNA [16], [17]. Finally, the resulting intron cDNA is integrated into the recipient DNA by host factors in late steps that minimally include the degradation or displacement of the intron RNA template strand, second (top)-strand DNA synthesis, resection of DNA overhangs, and ligation to seal DNA nicks [12].
In addition to its native host, the Ll.LtrB intron splices and retrohomes efficiently in a wide variety of other bacteria, including Escherichia coli, where it has been studied by using the facile genetic and biochemical methods available for that organism [5]. By screening E. coli mutants using two different plasmid-based retrohoming assays, we in collaboration with the Belfort laboratory previously identified candidate host factors that potentially function in the late steps in retrohoming, including RNase H1 and the 5′→3′ exonuclease activity of Pol I, both of which could contribute to degrading the intron RNA template strand; the host replicative polymerase Pol III, which may function in second-strand DNA synthesis; and DNA ligase A, which presumably seals strand nicks [12]. Decreased retrohoming frequencies were also found in mutants deficient in host exo- and endonucleases activities [RecJ, DnaQ (MutD), and SbcD], which could function to resect overhangs or resolve intermediates, and increased retrohoming frequencies were found for mutants deficient in RNases I and E and exonuclease III, which in wild-type strains may suppress retrohoming by degrading the intron RNA or nascent cDNA [12].
More recently, Coros et al. [22], [23] extended this work by screening an E. coli transposon-insertion library for mutants defective in group II intron retrohoming into chromosomal sites, using a donor plasmid to express an Ll.LtrB-ΔORF intron carrying a kanR marker. This screen identified additional host factors potentially involved in retrohoming, including polynucleotide phosphorylase (PNPase), the DNA helicase Rep, and MnmE (TrmE), which functions in tRNA modification, with additional host proteins (CyaA, SpoT, and AtpA) acting by affecting accessibility of chromosomal target sites or energy metabolism, and RNase E acting to impede retrohoming by degrading the intron RNA.
Although the genetic screens described above identified host genes in which mutations decrease retrohoming efficiency, it remains possible that some or many of these mutations affect retrohoming indirectly. Such indirect effects could result from mutations that impair the propagation or expression of the intron-donor plasmid, decrease the intracellular levels or activity of group II intron RNPs, or impede the accessibility of group II intron RNPs to DNA target sites. Additionally, all previous genetic screens relied upon the expression of an antibiotic-resistance marker to identify retrohoming events and are thus vulnerable to false positives arising from mutations that affect the expression of antibiotic-resistance (e.g., by affecting the expression of the antibiotic-resistance gene or cellular permeability to the antibiotic). In some cases, studies of “indirect” effects revealed by the genetic screens have provided rich information about host responses and the regulation of group II intron mobility [22], [23]. However, further insight into the retrohoming mechanism requires the identification of host factors that function directly in this process.
Here we used an E. coli genetic screen and Taqman qPCR assay that mitigate indirect effects to identify candidate host factors for Ll.LtrB retrohoming, and we tested their function in retrohoming by using a newly developed biochemical assay that combines group II intron RNPs with cellular extracts to reconstitute the complete retrohoming reaction in vitro. By using these multiple approaches, we confirmed some previously identified host factors but not others. Additionally, we found that replication restart enzymes play a key role in intron retrohoming by initiating second-strand DNA synthesis. Our findings indicate a novel mechanism for the major pathway of group II intron lariat retrohoming in bacteria, with features that may be shared by human LINE elements and other non-LTR retrotransposons.
To identify host factors that function in retrohoming of the Ll.LtrB group II intron, we used two complementary approaches that mitigate weaknesses of previous genetic screens. First, we used a plasmid-based group II intron retrohoming assay that controls for indirect effects to screen an E. coli mariner transposon-insertion library for mutants that have decreased or increased retrohoming efficiency. This screen used the intron-donor plasmid pALG3 and recipient plasmid pBRR3-ltrB (Figure 2A) and was done in E. coli host strain HMS174(DE3), which is RecA− and encodes an isopropyl β-D-1 thiogalactopyranoside (IPTG)-inducible T7 RNA polymerase for donor-plasmid transcription. The donor plasmid pALG3, which was newly constructed for this screen, uses a T7lac promoter to synthesize a precursor RNA, which contains an Ll.LtrB-ΔORF intron (i.e., an intron deleted for the LtrA ORF) flanked by short 5′ and 3′ exons, with the 3′ exon linked in frame to an ORF encoding GFP. The LtrA protein, which functions stoichiometrically to splice and mobilize the Ll.LtrB intron, is expressed from a position downstream of the GFP ORF, where it is co-transcribed with the intron RNA and translated using its own Shine-Dalgarno sequence. Because the expression of the Ll.LtrB-ΔORF intron and its splicing to produce RNPs are linked to GFP expression, mutants with decreased group II intron RNP production as a result of defects in donor plasmid replication, LtrA protein expression, or the expression and splicing of the Ll.LtrB intron RNA, are identified readily by decreased GFP fluorescence after IPTG induction.
To enable direct selection for retrohoming events, the Ll.LtrB intron in pALG3 carries a trimethoprim-resistance retrotransposition-activated genetic marker (TpR-RAM), which consists of a small trimethoprim-resistance gene inserted in the orientation opposite group II intron transcription, but interrupted by an efficiently self-splicing group I intron, the phage T4 td intron, in the forward orientation [24]. During retrohoming via an RNA intermediate, the td intron is spliced, thereby reconstituting the TpR marker and enabling its expression after the intron retrohomes into the DNA target site. By using this combination of TpR-RAM and GFP markers, mutants with transposon insertions that inhibit retrohoming without affecting Ll.LtrB expression or splicing are identified as TpS and GFP+.
The screen was done in 96-well plates under calibrated selective conditions in which the cell density for log phase cells provides a measure of retrohoming efficiency (Figure S1; Materials and Methods). Screening of 9,200 colonies by two rounds of 96-well plate assays identified 61 transposon insertions that reproducibly gave a >4-fold decrease in retrohoming efficiency compared to a wild-type control and had no defect in RNP production, as judged by a fluorescence-activated cell sorter (FACS) assay of GFP synthesis (Table S1). After mapping of transposon-insertion sites by thermal-asymmetric-interlaced (TAIL) PCR [25], we identified 67 candidate protein-encoding genes, whose disruption or altered expression due to the proximity of the transposon insertion results in decreased retrohoming efficiency. Six of these candidate genes were sites of multiple transposon insertions, and 12 were genes with nucleic acid-related functions found downstream of transposon-insertion sites within operons. An additional transposon insertion (C0719) that decreased retrohoming efficiency mapped to a site predicted to encode a small non-coding RNA (sRNA) (Table S1). All the candidates were confirmed by Southern hybridization to contain a single mariner-transposon insertion at the indicated genomic location (data not shown).
To complement the transposon-insertion screen, our second approach was to screen individual candidate strains for efficient integration into a chromosomal target site directly by using a Taqman qPCR assay to quantify both the 5′- and 3′-intron-integration junctions, thereby eliminating false positives that arise from mutations affecting expression of a drug-resistance phenotype (Figure 2B). Mobile group II introns can be retargeted to retrohome into different chromosomal DNA sites simply by modifying the intron RNA sequences that base pair to the DNA target sequence (see Introduction), a gene targeting technology known as “targetron” [19], [21], [26]. The Taqman qPCR assay uses an Ll.LtrB-ΔORF intron that was retargeted in this way to retrohome efficiently into a site in the rhlE gene, which encodes a non-essential DEAD-box protein whose disruption has no effect on cellular growth rate [21], [27]. The intron was expressed from the broad-host range donor plasmid pBL1, which has a different DNA replication origin than pALG3 and employs an m-toluic acid-inducible promoter; the latter functions independently of host factors and is activated by a freely permeable inducer (m-toluic acid) that does not require cellular transporters to enter the cell [28]. Additionally, the screen was carried out in mutant strains from the Keio collection in which deleted genes are replaced with a kanR marker, thereby mitigating polarity effects on downstream genes in operons [29]. The Keio strains were supplemented by temperature-sensitive mutants to test the contribution of essential genes.
We used the Taqman qPCR assay to test all 68 candidate host factors identified in our initial transposon-insertion screen, as well as 30 additional candidate proteins that act on nucleic acids, including all 21 such candidates identified in previous mutant screens [12], [22], [30]. Table 1 shows results of the Taqman qPCR assay for notable mutants, and Tables S1 and Table S2 show complete results for the mariner-transposon screen and Taqman qPCR assay, respectively. Among the 68 candidates identified in the initial transposon library screen (Table S1), only ten (dnaC, dnaT, gyrB, mdoB, paoD, rpoH, rpoN, tonB, ydcM, and yjjB) had statistically significant decreases in retrohoming efficiency in the Taqman qPCR assay, and only four (dnaC, dnaT, gyrB, and rpoH) had substantial decreases (10–67% of wild type retrohoming efficiency; Table 1 and Table S2). This poor correlation highlights the difficulty of distinguishing direct and indirect effects and the necessity of using multiple approaches to identify host factors that function in retrohoming.
Among the candidates identified as potential retrohoming factors in previous screens [12], [22], [30], the Taqman qPCR assay confirmed significant reductions in retrohoming efficiency in the Keio deletions of rnhA (RNase H1, the major cellular RNase H [31]); seqA (initiation of chromosomal DNA replication); sbcC (ATP-dependent exonuclease); hns (histone-like nucleoid structuring protein); and tus (DNA replication termination site-binding protein); as well as at restrictive temperatures in the temperature-sensitive mutants polAexts, which is defective in the 5′→3′ exonuclease but not the DNA polymerase activity of Pol I [32]; and dnaEts in the catalytic (α) subunit of the host replicative DNA polymerase Pol III [33]. Also in agreement with previous results, we found no strong decrease in retrohoming efficiency in a Keio deletion of rnhB (RNase H2 [34]).
In contrast to results of genetic assays, the Taqman qPCR assays found no decrease in retrohoming efficiency for Keio deletions of recJ (single-stranded DNA exonuclease [35]); dnaQ (Pol III ε subunit, which has the proofreading exonuclease activity [36]); rep and recQ (DNA helicases [37], [38]); recF (RecA-dependent recombination); stpA (H-NS-like DNA- and RNA-binding protein with RNA chaperone activity [39]); mnmE (trmE; tRNA modification); ligA and ligB (DNA ligases [40], [41]); and pnp (polynucleotide phosphorylase [42]). Additionally, the Taqman qPCR assay found no decrease in retrohoming efficiency for Keio deletions of the genes encoding DNA repair polymerases polB (Pol II [43]), dinB (Pol IV [44]), and umuC or D (Pol V [45]), whereas polB and dinB deletions in a different strain showed moderate decreases in previous genetic assays [12].
The new candidates that were identified in the transposon-insertion screen (Table S1) and confirmed to have substantial decreases in retrohoming efficiency in the Taqman qPCR assay (10–67% wild type; Table 1) were: DnaC and DnaT, which function in replication restart (identified as genes downstream of the transposon insertion in the yjjB operon [46], [47]); GyrB (DNA gyrase subunit B); and RpoH (RNA polymerase σ32 factor).
For several mutants in which inhibition of retrohoming was found in genetic assays but not in the Taqman qPCR assay, we subsequently found significant effects on top- or bottom-strand DNA synthesis in biochemical assays below (e.g., dinB, dnaQ, ligA, recJ, pnp, polB, and stpA). The disagreement between the genetic and Taqman qPCR assays for these mutants may reflect: (i) that qPCR monitors only short DNA regions at the intron-integration junctions; (ii) the longer time of the Taqman qPCR assay, which may give alternative enzymes a greater chance to act; or (iii) the different genetic backgrounds of the strains used in the two assays. The results again emphasize the need to use multiple assays to identify retrohoming factors, with biochemical support for a genetic assay in our view providing the most definitive identification.
The transposon-library screen in E. coli HMS174(DE3) identified eight retrohoming-deficient mutants that are TpS/GFP−, potentially indicating a defect in the production of Ll.LtrB RNPs (Table S3). Although we hoped that such mutants would identify host factors required for Ll.LtrB intron splicing, all eight of these mutants have transposon insertions that likely affect donor plasmid transcription (four in the lacUV5 promoter of the λDE3 prophage, one in the T7 RNA polymerase gene, and three in genes encoding membrane transporters that could affect IPTG induction or trimethoprim uptake: ugpA (glycerol-3-phosphate uptake transporter subunit); xylF (xylose transporter subunit); and yjbB (a putative transporter). Consistent with an effect on transcription, all eight mutants showed decreased GFP fluorescence when counter-screened with the control plasmid pALE, which lacks the Ll.LtrB intron and contains the ligated ltrB exons fused directly to GFP (Figure S2, Figure S3, and Table S3). A direct screen of the transposon library for splicing-defective mutants using plasmid pALG2, in which an Ll.LtrB-ΔORF intron lacking the TpR-RAM marker is linked to GFP expression [48], also identified only mutants that are defective in GFP expression with both the reporter construct containing the intron and the control reporter construct lacking the intron [T7 RNA polymerase of the λDE3 prophage; malE (maltose ABC transporter subunit); kdsD (arabinose-5-phosphate isomerase); dppD (dipeptide ABC transporter subunit); clcA (H+/Cl− exchange transporter); yfeN (conserved outer membrane protein); pgi (phosphoglucose isomerase; and yjbE (predicted protein)] (Table S4). These findings suggest either that splicing of the Ll.LtrB intron in vivo requires only the LtrA protein as it does in vitro [17] or that host-encoded splicing factors are essential proteins that are not readily identified in a transposon screen.
The TpR-RAM screen also identified five transposon insertions that give increased retrohoming efficiencies and potentially encode host factors that function to suppress retrohoming. Surprisingly, all five of these transposon-insertions mapped to three closely linked genes, rnlA (RNase LS), yfjK (DExH/D-box protein), and yfjL (protein of unknown function), which are part of a cryptic prophage (CP4-57) in E. coli K12 (Figure S4, Table S5; [49], [50]). In addition to higher group II intron retrohoming efficiencies indicated by high levels of TpR, all five disruptants showed increased levels of GFP fluorescence both with the intron-donor plasmid pALG3, in which splicing of the Ll.LtrB-ΔORF intron is required for GFP expression, and with the control plasmid pALE, which has ligated ltrB exons fused directly to GFP (Figure S4). Given the identity of the affected genes, these findings suggest that the increased retrohoming efficiencies and GFP fluorescence result from decreased rates of RNA degradation, leading to elevated levels of group II intron RNPs and GFP mRNA. The degradosome, which functions in mRNA turnover in E. coli, is a multiprotein complex consisting of RNase E (an endoribonuclease), PNPase (an exoribonuclease), RhlB (a DEAD-box RNA helicase), and enolase [51]. The close linkage of the three genes potentially involved in intron RNA degradation in our screen suggests that they may function together in a previously unknown RNA degradation pathway, possibly a second degradosome.
The decreased retrohoming efficiency resulting from a transposon insertion in the yjjB operon containing dnaC and dnaT in the TpR-RAM screen (see above; Table S1) focused our attention on replication restart proteins as attractive candidates for playing a role in the late steps of retrohoming. We therefore carried out systematic Taqman qPCR assays of replication restart mutants and found significant reductions in retrohoming in Keio deletions of priA, priC, and dnaT, and in a temperature-sensitive mutant of dnaB (Table 1). PriA and PriC are key proteins that independently recognize stalled or collapsed replication forks in the three major E. coli replication restart pathways (denoted the PriA-PriB, PriA-PriC, and PriC-Rep pathways; [47], [52]), while DnaT interacts with PriA and PriC to load the replicative DNA helicase DnaB [53], [54]. We also found decreased retrohoming efficiencies in temperature-sensitive mutants of several essential genes that function in replication restart, including those encoding DnaC, which interacts with DnaB prior to loading [55]; DnaG, the DNA primase [56]; and the single-stranded DNA binding protein Ssb, which has been shown to promote the formation of the primosome at the chromosomal replication origin (oriC) and interacts with PriA to stimulate the loading of DnaB during replication restart [57], [58]. However, deletion of the genes encoding PriB, an auxiliary component of the PriA-dependent pathway, and Rep, which functions in conjunction with PriC [59], showed only small (1–21%) reductions in retrohoming efficiency.
We also carried out Taqman qPCR assays of retrohoming in a different set of replication restart mutants in the genetic background of E. coli SS996, a recA+ strain containing a GFP reporter for SOS induction. The results indicated that the decreased retrohoming efficiencies in the affected replication restart mutants are not allele specific and are larger than can be accounted for by the proportion of cells undergoing the SOS response (Figure S5; different alleles tested for all genes except dnaE and dnaB). Additionally, the mutant strain SS4610 (lexA51::Tn5), which has a lexA null allele and is constitutively induced for the SOS response in nearly 100% of cells [60], showed only minimally decreased retrohoming frequencies in the Taqman qPCR assay (81–87% wild type; Figure S5 and Table 1). Collectively, the above findings indicate that the decreased retrohoming efficiency in the replication restart mutants is not a secondary effect of cell cycle arrest during SOS induction and indicate a requirement for replication restart proteins in group II intron retrohoming.
To further test the function of individual host proteins, we developed a biochemical assay in which host factors function together with group II intron RNPs to reconstitute the complete retrohoming reaction in vitro. This assay uses an E. coli S12 extract similar to those used for in vitro transcription and translation [61], [62]. Figure 3 shows experiments in which 5′ top- or bottom-strand labeled, 73-bp DNA oligonucleotide substrates containing the Ll.LtrB target site were incubated with group II intron RNPs in the presence of the extract, dNTPs (dATP, dCTP, dGTP, and dTTP), ATP, and an ATP-regenerating system (phosphoenolpyruvate+pyruvate kinase) at 37°C. The products were then analyzed in a denaturing polyacrylamide gel before and after digestion with RNases A+H to degrade the reverse spliced intron RNA leaving only 5′-labeled DNA products.
Time courses with the 5′-labeled bottom-strand substrate, which monitors cDNA synthesis, showed that an RNase-resistant band of the size expected for full-length bottom strand (988 nt) appeared after 5 min and accumulated for up to 30 min, along with a series of smaller bands (Figure 3A). These smaller bands are likely incomplete cDNAs rather than degradation products, as controls showed that exogenous 32P-labeled ssDNA and dsDNA corresponding to the retrohoming products were not degraded when incubated in the extracts under the same conditions (Figure S6).
Time courses with the 5′-labeled top-strand substrate, which monitors second-strand DNA synthesis, showed that an RNase-resistant band of the size expected for full-length top strand appeared later (10 min) and continued to accumulate during the reaction (Figure 3A, bottom right panel). We confirmed by primer extension that both the 5′- and 3′-junctions in the newly synthesized top strand DNA are continuous (Figure 3B).
Figure 3C shows that the appearance of labeled top- and bottom-strand products is dependent upon the addition of RNPs (lanes 4 and 8) and that top-strand DNA synthesis is completely dependent upon the presence of extract (lane 6). ATP increased the levels of reverse splicing and cDNA synthesis and was required for top-strand DNA synthesis in the extracts (cf., lane 1 and 5 with lanes 3 and 7), indicating that energy-dependent processes, perhaps involving DNA or RNA helicases, are involved at both stages of the reaction. Further, RNPs containing a mutant LtrA protein that lacks RT activity (RT−; YADD→YAAA) carried out reverse splicing, but showed no detectable cDNA synthesis, indicating that bottom-strand synthesis is dependent upon the group II intron RT activity and is not done by a host polymerase in the extracts (Figure 3D).
Finally, the E. coli extract assay enabled us to directly analyze bottom- and top-strand DNA synthesis in retrohoming-deficient mutants identified in genetic screens. In these experiments, extracts from Keio deletion or temperature-sensitive mutants were compared with those from their parental wild-type strains. In most cases, extracts were prepared from cells grown continuously at 37°C, which is a semi-permissive temperature for most of the temperature-sensitive mutants. The priA deletion strain was grown at 30°C to avoid accumulation of suppressor mutations, and five temperature-sensitive mutants (dnaEts, gyrBts, ligAts, rpoHts, and ssbts) that could not grow at 37°C along with their parental wild-type strains were grown at 30°C and then shifted to 37°C for 2 h before preparing the extracts. For all strains, the extract assays were done at 37°C for relatively short times (15 min) to remain within the linear range and minimize dephosphorylation of the labeled DNA substrate for several mutants whose extracts appear to have elevated phosphatase activity (DnaQ, DnaT, PriA, and DnaB). Table 2 and Table S6 summarize quantitation of the assays, and representative assays for notable mutants are shown in Figure 4 and for the remaining mutants in Figure S7. All values shown in Table 2 and Table S6 were reproducible to within <30% in replicate experiments.
First, the biochemical assays confirmed the function of several candidate retrohoming factors that were expected to be required for top-strand DNA synthesis, including RNase H1 (rnhA), the 5′→3′ exonuclease activity of Pol I (polAexts), and the replicative polymerase Pol III (dnaQ, dnaEts). Extracts from Keio deletions or mutants having temperature-sensitive defects in these activities showed high levels of reverse splicing and cDNA synthesis, but strongly decreased top-strand DNA synthesis (rnhA and dnaQ, 2% wild type; polAexts, 34% wild type; and dnaEts, 10% wild type; Figure 4A, Figure 4C, and Table 2). The ligAts mutant also showed a substantial decrease in top-strand DNA synthesis in the extract assays (45% wild type; Figure 4C), consistent with a role for DNA ligase A in sealing nicks. The incomplete inhibition of top-strand DNA synthesis in the temperature-sensitive mutants may reflect residual activity at the semi-permissive temperature used in the experiment. The Keio deletion of the second E. coli RNase H gene, rnhB, which had no effect on retrohoming in vivo, did not inhibit top-strand synthesis in the in vitro assays (153% wild type; Figure 4A). However, extracts from both the rnhA and rnhB deletions showed elevated levels of full-length intron cDNAs (239 and 218% wild type, respectively), possibly reflecting that RNase H2 makes some contribution to degrading the intron RNA template in vitro.
Among the replication restart mutants, we found substantial decreases in top-strand synthesis in extracts from Keio deletions priA (1% wild type), priC (60% wild type), and dnaT (50% wild type), as well as the temperature-sensitive mutants of dnaB, which encodes the replicative DNA helicase (0%), and dnaC, which interacts with DnaB prior to loading (18% wild type) (Figure 4A, Table 2). The greater effect of the Keio priA deletion in vitro than in vivo may reflect that the short time of the in vitro assays (15 min) favors intermediates with short gaps that are recognized by PriA, while the longer time of the in vivo assays (1 h) favors intermediates with longer gaps that are recognized by PriC [47]. In agreement with Taqman qPCR assays, we found no effect for the deletion of the genes encoding PriB, an accessory protein that facilitates PriA-DnaT complex formation (116% wild type; Figure 4B) [63], nor the DNA helicase Rep (96% wild type; Figure 4A), which ordinarily functions in conjunction with PriC [59].
We also found severe defects in top-strand DNA synthesis in the extracts from temperature-sensitive mutants of two essential proteins that function in replication restart pathways, the single-stranded DNA-binding protein Ssb (ssbts; 8% wild type) [57], [58] and the DNA primase (dnaGts, 0%; [56]) (Figure 4C). Surprisingly, the DNA primase mutant (dnaGts) was also strongly inhibited in bottom-strand cDNA synthesis (<1% wild type; Table 2; see Discussion). Importantly, the LexA SS4610 (lexA51::Tn5) mutant, which has a constitutively induced SOS response [60], showed no significant decrease in top- or bottom-strand DNA synthesis in the extract assays (92 and 121% wild type, respectively; Table 2, Figure S7), consistent with the minimal effect of this mutation on retrohoming in vivo (Table 1 and Figure S5).
Only four other protein mutants showed significantly decreased retrohoming in the extract assays (<70% wild-type top-strand synthesis): the Keio deletions of dinB (Pol IV; 63% wild type), seqA (51% wild type), and stpA (42% wild type; Figure 4A); and a pnp mutant that retains <10% of the wild-type PNPase activity (42% wild type; Figure 4B). Three of these mutants along with the Keio deletion of polB also showed substantially decreased synthesis of full-length bottom strands (dinB, 32% wild type; pnp, 45% wild type; polB, 43% wild type; and stpA, 23% wild type; Table 2). The effect of the DNA repair polymerase mutations is consistent with a possible role in helping to initiate at or traverse the intron RNA/DNA junctions in retrohoming intermediates [12] and/or in initiating replication restart before being replaced by Pol III [64]. PNPase, a component of the E. coli RNA degradosome, has multiple activities that could affect retrohoming, including 3′→5′ exoribonuclease, 3′-terminal oligonucleotide polymerase, high affinity binding to ssRNA and ssDNA, and lower affinity binding to dsDNA [65], [66]. StpA, an H-NS-like DNA- and RNA-binding protein that has RNA chaperone activity, could affect retrohoming by acting on either the target DNA or intron RNA [39], [67], [68]. How SeqA, a negative regulator of the initiation of chromosome replication [69], might contribute to retrohoming is unclear.
The RecJ deletion (single-stranded DNA exonuclease) is noteworthy for showing substantially decreased synthesis of full-length bottom strands (37% wild type; Figure 4A; Table 2) with no decrease in top-strand synthesis. RecJ mutants were identified as retrohoming-deficient both in initial genetic assays [12] and in the transposon screen in this work (Table 1), although not in the Taqman qPCR assays (see above). The effect of the recJ deletion on bottom-strand synthesis in the extract assays is consistent with its previously suggested function in resecting the 5′ overhang of the bottom-strand resulting from the staggered double-strand break made by group II intron RNPs [12].
Notably, we also found strong inhibition of top strand synthesis with extracts from the transposon-insertion mutant C0719, which is at the site of a predicted sRNA (17% wild type; Figure 4B and Table 2). Both reverse splicing and full-length cDNA synthesis were also decreased in this strain (57 and 69% wild type, respectively). How an sRNA might affect retrohoming warrants further investigation.
None of the other mutants tested strongly inhibited retrohoming in the in vitro assays (<70% wild type top-strand DNA synthesis; Figure 4, Table 2, Figure S7, and Table S6). A number of these mutants showed decreased retrohoming efficiencies in in vivo assays and may affect retrohoming indirectly by affecting chromosome structure, DNA replication, DNA target site accessibility, or energy production, e.g., gyrBts (DNA gyrase subunit B); hns (histone-like nucleoid structuring protein); rpoHts and rpoN (RNA polymerase sigma factors); tonB (membrane protein involved in energy production); and tus (DNA termination site binding protein). Notably, the gyrBts mutant showed decreased retrohoming efficiencies in vivo, but elevated levels of both top- and bottom-strand synthesis in vitro (261 and 215% wild type, respectively), possibly reflecting that DNA gyrase impedes retrohoming in wild-type extracts in vitro by unwinding some proportion of the dsDNA oligonucleotide substrate.
Here, we used genetic and biochemical approaches to identify E. coli host factors that function in group II intron retrohoming. First, we used a plasmid-based genetic assay that controls for indirect effects to screen a transposon-insertion library for mutants in which the Ll.LtrB group II intron shows decreased or increased retrohoming efficiency. We then used a Taqman qPCR assay to quantify retrohoming into a chromosomal site in Keio deletions or temperature-sensitive mutants of the candidates identified in this and previous screens. Finally, we compared retrohoming activity in wild-type and candidate mutant strains by using a new biochemical assay that combines Ll.LtrB RNPs with E. coli extracts to reconstitute the complete retrohoming reaction in vitro. Although the initial transposon screen remained vulnerable to false positives, it yielded a manageable group of candidates, whose function in retrohoming was verified by the more direct Taqman qPCR and/or biochemical assays.
Considered together, our results suggest a model for retrohoming of Ll.LtrB intron lariat RNAs in E. coli shown in Figure 5. In initial previously characterized steps, Ll.LtrB RNPs recognize the double-stranded DNA target site and the intron RNA reverse splices into one DNA strand, while the IEP cuts the opposite DNA strand and uses the cleaved strand as a primer for reverse transcription of the reverse-spliced intron RNA [5], [16]. The major host RNase H, RNase H1 encoded by rnhA, degrades the intron RNA template strand during or after cDNA synthesis, leaving residual RNA fragments that could serve as primers for top-strand DNA synthesis. Crucially, after synthesis of a full-length intron cDNA, either the group II intron RT or host DNA polymerase extends bottom-strand synthesis into the 5′ exon, yielding a branched intermediate that is recognized by the replication restart proteins PriA or PriC, which act preferentially on intermediates with short or long gaps between the branch and the 3′ end of the nascent strand [47]. These replication restart proteins then initiate a replisome-loading cascade leading to top-strand DNA synthesis by the host replicative polymerase, Pol III. We find that the 5′→3′ exonuclease activity of Pol I is required for second-strand synthesis during retrohoming, presumably to degrade RNA primers attached to newly synthesized DNA, and Pol I DNA polymerase activity could additionally contribute by helping to fill gaps, both functions of Pol I in host cell DNA replication [70], [71]. Surprisingly, although bottom-strand cDNA synthesis in the extracts is completely dependent upon the RT activity of the LtrA protein, it was nevertheless strongly inhibited in the DNA primase mutant dnaGts, suggesting a previously unsuspected contribution of host factors to initiating cDNA synthesis (see below). The genetic screens also revealed a new putative E. coli RNA degradation pathway that impedes retrohoming and whose disruption leads to increased retrohoming efficiencies.
Importantly, our results confirm that the Ll.LtrB group II intron relies on host DNA polymerases for second-strand DNA synthesis, with a major role for the host replicative polymerase Pol III. The involvement of Pol III in second-strand DNA synthesis was postulated previously based on two findings: (i) that LtrA has low DNA-dependent DNA polymerase activity on artificial substrates in vitro [12], [13], and (ii) that DnaQ and DnaEts mutants are deficient in retrohoming in plasmid-based genetic assays [12]. Here, biochemical assays with cell extracts show that second-strand synthesis is completely dependent upon host DNA polymerases and is strongly inhibited in extracts from DnaQ and DnaEts mutants (2 and 10% wild-type activity, respectively). Notably, although on-going DNA replication may contribute to retrohoming in vivo, we observed the synthesis of a complete second-strand DNA in vitro in the absence of DNA replication.
A major function for RNase H1 in retrohoming is indicated by the findings that a Keio deletion and other mutations in the rnhA gene strongly inhibit retrohoming in genetic, Taqman qPCR, and biochemical assays in this work, and in two different genetic assays used in previous work [12]. By contrast, mutations in the rnhB gene encoding RNase H2 do not significantly inhibit retrohoming or top-strand synthesis in these assays (this work and [12]). Biochemical analysis using extracts from the Keio rnhA deletion strain show that deficiency of RNase H1 results in the accumulation of an intermediate containing the reverse spliced intron RNA, as expected, and that inability to degrade the intron RNA template strand strongly inhibits top-strand DNA synthesis. Extracts from the Keio rnhB deletion also showed some accumulation of the reverse-spliced intermediate, but no deficiency in top-strand synthesis.
A major finding is that replication restart proteins function in retrohoming and are required for second-strand DNA synthesis. We hypothesize that these proteins recognize the branched intermediate formed after RNase H degradation of the intron RNA template strand and extension of intron cDNA synthesis into the 5′ exon and then initiate replisome loading and second-strand DNA synthesis by Pol III by mechanisms similar or identical to those ordinarily used for replication restart at stalled or collapsed replication forks. The replication restart components found here to function in retrohoming by both in vivo and in vitro assays include PriA and PriC, the host proteins that initiate replication restart by recognizing stalled or collapsed replication forks [47]; the accessory proteins DnaC and DnaT [72]; and the replicative helicase DnaB [54]. Our biochemical assays with mutant extracts show directly that all these components are required for second-strand DNA synthesis. During replication restart, PriA recognizes a branched intermediate in which the 3′ OH of the nascent leading strand is close to the replication fork (no gap or a gap of <3 nts), while PriC recognizes an intermediate with a larger gap (>7 nts) [47]. During retrohoming, longer or shorter gaps in the branched intermediate could result from more or less resection of a stalled nascent bottom strand after dissociation of the RT prior to reinitiation of DNA synthesis by a host DNA polymerase. Although the top strand of the retrohoming intermediate contains annealed RNA fragments that result from RNase H digestion and may thus resemble a nascent lagging strand, the location of this strand relative to the branch differs from that at a replication fork, and it is unclear how or if it might also contribute to recognition by PriA or PriC.
The genetic and biochemical assays also indicate a major role in retrohoming for two other essential proteins that function in conjunction with replication restart machinery, the single-stranded DNA binding protein Ssb and the primase DnaG, with mutations in these proteins inhibiting both retrohoming in vivo and top-strand DNA synthesis in vitro. Ssb binds ssDNA regions after unwinding by Rep or PriA and has been shown to physically interact with PriA to stimulate the loading of DnaB at stalled forks [47], [58]. DnaG synthesizes short RNA primers, which are used for initiation of DNA synthesis by Pol III, and triggers the release of DnaC from DnaB [73]. The very stringent requirement for the helicase DnaB in the in vitro retrohoming reaction with a small DNA oligonucleotide substrate could reflect that in addition to DNA unwinding, it is needed to recruit the primase DnaG.
In contrast to other replication restart components, we found no contribution to group II intron retrohoming for PriB, an accessory protein in the PriA-PriB pathway, or Rep, which ordinarily functions together with PriC on the stalled fork by unwinding the dsDNA, especially when the 5′ end of the newly synthesized lagging strand is close to the fork [59], [74]. The dispensability of these factors presumably reflects that PriA can function independently of PriB in the PriA-PriC pathway and that PriC can load DnaB on stalled forks independently of either PriA or Rep [47], [59], [74]. Although the lack of requirement for Rep in vitro could also reflect that the biochemical assay uses a small DNA oligonucleotide substrate, it is nevertheless consistent with the lack of requirement for Rep in our in vivo assays (Table 1; in agreement with [12] but not [22]). An intriguing possibility is that a replisome assembled by replication restart proteins at the site of a group II intron insertion initiates a new round of host DNA replication from this location, thereby rapidly fixing the group II intron insertion into the genome.
The genetic and biochemical assays are consistent with the previously suggested role for RecJ, a 5′→3′ DNA exonuclease, which may resect the 5′ overhang on the bottom strand resulting from the staggered double-strand break made by the group II intron RNP [12]. A surprising finding, however, was that bottom-strand DNA synthesis in extracts requires not only the group II intron RT, but is also strongly inhibited in extracts from DnaG primase mutants and moderately inhibited in extracts from DinB and PolB DNA repair polymerase mutants. The DnaG mutant extracts showed strongly decreased synthesis of even short cDNAs (Figure 4C), suggesting that DnaG may be needed for initiation of TPRT, possibly by functioning in conjunction with DNA repair polymerases to copy the 5′-top-strand DNA overhang before the group II intron RT is engaged to copy the reverse-spliced intron RNA. As LtrA can by itself efficiently initiate TPRT directly from the bottom-strand cleavage site in in vitro reactions with purified RNPs [16], [75], we speculate that host proteins, such as Ssb or RecA, may block initiation by the foreign group II intron RT in extracts, whereas host DNA repair enzymes have mechanisms for overcoming such blocks. Either or both mechanisms for initiation of TPRT could be employed in vivo.
We also identified a number of host proteins in which mutations inhibit retrohoming in vivo, but not in vitro. A number of these proteins act on chromosomal DNA or in transcription (e.g., GyrB, Hns, RpoH, SbcC, Tus) and could impact group II intron retrohoming in vivo by affecting chromosome structure, DNA replication, or target site accessibility. Also affecting retrohoming in vivo but not in our in vitro assay is MnmE, which functions in tRNA modification and may affect the activity or intracellular levels of group II intron RNPs (see also [22], [30]). The failure to observe a decrease in retrohoming activity in some mutant extracts could also be due to replacement of the required activity by other host enzymes.
In addition to mutants that decrease retrohoming efficiency, our transposon-library screen also identified host genes whose disruption leads to increased retrohoming efficiencies. Although we hoped such mutants would identify a variety of host defense factors that function in different ways, all five transposon-insertions that increased retrohoming efficiency in our screen mapped to three closely linked genes associated with a cryptic prophage: rnlA, which encodes RNase LS; yfjK, which encodes a DExH/D-box helicase; and yfjL, which encodes a protein of unknown function. The identity of these genes and the finding that their disruption also leads to elevated GFP expression from a control reporter construct that lacks the Ll.LtrB intron suggest that they suppress retrohoming by degrading group II intron RNAs. Previous studies showed that mutations in RNase E, an essential protein, increase retrohoming frequencies by inhibiting group II intron RNA degradation [12], [22]. Together, these findings indicate that cellular RNases function as a major host defense mechanism for suppressing retrohoming. Additionally, our findings identify a new putative RNA degradation pathway in E. coli K12 that may have been acquired from another bacteria via insertion of a temperate phage and may constitute a second degradosome. The suppression of group II intron mobility by intron RNA degradation in bacteria may be analogous to the suppression of mobility of LINE-1 elements by sequestration into stress granules in human cells [76], [77].
Given our findings for Ll.LtrB retrohoming, we anticipate that replication restart proteins may also function in alternate group II intron retromobility pathways in which a nascent strand at a DNA replication fork rather than a cleaved DNA strand is used to prime reverse transcription of the intron RNA [78]–[80]. In these pathways, which are used by group II introns whose IEPs lack DNA endonuclease activity, reverse splicing is thought to result in the insertion of a group II intron RNP into a DNA target site ahead of a replication fork, with the RT positioned to use either a nascent leading or lagging strand as a primer for reverse transcription, depending upon the strand into which the intron inserted [80]. The stalling of the replication fork when it encounters the inserted group II intron RNP may lead first to dissociation of the replisome, enabling the group II intron RT to access the nascent DNA strand for the priming of cDNA synthesis, and then contribute to its re-recruitment via replication restart proteins for second-strand synthesis and continuation of host DNA replication. We note that yeast mtDNA group II introns primarily use a recombination mechanism rather than replication restart for cDNA integration (see Introduction), and the extent to which the replication restart, DNA recombination, or other pathways are used for the retromobility of different group II introns in different bacteria remains to be elucidated.
Key features of the Ll.LtrB intron retrohoming mechanism delineated here may be relevant to the propagation of LINES and other non-LTR-retrotransposons in eukaryotic nuclear genomes. Non-LTR-retrotransposons, which are thought to be evolutionary descendants of mobile group II introns, encode closely related RTs and use an analogous TPRT mechanism for cDNA synthesis [1], [7]. Like mobile group II introns, most non-LTR-retrotransposons do not encode RNase H and presumably rely on a cellular enzyme to degrade the RNA template strand after cDNA synthesis [7]. The mechanism used for second-strand DNA synthesis by non-LTR-retrotransposons is unknown, but given that non-LTR-retrotransposons carry out reverse transcription in the nucleus could well involve the use of a host DNA polymerase and replication restart proteins as found here for group II introns. The use of replication restart for second-strand synthesis by LINE-1 elements is consistent with their ability to retrotranspose in non-dividing cells [81]. In contrast to non-LTR-retroelements, retroviruses and LTR-containing retrotransposons carry out reverse transcription in the cytosol and rely on RTs that have acquired an RNase H domain and efficient DNA-dependent DNA polymerase activity to synthesize a pre-integration complex containing dsDNA that must then enter the nucleus for integration into the genome [7]. These evolutionary advances, which enable LTR-containing retroelements to carry out major steps of their replication pathway in the cytosol, may contribute to their greater propensity to be transferred horizontally between species and evolve into infectious viruses.
Finally, our results have implications for replication restart pathways. In E. coli, replication restart occurs on stalled or collapsed DNA replication forks and is thus dependent upon on-going DNA synthesis. Surprisingly, our extract assays indicate that replication restart components can synthesize a complete second-strand DNA without on-going DNA replication. These findings resemble recent results for bacteriophage Mu where PriA was found to be required for filling in 5-bp gaps at each end of the Mu insertion in the absence of DNA replication [82]. Thus, replication restart proteins may play a more general role both in the repair of DNA damage and propagation of mobile elements than was thought previously, including as an integral part of the group II intron retrohoming mechanism.
E. coli HMS174(DE3) (Novagen) was used for the transposon library screen and retrohoming assays and DH5α was used for cloning. The construction of the mariner transposon library in HMS174(DE3) was described previously [83]. E. coli Keio deletions and their parental wild-type strain BW25113 were obtained from the National BioResource Project (National Institute of Genetics, Japan). Wild-type SS996 and mutant strains in this genetic background were obtained from Dr. Steven Sandler (University of Massachusetts) [84]. A complete listing of strains and genotypes is given in Table S7.
Cells were grown in Luria-Bertani (LB), Mueller-Hinton (MH) [85], or 2xYT medium [86], as specified for individual experiments. Antibiotics were added at the following concentrations: ampicillin, 100 µg/ml; chloramphenicol, 25 µg/ml; kanamycin, 40 µg/ml; rifampicin, 50 µg/ml; trimethoprim (Tp), 10 µg/ml. Thymine was added at 2 µg/ml.
The intron-donor plasmids pALG2 (Figure S2A; [48]) and pALG3 (Figure 2A) have the vector backbone of pACYC184 with a camR marker and use a T7lac promoter to express an ltrB/GFP fusion cassette, followed by the LtrA ORF. The ltrB/GFP fusion cassette consists of the Shine-Dalgarno sequence and the first eight codons of the phage T7 Φ10 gene linked in-frame to a segment of the L. lactis ltrB gene [58-bp exon 1 (E1), 915-bp Ll.LtrB-ΔORF intron, and 38-bp exon 2 (E2)], with E2 linked to codons 2 to 238 of the GFP ORF. The GFP ORF is derived from pGFPuv (Clontech), which has the amino acid substitutions F64L and S65T to improve performance in FACS assays [87]. The LtrA ORF is cloned downstream of the ltrB/GFP fusion cassette and has its own Shine-Dalgarno sequence. pALG3 additionally contains a trimethoprim-resistance retrotransposition-activated marker (TpR-RAM [24]) inserted at the MluI site in DIV of the Ll.LtrB-ΔORF intron, enabling selection for retrohoming events.
The intron-recipient plasmid pBRR3-ltrB (Figure 2A) is a derivative of pBR322 with an ampR marker [26]. It contains a 45-bp wild-type ltrB target site cloned upstream of a promoterless tetR gene, enabling detection of mobility events either by using the TpR-RAM marker, as in the present work, or by integration of an intron carrying a phage T7 promoter for reporter gene activation [26].
The control plasmid pALE is a derivative of pALG2 that lacks the Ll.LtrB-ΔORF intron and has the ligated ltrB exon sequences linked directly to GFP (Figure S2B; [48]).
pBL1Cap is a broad host range intron-donor plasmid that uses an m-toluic acid-inducible promoter to express the Ll.LtrB-ΔORF intron and flanking exons. It was derived from the broad host range intron-donor plasmid pBL1 [28] by replacing the tetR marker with a camR marker (1.5-kb NheI/PshAI fragment of pACD3 blunt ended and cloned in place of tetR between the FspI sites of the pBL1).
pBL1-rhlE is a derivative of pBL1Cap that expresses an Ll.LtrB-ΔORF intron that was retargeted to insert into a site in the antisense strand of the rhlE gene [21].
The intron-donor plasmid pALG3 and recipient plasmid pBRR3-ltrB were co-transformed into HMS174(DE3) containing random mariner-transposon insertions [83], and the cells were plated on LB medium containing ampicillin and chloramphenicol to select for the markers on the plasmids. Individual colonies were resuspended in 500 µl of LB medium containing the same antibiotics in 96-deep-well plates and incubated overnight at 37°C. A portion (20 µl) of the overnight culture was transferred into 500 µl of fresh LB medium containing ampicillin and chloramphenicol in new 96-deep-well plates, incubated with shaking at 37°C for 2 h, and then induced with 0.5 mM IPTG for 3 h at 30°C. After induction, 105 cells from each sample were transferred to another set of 96-deep-well plates containing 600 µl of MH medium or MH medium with thymine and trimethoprim. The plates were then incubated overnight at 30°C, and the O.D.595 of individual wells was read by a DTX880 multimode plate reader (Beckman Coulter). DNA was isolated by using a Bacterial Genomic DNA Prep kit (Qiagen) and transposon-insertion sites were mapped by Thermal-Asymmetric-Interlaced (TAIL) PCR [25]. Southern hybridizations to confirm that each mutant contained a single transposon insertion at the identified site were as described [83].
E. coli strains carrying plasmids with ltrB/GFP fusions cassettes were grown to log phase (O.D.595 = 0.2–0.4) and induced for 3 h at 30°C with 0.1 mM IPTG (pALG2 or pALE) or 0.5 mM IPTG (pALG3). Wild-type and mutant SS996 (PsulA-GFP) strains were grown at 30°C to O.D.595 = 0.2–0.3 and induced with 4 mM m-toluic acid for 1 h at 37°C. Cells were collected by centrifugation, resuspended in phosphate-buffered saline (140 mM NaCl, 2.7 mM KCl, 9 mM Na2HPO4, 1.6 mM KH2PO4, pH 7.4), and analyzed by using a FACS Caliber (Becton Dickinson), with filter FL1 set at 530±30 nm. FACS data were analyzed with the CELLQuest Pro program (Becton Dickinson).
Retrohoming frequencies in Keio deletion and temperature-sensitive mutant strains were determined by using a retargeted Ll.LtrB-ΔORF intron (rhlE481a) that inserts at a site in the chromosomal rhlE gene [21] and quantifying the 5′- and 3′-integration junctions relative to the number of rhlE genes by Taqman qPCR. For this assay, E. coli strains were transformed with pBL1-rhlE and grown overnight at 37°C in LB medium containing chloramphenicol. The overnight culture was subcultured into fresh medium and incubated at 37°C until O.D.595 = 0.2–0.3. Then, triplicate 3-ml cultures were induced with 4 mM m-toluic acid for 1 h at 37°C. The priA Keio deletion strain was grown and induced at 30°C to avoid the accumulation of suppressor mutations. Five temperature-sensitive mutants (dnaEts, gyrBts, ligAts, rpoHts, and ssbts) that could not be grown at 37°C were grown at 30°C and then shifted to 37°C for 1 h followed by a 1-h induction with m-toluic acid. After induction, cells were lysed immediately and DNA was extracted by using a DNeasy Blood and Tissue Kit (Qiagen).
Taqman qPCR was carried out on 10 ng of total DNA in 384-well plates using a Universal qPCR Master Mix kit (Applied Biosystems), with the following primer/probe sets: (i) 5′-integration junction. P5-forward 5′-GGTGCAAACCAGTCACAGTAATG; reverse 5′-GTCAGCTTCATCGAGGACGAG; Taqman probe 5′-CAAGGCGGTACCTCC; (ii) 3′-integration junction. P3-forward 5′-ATAAAGCCCATGTCGAGCATG; reverse 5′-TGTAAGATAACACAGAAAACAGCCAA; Taqman probe 5′-TGCGCCCAGATAGGGTGTTAAGTCAAGTAGT; (iii) rhlE gene. rhlE-forward 5′-CAGCAACGTCCCGGGG; reverse 5′-ACGCAGTTTCATCATCTGCG; Taqman probe 5′-CCACCAGCACATCAACGCCGC. Taqman qPCR primers and probes with a 5′ FAM (6-carboxyfluorescein) label and 3′ MGB (dihydrocyclopyrroloindole tripeptide major groove binder) quencher were obtained from Applied Biosystems. Standard curves were generated by serial ten-fold dilutions of a TOPO-2.1 vector carrying a cloned DNA fragment containing the Ll.LtrB-ΔORF intron integrated within the rhlE gene. All primer/probe sets had >90% amplification efficiency over the concentration range of the standard curve. Retrohoming frequencies were calculated as the numbers of 5′- and 3′-integration junctions relative to the number of rhlE genes after subtraction of background signal without m-toluic acid induction and are the mean ± standard error of the mean (S.E.M.) for triplicate 1-h inductions.
E. coli S12 extracts were prepared by a modification of procedures used to prepare extracts for in vitro transcription and translation [61], [62]. Cells were grown at 37°C in 200 ml of 2xYT medium at 200 reciprocations per min to O.D.595 = 0.8–1.0, harvested by centrifugation, and washed three times with 20 ml of buffer A [10 mM Tris-acetate, pH 8.2, 14 mM magnesium acetate, 60 mM potassium glutamate, 1 mM dithiothreitol (DTT), and 0.05% (v/v) 2-mercaptoethanol (2-ME)]. The five temperature-sensitive strains (dnaEts, gyrBts, ligAts, rpoHts, and ssbts) that could not be grown at 37°C were instead grown to O.D.595 = 0.4–0.6 at 30°C, then shifted to 37°C and incubated for an additional 2 h. The priA deletion strain was grown at 30°C. After growth, the cells were resuspended in 1.27 ml of buffer B (buffer A without 2-ME) per g wet weight and disrupted by using a BeadBeater [3.24 g of 0.1-mm glass beads (BioSpec Products) per g wet weight; 3 cycles of 1 min at 4°C followed by 1-min incubation on ice]. The crude lysate was centrifuged twice (12,000× g for 10 min at 4°C), and the supernatant was removed and pre-incubated at 37°C for 30 min to complete endogenous reactions and release required components. Extracts prepared without this pre-incubation step showed no detectable top-strand DNA synthesis in retrohoming assays. The extracts were divided into 50-µl aliquots and stored at −80°C. Cell extracts from different strains were checked by Coomassie blue staining of SDS-polyacrylamide gels to confirm that they contained equal amounts of total protein.
The DNA substrates used for extract assays consisted of the top-strand oligonucleotide (5′- GCAACCCACGTCGATCGTGAACACATCCATAACCATATCATTTTTAATTCTACGAATCTTTATACTGGCAAAC) and the bottom-strand oligonucleotide (5′- GTTTGCCAGTATAAAGATTCGTAGAATTAAAAATGATATGGTTATGGATGTGTTCACGATCGACGTGGGTTGC). The strands were annealed by mixing 1 µM of each strand in RNase-free water, heating to 100°C for 5 min, and slowly cooling to room temperature. DNA oligonucleotides were 5′-labeled with 32P using T4 polynucleotide kinase (New England Biolabs), according to the manufacturer's protocol.
Assays were carried out in 20 µl of reaction mixture containing 50 nM DNA substrate, 3 µl of in vitro reconstituted Ll.LtrB RNPs (5–10 µg based on O.D.260; RNPs prepared as described in ref. [13]), 6 µl of S12 extract, 20 µM carrier DNA oligonucleotide (5′-GTGATGTCTGAAAAGAACGGGAAG) as protection against DNase activity, 56.4 mM Tris-acetate buffer, pH 7.5, 100 mM potassium acetate, 35.9 mM ammonium acetate, 24 mM magnesium acetate, 1.5 mM ATP, 1 mM each of dATP, dCTP, dGTP, and dTTP (collectively denoted dNTPs), 500 µM of CTP, GTP, and UTP, 5 mM phosphoenolpyruvate, 50 µg/ml pyruvate kinase, 2 units/µl RNaseOUT (Invitrogen), and 1% (v/v) protease inhibitor cocktail [made by dissolving a mini EDTA-free tablet (Roche) in 1 ml of RNase-free water]. For time-course experiments, the reactions were scaled up to 100 µl, and a 20-µl portion was withdrawn at each time point. Reactions were initiated by adding labeled DNA substrate, incubated at 37°C for times specified for individual experiments, and terminated by extraction with phenol-chloroform-isoamyl alcohol (phenol-CIA; 25∶24∶1 by volume). After digestion with proteinase K (2 µg/µl, 30 min at 37°C) and re-extraction with phenol-CIA, nucleic acids were ethanol precipitated and dissolved in 20 µl of nuclease-free water. For RNase treatment, 0.4 units RNase H (Invitrogen) and 0.1 µg RNase A (Roche) were added, and the sample was incubated for 30 min at 37°C before the ethanol-precipitation step. Samples were analyzed in a denaturing 6% polyacrylamide gel, which was dried and scanned with a PhosphorImager.
For primer-extension analysis, DNA products were separated in a 1% low-melting point agarose gel (Fermentas). A gel slice containing DNAs of 0.85–1.2 kb was heated to 70°C for 10 min and then digested with agarase (Fermentas) for 30 min at 42°C. The released DNAs were ethanol precipitated in the presence of glycogen carrier (Fermentas), washed with 70% ethanol, and dissolved in 10 µl of RNase-free water. For primer extension, a 1-µl portion of the DNA was incubated with 10 µM labeled primer in a 10-µl PCR reaction (95°C, 5 min; then 10–20 cycles of 95°C, 15 sec, 58°C, 30 sec, slow ramp to 72°C, 60 sec; 72°C, 4 min) in an ABI 9700 PCR apparatus. Primers were FB, 5′-TCGATCGTGAACACATCCATAAC; 5T, 5′-TGCTCTGTTCCCGTATCAGCT; and FT, 5′-GTTTGCCAGTATAAAGATTCGTAGAA. Samples were analyzed in a denaturing 6% polyacrylamide gel, which was dried and scanned with a PhosphorImager.
|
10.1371/journal.pntd.0005706 | Efficacy and safety of available treatments for visceral leishmaniasis in Brazil: A multicenter, randomized, open label trial | There is insufficient evidence to support visceral leishmaniasis (VL) treatment recommendations in Brazil and an urgent need to improve current treatments. Drug combinations may be an option.
A multicenter, randomized, open label, controlled trial was conducted in five sites in Brazil to evaluate efficacy and safety of (i) amphotericin B deoxycholate (AmphoB) (1 mg/kg/day for 14 days), (ii) liposomal amphotericin B (LAMB) (3 mg/kg/day for 7 days) and (iii) a combination of LAMB (10 mg/kg single dose) plus meglumine antimoniate (MA) (20 mg Sb+5/kg/day for 10 days), compared to (iv) standard treatment with MA (20 mg Sb+5/kg/day for 20 days). Patients, aged 6 months to 50 years, with confirmed VL and without HIV infection were enrolled in the study. Primary efficacy endpoint was clinical cure at 6 months. A planned efficacy and safety interim analysis led to trial interruption.
378 patients were randomized to the four treatment arms: MA (n = 112), AmphoB (n = 45), LAMB (n = 109), or LAMB plus MA (n = 112). A high toxicity of AmphoB prompted an unplanned interim safety analysis and this treatment arm was dropped. Per intention-to-treat protocol final analyses of the remaining 332 patients show cure rates at 6 months of 77.5% for MA, 87.2% for LAMB, and 83.9% for LAMB plus MA, without statistically significant differences between the experimental arms and comparator (LAMB: 9.7%; CI95% -0.28 to 19.68, p = 0.06; LAMB plus MA: 6.4%; CI95% -3.93 to 16.73; p = 0.222). LAMB monotherapy was safer than MA regarding frequency of treatment-related adverse events (AE) (p = 0.045), proportion of patients presenting at least one severe AE (p = 0.029), and the proportion of AEs resulting in definitive treatment discontinuation (p = 0.003).
Due to lower toxicity and acceptable efficacy, LAMB would be a more suitable first line treatment for VL than standard treatment. ClinicalTrials.gov identification number: NCT01310738.
ClinicalTrials.gov NCT01310738
| Visceral leishmaniasis remains a worldwide public health concern with high mortality even when proper treatment is instituted. There is a need to develop efficacious, safer and shorter treatment alternatives as the current options suffer from high toxicity and long treatment duration. Combination therapies emerge as an alternative, and WHO has encouraged the conduct of studies to evaluate drug combinations where evidence for current treatment regimens is not available. In Brazil, there is no local evidence to support the current treatment recommendations. Therefore, a clinical trial was conducted in five hospitals in Brazil to evaluate the efficacy and safety of the current available treatments—meglumine antimoniate, amphotericin B deoxycholate and liposomal amphotericin B—and of a combination of liposomal amphotericin B, single dose, and meglumine antimoniate, in a shorter administration regimen. Preliminary safety results led to the discontinuation of the amphotericin B deoxycholate treatment arm and a planned interim analysis resulted in the trial interruption. The final results, comparing liposomal amphotericin B and liposomal amphotericin B plus meglumine antimoniate to the standard meglumine antimoniate treatment, did not show a statistically significant difference in cure rates, though cure rate was higher in the liposomal amphotericin B group. Liposomal amphotericin B treatment showed a better safety profile compared to meglumine antimoniate. These results will support future changes in treatment protocols in Brazil and potentially in Latin America.
| Visceral leishmaniasis (VL) is a public health problem in developing countries. The overall annual incidence is estimated at approximately 200,000 to 400,000 cases [1], with a case-fatality rate of 10% per year (i.e. 20,000 to 40,000 deaths per year) [2]. In Latin America, VL is a zoonosis caused by L. infantum with dogs being the main reservoir. More than 95% of VL reported cases in Latin America occur in Brazil, with more than 40% of those in children under 10 years of age [3,4]. In Brazil, the disease has spread from the rural North East region throughout the national territory over the past 30 years, causing epidemics in the fast-growing outskirts of many large and medium-sized cities [5]. Between 1990 and 2014, Brazil reported a total of 78,433 cases, with an average of 3,137 new cases per year [6]. Recent studies have demonstrated increasing case-fatality rates [7] and poor prognosis has been associated with the presence of jaundice, thrombocytopenia, hemorrhage, HIV coinfection, diarrhea, age <5 and age >40–50 years, severe neutropenia, dyspnea and bacterial infections [8]. In Brazil, the current first-line treatment for VL is meglumine antimoniate—MA (20 mg Sb+5/kg/day for 20 days), with amphotericin B deoxycholate—AmphoB (1 mg/kg/day for 14 days) or liposomal amphotericin B—LAMB (3 mg/kg/day for 7 days) being second-line treatments [5]. However, there is no local evidence on the efficacy and safety of these recommended treatment regimens [9]. The national treatment recommendations are based on an expert consensus based on data obtained from studies carried out in other VL-endemic geographies (e.g. Europe, Asia) and local medical practice. As recommended treatments present serious patient management limitations due to toxicity, a need for parenteral administration and hospitalization, the risk of resistance development, and cost, alternative drugs and treatment regimens for VL are needed. Although a series of new, highly active compounds have been identified, these are still in pre-clinical or earlier phases of development. In the short term, organizations such as the Drugs for Neglected Diseases Initiative (DNDi) have adopted a strategy of investigating combination therapies in order to improve current treatment regimens. Clinical trials conducted on the Indian subcontinent and in Africa have shown the efficacy and safety of different treatment combinations: sodium stibogluconate (SSG) plus paromomycin (PM) in East Africa [10], and LAMB plus miltefosine, LAMB plus PM and miltefosine plus PM in India [11].
Based on this evidence, the WHO Expert Committee on the Control of Leishmaniasis reviewed VL treatment guidelines, and recommended the use of SSG plus PM for the treatment of VL in East Africa and drug combinations based on LAMB, miltefosine and PM on the Indian subcontinent. The committee also encouraged the conduct of studies to evaluate drug combinations in those endemic geographies where data on various treatment regimens were not available [12].
Given the potential drug susceptibility variations due to different Leishmania species in Latin America, evidence from other regions cannot be extrapolated to Brazil. In order to provide evidence for the rational use of VL drugs and treatment regimens, the Brazilian Ministry of Health requested and sponsored a clinical trial to assess the safety and efficacy of the three treatments recommended for VL in Brazil. The initial protocol was reinforced with a combination arm of LAMB single dose plus MA for ten days, suggested by DNDi in 2009, after consultation with local experts, investigators and representatives of the National Control Program of the Ministry of Health. The objective of the trial was to assess the efficacy and safety of AmphoB, LAMB and a combination of LAMB plus MA, compared to the standard MA treatment.
The study protocol and further amendments were approved by the institutional Ethics Committees of all the partner institutions: University of Brasilia, Brasilia; Federal University of Piauí, Teresina; Montes Claros State University, Minas Gerais; Pediatric Hospital Joao Paulo II—FHEMIG, Belo Horizonte, Minas Gerais; René Rachou Research Centre, FIOCRUZ, Minas Gerais; Federal University of Sergipe, Aracaju; Sao José Hospital of Infectious Diseases, Fortaleza, Ceará.
The study was conducted in accordance with the declaration of Helsinki, the International Committee on Harmonization guidelines for Good Clinical Practices (ICH—GCP) and all applicable local requirements for the conduct of research on human subjects. A written ‘Informed Consent Form’ (ICF) was obtained from all study participants before enrollment in the trial. Written informed consent was provided by the parents or legal representative of study participants under the age of 18 years and, additionally, minors aged between 12 and 17 years also provided a written assent form. The study was registered in ClinicalTrials.gov under the identification number NCT01310738.
The study was designed as a multicenter, randomized, open label, controlled trial to evaluate the efficacy and safety of AmphoB, LAMB and the combination LAMB plus MA, compared to standard treatment with MA. The patient allocation ratio was 1:1:1:1.
The study started in January 2011. In September 2012 an unplanned interim safety analysis was performed that led to the suspension of AmphoB arm. In April 2014, enrollment of the remaining three treatment arms was interrupted, in accordance with the Data Safety Monitoring Board (DSMB) recommendation, after a planned efficacy and safety interim analysis had been completed. The last 6 months patient follow-up evaluation was concluded in October 2014.
Participants were initially enrolled at four clinical trial sites: Federal University of Piauí, Teresina; Montes Claros State University, Montes Claros; Pediatric Hospital Joao Paulo II—FHEMIG, Belo Horizonte; and Federal University of Sergipe, Aracaju, with the inclusion of a fifth trial site in 2012, Sao José Hospital of Infectious Diseases, Fortaleza, Ceará.
Inclusion criteria were defined as male and female patients aged from > 6 months to < 50 years, with a confirmed diagnosis of VL defined as fever > 37.8°C for at least one week, associated with hepatomegaly or splenomegaly, and a positive result in at least one of the following laboratory diagnostic tests: direct microscopic examination of bone marrow (BM) aspirate or parasite culture, polymerase chain reaction (PCR) on BM aspirate or peripheral blood samples, or the rK39 immunochromatographic rapid test (Kalazar Detect; InBios International Inc., Seattle, USA; IT—LEISH, Bio-Rad, France). Exclusion criteria were pregnancy; HIV infection; underlying chronic or acute disease which would preclude evaluation of the participant’s response to study medication (e.g. diabetes, cardiac, renal, or hepatic impairment, schistosomiasis, malaria, tuberculosis); co-morbidities that may cause alterations of the immune system; any concomitant use of medication that may interfere with the therapeutic response or cause detrimental pharmacological interactions; previous treatment with any anti-leishmanial drugs in the past 6 months; drug abuse; previous history of hypersensitivity reaction to tested interventions; any condition that may hinder compliance with the planned scheduled visits; relapse cases; clinical signs of severe VL disease, such as generalized edema, jaundice, toxemic signs, and severe malnutrition; serum creatinine and bilirubin above the upper normal limit (UNL); prolonged prothrombin time with international normalized ratio (INR) > 2.0; or a platelet count < 20,000/mm³.
The four study arms were (i) MA (Glucantime—Sanofi), 20 mg Sb+5/kg/day intravenous (IV) for 20 days, with a maximum of 1,215 mg pentavalent antimony (Sb+5; three 5 mL vials) per day as the reference arm; (ii) AmphoB, (Anforicin B—Cristália), 1 mg/kg/day IV for 14 days; (iii) LAMB (AmBisome—Gilead), 3 mg/kg/day IV for 7 days; and (iv) a combination of LAMB, 10 mg/kg/day IV, single dose at first day, followed by MA, 20 mg Sb+5/kg/day IV, for 10 days, with a maximum of 1,215 mg of pentavalent antimony (three 5 mL vials) per day, as intervention arms. The choice of MA as the reference arm was based on the prevailing Brazilian VL treatment guidelines at the time when the study was started in 2011 [5], as well as data obtained from a clinical trial that had been concluded in 2009 [13]. AmphoB and LAMB regimens were included as these were routinely used in Brazil and had been included in 2011 into the new treatment guidelines to reduce the VL fatality rate [14]. The same regimen of AmphoB had also been used in the above-referenced clinical trial conducted in 2009 [13]. The selection of the combination therapy with LAMB single dose plus MA was based on the lack of local efficacy evidence for miltefosine and PM, and the satisfactory efficacy rates reported for MA, the first line treatment in the country. Since MA toxicity is dose and time-dependent, with occurrence of cardiotoxicity after three to four weeks of exposure [15], a reduction of treatment duration was expected to result in a significant improvement in safety. Although, based on the current medical practice in Brazil, a maximum daily dose of antimony was used to avoid toxicity, the upper limit dose would not have significantly impacted trial outcome due to the overwhelming proportion of pediatric participants in the study. The LAMB regimen was selected due to the limited data available for Brazil [16] and the encouraging results obtained by Sundar et al. (2010) in India [17]. We decided to use a combination regimen, considering the safety concerns with LAMB doses above 10 mg/kg, the dose used in India as a single-dose administration.
Participants needing rescue treatment were given: LAMB, 3 mg/kg/day IV for 7 days. All participants received supervised treatment.
After the first year of the study, the investigators raised a concern about the number of serious adverse events (SAEs) and treatment discontinuations observed in the AmphoB treatment arm. In order to protect the safety of VL patients, an unplanned safety interim analysis was performed in 2012 to assess the toxicity profile in 79 participants randomized to the four treatment arms (17 to MA, 20 to AmphoB, 22 to LAMB and 20 to LAMB plus MA). The analysis, which included adverse events (AE) related to study medication, showed that treatment with AmphoB was associated with statistically significant higher number of AEs, an increased risk of AEs that led to temporary suspension or treatment discontinuation, and AEs that required medical intervention. AmphoB treatment also showed a higher risk of SAEs when compared to LAMB monotherapy and in combination with MA. Consequently, the DSMB recommended dropping the AmphoB treatment arm from the trial.
As the trial objective was to compare three interventions with the standard treatment (MA), the sample size was determined to allow for detection, with 80% power (1- β) and 5% significance level (α = 0.05), of a at least 8% difference in efficacy of each treatment arm in relation to the reference arm. Assuming a 90% efficacy in the MA reference arm, and adjusting for a 10% loss to follow-up and 10% to maintain the power of comparison between the patient subgroups with parasitological diagnostic confirmation and only a positive rk39 diagnostic test (as defined in the inclusion criteria), it was estimated 165 participants would be required per treatment arm, for a total sample size of 660 participants.
After the AmphoB arm was stopped due to its higher toxicity, the trial’s sample size was reviewed. Adjustment for loss to follow-up was reduced to 5%, in accordance with the observations from the trial up to that moment. No more adjustment was made for those patients diagnosed with only a positive rK39 test, due to the wide validation of these tests by TDR/WHO on three continents, which allows for the inclusion of patients with clinical signs and a positive immunochromatographic test [18]. A final total sample size of 426 was calculated, with 142 participants per treatment arm.
Computer-generated randomization into the four treatment arms was done using the software Quickcalcs—online calculators for scientists (www.graphpad.com/quickcalcs/randomize1.cfm). Blocks of 28 treatment allocations were generated and placed in sealed, opaque envelopes that were sent to each clinical trial site, and only opened by trial clinicians or site investigators when a participant was included in the trial.
After the AmphoB arm was withdrawn, if an enrolled patient was allocated to this treatment arm, the subsequent envelope containing a new code was designated to the patient until they were allocated to any of the three remaining arms. This type of approach did not allow for blinding.
Patients attending the clinical trial sites who were suspected of having VL or who were referred for VL diagnosis confirmation and treatment were submitted to clinical examination, laboratory tests, rk39 immunochromatographic rapid testing and BM aspiration, as per routine case management procedures at the sites. Patients with confirmed VL diagnosis by rK39 test and/or a positive microscopic examination of the BM aspirate were then invited to participate in the trial. Molecular diagnosis was performed in peripheral venous blood and bone marrow samples collected at baseline. Polymerase chain reaction (PCR) was applied in these samples targeting a 120bp minicircle conserved region of Leishmania DNA kinetoplast, as previously reported [19] and the assays were performed later on at the Laboratório de Pesquisa Clínica, Centro de Pesquisas René Rachou, FIOCRUZ, Minas Gerais and at the Leishmaniasis Laboratory of Natan Portella Institute in Teresina, Piauí.
During the informed consent process, they were informed of the trial objectives, procedures, interventions, risks and benefits. Participants then underwent thorough medical history and physical examination, including weight, height, vital signs, and spleen and liver size measurements. Complementary laboratory tests to confirm eligibility, and electrocardiogram (ECG) and chest radiography were performed. Participants who met the trial inclusion criteria were then randomly allocated to one of the four treatment arms.
After treatment was started at day (D) 0, participants were examined daily until they were discharged from hospital at the end of the interventions’ treatment period, and scheduled clinical and laboratory assessments were performed at D0, D3, D7, D14, D21, D30, D60, D90 and D180. ECG was performed at D7, D14 and D21 (and at additional time-points if indicated) and was evaluated by trial physicians. After a safety alert was raised by GlaxoSmithKline (GSK) regarding a potential risk of cardiac arrhythmia due to possible drug interaction between SSG and AmphoB [20], an additional safety assessment, consisting of potassium and serum creatinine measurement and ECG, was included at D1 for participants randomized to the LAMB plus MA combination arm.
The primary efficacy endpoint was cure at 6 months follow-up, defined as complete remission of clinical signs and symptoms up to 3 months after the beginning of treatment, associated with normalization of hematological abnormalities observed at baseline, without evidence of relapse up to 6 months. Abnormal hemoglobin level at three and six months of follow up was taken into consideration when accompanied by other hematological abnormalities. Isolated anemia was not considered as a treatment failure criterion unless other hematological abnormalities or clinical signs and symptoms were present. Any abnormal white blood cells or platelet counting at D90 and D180 precluded clinicians of declaring the cure endpoint. Secondary efficacy endpoints were (i) clinical improvement at D30, defined as fever clearance, improvement or non-worsening of the hematological parameters, and spleen size reduction of any magnitude until D30; (ii) early therapeutic failure, defined as lack of clinical and/or laboratory responses described for ‘clinical improvement at D30’; (iii) therapeutic failure, defined as the absence of improvement or cure, or the need for early treatment interruption and rescue treatment; and (iv) relapse, defined as reappearance of symptoms after a period of improvement or complete remission, once specific VL treatment was completed, and occurring up to the 6 months follow-up visit. Other secondary endpoints included time to fever clearance from D0, mean proportion of spleen reduction at D30, and proportion of participants presenting splenomegaly at D60.
Safety endpoints included clinical and laboratory safety assessments of the different treatment arms as compared with standard treatment with MA. Safety was monitored daily during the hospitalization period and at the scheduled follow-up visits, through clinical examination and registry of participants´ complaints. Complete blood count, biochemistry tests and ECG, were performed according to scheduled evaluation times and whenever indicated by the investigator. AEs were recorded during the whole trial period and evaluated according to the Division of AIDS Table for Grading the Severity of Adult and Pediatric Adverse Events (DAIDS AE grading table), version 1.0, December 2004 [21]. Trial physicians performed AE causality assessment, and AEs were classified as related or not-related to the trial medications.
Participants withdrawn from the trial due to the occurrence of an AE received rescue treatment as described above, and were considered as an early treatment failure. Study coordination and Institutional Ethics Committees were notified of all SAEs, and an independent medical monitor (IMM) evaluated all SAE reports.
Data were registered in a specific Case Report Form (CRF) from inpatients’ records and standard trial records were designed for the scheduled follow-up visits. Trial coordination designated monitors performed source document verification for all CRFs on site. Double data entry was performed in the Statistical Package for the Social Sciences—SPSS (IBM Statistics version 22).
The primary efficacy endpoint of cure at 6 months was analyzed as to the intention-to-treat (ITT) and per-protocol (PP) approaches. The ITT analysis included all participants randomized to the remaining three interventions, except for one participant randomized to the MA arm and withdrawn from the trial at D1 after a stool tested positive for schistosomiasis. Therefore, the ITT analysis included a total of 332 participants. The PP analysis excluded participants lost to follow-up and withdrawn from the trial because of the occurrence of AE/SAE.
Descriptive data were summarized using percentages, medians or means with their respective 25 and 75 percentiles or standard deviations as appropriate. Parametric (student T test) and non-parametric statistics (Mann Whitney U test) were used for comparison of continuous data, taking into consideration the normality of data distributions evaluated using the Kolomogorov-Smirnoff test and visual exploration of plotted data. Categorical comparisons were performed using the chi-square test or the exact Fisher test as appropriate. For all comparisons, 95% CI were calculated for the observed differences, as recommended by Altman et al. [22]. All comparisons were performed using SPSS software (IBM statistics version 22).
In Setember 2012, an unplanned interim safety analysis showed that treatment with AmphoB was associated with higher toxicity, resulting in the interruption of the allocation of participants to this intervention arm, following DSMB recommendations.
In 2014, a planned efficacy and safety interim analysis was performed after 50% of the recruited participants completed 6 months follow-up. The data showed cure rates as per ITT of 77.1% for MA standard arm treatment, 86.1% for LAMB monotherapy, and 83.1% for LAMB plus MA combination treatment. The ITT efficacy analysis comparing the intervention groups against the MA comparator did not show statistical significance (S1 Table). Considering the cure rate observed in both intervention arms and the cure rate observed in the MA comparator arm, efficacy assumptions would require adjustment leading to a significant increase of the sample size per treatment arm in order to maintain the 80% study power, hindering the conduct of the trial. Therefore, the DSMB recommended that patient recruitment be stopped and the trial was interrupted in April 2014, with a total of 378 patients randomized to four arms, who were followed until completion of the 6 months follow-up period. This publication presents the final efficacy and safety results of the trial.
As shown in Fig 1, a total of 1,222 patients with confirmed VL were screened, of which 378 (31%) were randomized to the four treatment arms. The main screening failure reasons were age outside the trial range (18.1%), abnormal laboratory tests as defined in the exclusion criteria (11.5%), severity of the illness (11.0%), specific treatment exposure before trial enrollment (10.1%), and a positive HIV test (9.8%). The sample was composed mainly by pediatric participants who were at least moderately ill.
Among the 332 participants considered for these analyses, eighteen mild to moderate protocol violations were recorded during the study period. Three participants were randomized in spite of fulfilling exclusion criteria: (i) one in the LAMB group with bilirubin level above the UNL, and (ii) two in the MA group, of whom one had schistosomiasis diagnosed after randomization and of whom one was diagnosed with severe malnutrition. For two participants randomized to the MA group bilirubin or creatinine levels was not measured at baseline. Eight participants received one or two more doses of the treatment medication and two participants had treatment with MA with an interval between doses exceeding 72 hours. One participant did not receive the last MA dose and one did not receive the last two doses. Another patient randomized to the combination treatment arm received the single dose LAMB after completing MA administration. Though characterizing protocol deviations, all those participants were maintained in the PP analysis with their actual outcome registry as per evaluation at 6 months follow-up.
Nine participants had their six months follow-up visit after the allowed window (ranging from one day to 43 days) and were included in the ITT analysis as cured since if they were free of relapse after the allowed period, they were also free of relapse at 6 months. One patient had the six months visit 12 days before the allowed window, and as it was verified that he continued to be free of relapse after 6 months, he was also considered as cured for ITT analysis.
As shown in Table 1, treatment groups are comparable for baseline demographic, clinical and laboratory variables.
VL diagnosis was confirmed by rk39 rapid test and by direct examination of bone marrow aspirate in 310 out of 331 (93.9%) and 175 out of 283 (61.8%) participants respectively. Bone marrow cultures were positive in 130 out of 275 participants (47.3%). PCR on peripheral blood and bone marrow samples showed positive results in 229 out of 240 participants (95.4%) and PCR in peripheral blood was positive for 213 out of 247 participants (86.2%).
As shown in Table 2, the efficacy at 6 months follow-up (primary endpoint) as per ITT analysis was 77.5% in the MA arm, 87.2% in the LAMB arm and 83.9% in the LAMB plus MA arm. All patients who were designated as initial failure and relapse, who were withdrawn due to the occurrence of AE/SAE, and who were lost to follow-up were considered as treatment failure in the ITT analysis (S2 Table). There was no statistically significant difference in the cure rates between each treatment intervention and the comparator.
The per protocol analysis (Table 3) that excluded participants withdrawn from the study due to AE/SAE and patients lost to follow-up shows a cure rate at 6 months of 94.5% in the MA arm, 92.2% in the LAMB arm and 98.9% in the LAMB plus MA arm. Again, the difference between each treatment arm and the comparator was not statistically significant.
As shown in Table 4, the rate of early withdrawal due to the occurrence of AE/SAE was 13.5% for MA (15/111), 0.92% for LAMB (1/109) and 8.9% for the combination LAMB plus MA (10/112), with a statistically significant difference between the LAMB treatment arm and the comparator (P- value < 0.001).
There was no statistically significant difference between each treatment arm and the comparator for the secondary endpoints of clinical improvement until D30 and relapse until D180 (S3 and S4 Tables). Survival analysis of time until fever clearance for 287 patients for whom data was available (Fig 2) did not show statistical differences between treatment arms (S5 Table). In all treatment arms there was a statistically significant reduction in spleen size between screening and D30, with a similar absolute difference in the mean reduction of spleen size of around 6 cm in all groups (S6 Table). At D60, there was no statistically significant difference between treatment intervention and the comparator in the proportion of participants still presenting splenomegaly (S7 Table). The data on spleen size reduction should be handled with caution because of limitations in spleen size measurement due to the inter and intra-observer variability, as there was no reliability assessment performed between study sites and between observers in the same trial site at the beginning of the trial.
Safety data were analyzed per participant and per event. The per participant analysis (Table 5) shows that 239 participants out of 332 (71.2%) presented at least one medication-related AE and that 66/332 (19.9%) presented at least one medication-related SAE. 61 participants out of the 66 who presented SAE (92.4%) presented one SAE, and 5 participants presented two SAEs (7.6%). There was no statistically significant difference in the percentage of participants presenting AE or SAE between each treatment arm and the comparator. The patient withdrawn in D1 due to a diagnosis of schistosomiasis did not present AE during his participation in the trial.
When analyzing AE severity, results show a statistically significant difference in the proportion of participants who presented at least one AE classified as severe (grade 3 and 4 according to DAIDS AE grading table [21] between the LAMB arm and comparator.
In the per event approach (Table 6), a total of 568 medication-related AEs were observed among the 239 patients who presented related AEs, of which 71 fulfilled the definition of SAE. Of these 568 medication-related AEs, 208 (36.6%) occurred in participants randomized to the MA arm, 143 (25.2%) in participants randomized to the LAMB arm and 217 (38.2%) in patients randomized to the LAMB plus MA arm. There was a smaller proportion of related AEs in the LAMB arm compared with MA comparator (p = 0.045). As far as SAEs are concerned, 21/71 (29.6%) occurred in the MA arm, 22/71 (31.0%) in the LAMB arm and 28/71 (39.4%) in the LAMB plus MA arm. There was no statistically significant difference between treatment arms and the comparator.
There were 51/568 (9%) of medication-related AEs resulting in temporary or definitive treatment suspension. For 22 episodes (3.9%), treatment interruption was transitory and for 29 occurrences (5.1%) treatment interruption was definitive. The proportion of medication-related AEs that resulted in definitive treatment suspension and need for rescue treatment in the LAMB intervention arm was shown to be significantly lower when compared with MA standard treatment (p = 0.003). There was no statistical difference in the proportion of definitive treatment suspensions between the LAMB plus MA arm and the comparator, nor in the proportions of transitory interruptions between each treatment arm and the comparator. In the MA and LAMB plus MA arms, the AES / SAEs that led to treatment discontinuation were mostly due to the increase of pancreatic and liver enzymes (10/16 or 62.5% and 6/12 or 50% respectively) and cardiac toxicity (2/16 or 12.5% and 4/12 or 30% respectively). One patient was discontinued in the LAMB arm due to a grade 4 increase in liver enzymes.
One of the 71 reported medication-related SAEs resulted in death. This participant was randomized to the MA treatment arm, presented respiratory and hemodynamic worsening on D3, was transferred to an intensive care unit and received rescue treatment with LAMB, but died of presumed sepsis on D11. Another patient randomized to the LAMB plus MA arm died on D2, due to a generalized infection by Pseudomonas aeruginosa confirmed at autopsy. This death was evaluated by the IMM as not being related to the study medication.
Treatment-related AEs reported in this trial are consistent with the expected toxicity profile of the drugs. S8 Table in supporting information shows treatment-related AEs per system organ class and intervention arm. Most of the treatment-related SAEs resolved during the hospitalization period. Participants presenting AE/SAE that led to their withdrawal from the trial and administration of rescue treatment were followed until complete resolution of the AE/SAE. All non-fatal medication-related SAEs that did not result in treatment interruption resolved within the 6 months follow-up period.
The present trial is the largest clinical trial ever conducted in the Americas to evaluate the efficacy and safety of the current recommended treatments for VL. The ITT analysis reported herein demonstrates that the overall cure rate obtained in the MA arm was lower than expected (77.5%). This precluded the possibility of proving a statistically significant difference compared with the LAMB arm (87.5%) in spite of the 9.7% crude efficacy difference between those groups favoring the LAMB arm. This is noteworthy: prior to the present trial, the general specialist opinion was that MA could achieve higher efficacy rates. The cure rate observed in the combination arm (LAMB plus MA) was also better than MA, but the crude efficacy difference was lower than for LAMB alone (6.4%).
The PP analysis, which focused on patients completing treatment, showed higher cure rates than those observed with the ITT approach. The PP analysis was essential to suggest that, once overcoming the important toxicity issues related to the administration of MA, patients exposed to MA in monotherapy or in combination present a tendency towards a higher cure rate. Those data reinforce the relevance of toxicity issues during VL treatment as important determinants of final outcomes and also call attention to the need for rescue treatment with the less toxic therapeutic option.
These final trial results are similar to those obtained in the planned efficacy and safety interim analysis that showed efficacy at the end of 6 months follow-up by ITT and PP analyses of 77.1% and 94.7% for MA, 86.1% and 91.2% for LAMB, and 83.1% and 98.3% for LAMB plus MA combination treatment, respectively, with no statistical significant differences in the cure rate between each treatment arm and the comparator for either set of analyses.
The absence of statistical significance between treatment arms and the comparator in the final analysis was expected. Thus, although a greater crude efficacy difference (9.7%) was observed than the 8% assumed a priori when designing the trial, the lower than expected MA efficacy (90%) meant that the study was not sufficiently powered.
Another study limitation is the fact that only non-severe cases of the disease and non-HIV infected participants were included, and therefore external validity of the trial results is only applicable to non-HIV infected patients who do not present severe illness.
Although efficacy evidence is scarce in Brazil and comes mostly from observational studies, high cure rates are reported in patients treated with MA. A retrospective study conducted at the Hospital of the University of Mato Grosso do Sul, Central-West Region of Brazil in 111 children treated with MA, showed a 96.9% efficacy rate in mild to moderate cases (93/96 patients who completed the standard 20 to 30 days treatment course) and over 64% in severely ill patients (7/11 patients). [23]. However, another retrospective observational study conducted in 89 patients with VL admitted in the Teaching Hospital Dr. Hélvio Auto in Maceió, State of Alagoas and treated with standard dose of MA for 1 to 40 days, for an average of 24.42 days, reported a 83.14% cure rate at 6 months for those 74 patients that had completed treatment; the study also showed that in 13.5% of the patients treatment had to be changed to AmphoB due to the occurrence of adverse reactions and three patients died (3.37%) [24]. Results of the present trial show an ITT efficacy for LAMB inferior to 90%, similar to previous studies: a clinical trial conducted simultaneously in India, Kenya and Brazil showed a 87% (13/15) efficacy of a LAMB regimen of 2 mg/kg/day on days 1 to 10 [16], while an efficacy superior to 90% has been observed in clinical trials conducted in other countries [25, 26]. Trials conducted in India in uncomplicated VL with different dose regimens showed LAMB efficacy greater than 90% [17, 27, 28, 29, 30]. Among these, two open-label, randomized trials using LAMB doses of 5 mg/kg reported efficacies of 91% at six and nine months (LAMB single dose) and 93% at six months (LAMB 5 mg/kg over 5 days) [27, 30].
In 2014, based on the efficacy results of single dose LAMB [17] and combination therapies of LAMB plus miltefosine, LAMB plus PM and miltefosine plus PM [11], the Indian National Road Map for Kala-Azar Elimination recommended the use of 10 mg/kg single dose LAMB as a first line treatment for VL patients in India, with PM and miltefosine as a second option at all levels [31].
In European countries, clinical trials have also shown efficacy of LAMB superior to 90% [32, 33, 34]. In Italy, an open label, dose-finding study in 88 patients showed 91% efficacy at 12 months with a LAMB regimen of 3 mg/kg on days 1 to 4 and 10 [33], though efficacy of 75% was found with the same total dose of 15 mg/kg in another dose-finding trial conducted in the same country in 106 children [32]. In both studies, total LAMB dose of 18 mg/kg (3 mg/kg on days 1 to 5 and 10) yielded a 98% efficacy rate at 12 months. LAMB total doses of 20 mg/kg and 21 mg/kg, as used in the present trial, have showed cure rates ranging from 90% to 100% [25, 32, 34].
Though the evidence is scarcer, LAMB does not appear as efficacious in Africa. While total doses of 14mg/kg and 10 mg/kg showed respectively 100% and 90% efficacy in a small number of patients in Kenya [16], recently, a multi-center, open-label, non-inferiority, randomized controlled trial conducted in 124 patients in Ethiopia and Sudan to compare the efficacy and safety of a single dose and multiple doses of LAMB showed efficacy at 6 months of 40% for a LAMB single dose of 7.5 mg/kg, 58% for a LAMB single dose of 10 mg/kg, and 85% for a LAMB regimen of 3 mg/kg on days 1–5, 14, and 21 [35]. The trial was terminated because of the low efficacy of the tested regimens.
Though results from other countries cannot be extrapolated to Brazil, the results of this trial suggest that there is room to improve LAMB efficacy, increasing the total dose by adjusting daily doses or length of treatment. Also, the hypothesis that parasite species (L. donovani in India and Africa, and L. infantum in Latin America and the Mediterranean region) and their susceptibility to drugs are determinants of the therapeutic outcomes observed across continents should be better explored in future trials. Furthermore, other alternative drug combinations could be explored as new chemical entities are progressing through development.
For the treatment of VL caused by L. infantum, WHO recommends, in order of preference, LAMB 3–5 mg/kg per daily dose given over a 3–6 days period, up to a total dose of 18–21 mg/kg [12]. PAHO incorporated this recommendation for the treatment of VL in the Americas, though the quality of evidence was evaluated as very low [36].
In Brazil, LAMB is indicated for patients aged < 1 year and > 50 years, severe illness based on severity score, renal, hepatic or cardiac insufficiency, HIV-coinfected patients, or other conditions leading to immunodeficiency, therapeutic failure to MA or other contraindications of use of MA [14, 37, 38]. The widening of LAMB indications in the country was motivated in part by the results of the unplanned safety analysis performed for the present trial, which showed increased toxicity in patients exposed to AmphoB.
Relapse was evaluated as a secondary endpoint; nonetheless, considering the scarcity of the event observed in the study (nine cases in 315 patients followed until six months), this issue remains to be addressed in future studies, principally in patients exposed to LAMB who presented the higher absolute number of relapse events in this trial (5/104) or when new interventions are assessed.
Regarding safety, there was no statistical significance between each treatment arm and the comparator in terms of the proportion of patients who presented medication-related AE/SAE, though the proportion of patients presenting AEs was lower in the LAMB intervention. However, patients exposed to LAMB presented a statistically significant lower frequency of AEs as compared to standard treatment. The proportion of AEs and SAEs among the total events observed in the trial was higher in the combination arm (217/568 or 38.2% and 28/71 or 39.4%, respectively) and SAE occurrence was mostly associated with MA toxicity.
A statistically significant difference between LAMB and the comparator MA arm was observed when comparing the proportion of patients presenting at least one severe AE (p = 0.029), as well as in the rate of early withdrawal due to the occurrence of AE/SAE (P < 0.001) and in the proportion of medication-related AEs that resulted in definitive treatment suspension and the need for rescue treatment (p = 0.003). Although there was no statistical difference in the proportion of transitory treatment suspensions between LAMB and comparator MA arm, participants exposed to LAMB presented a lower absolute number of temporary treatment interruptions (three vs eight in the MA arm and 11 in the LAMB plus MA arm).
The frequency of treatment interruption reported in the analysis of early withdrawal due to AE/SAE was lower than the frequency of definitive interruption caused by AE/SAE, because in the first analysis the participant was the analysis unit, while in the second, AE/SAE was the analysis unit; therefore, the occurrence of various concomitant AEs might have justified the treatment suspension in a single participant.
Overall, LAMB intervention showed an acceptable cure rate, and also showed a statistically significant better safety profile. This better tolerability profile, associated with a shorter administration time (seven days regimen vs 20 days for MA and 11 days for LAMB plus MA) and thus a reduced hospitalization period, suggest that LAMB monotherapy could be recommended as a more suitable first line treatment option for VL in Brazil and possibly in other Latin American countries where patients show a similar clinical profile.
The cost of medication is still a limiting factor for widening the use of LAMB. However, an agreement between the Brazilian Ministry of Health and the manufacturer of LAMB (Gilead) allows the drug’s purchase at a WHO negotiated price of 18 USD per 50 mg vial. Furthermore, the possibility of pooled purchase of the product through the PAHO Strategic Fund offers an opportunity for further price reduction. On the other hand, the higher costs of LAMB treatment should be put into context: toxicity observed with the current standard MA treatment results in costs associated with drug toxicity handling, longer hospitalization due to longer administration times, and ultimately, if treatment is interrupted, the need for rescue treatment. Such cost difference in treatment regimens should be evaluated in a cost-effectiveness study, which could provide additional evidence for the Ministry of Health in its potential revision of national VL treatment guidance.
The results of the trial point towards a recommendation for the use of LAMB as the first line treatment for VL in Brazil and probably in other Latin American countries, due to its acceptable efficacy profile and its lower toxicity and shorter administration time as compared to MA and LAMB plus MA combination therapy.
|
10.1371/journal.pntd.0004392 | Ongoing Transmission of Onchocerca volvulus after 25 Years of Annual Ivermectin Mass Treatments in the Vina du Nord River Valley, in North Cameroon | Recent reports of transmission interruption of Onchocerca volvulus, the causing agent of river blindness, in former endemic foci in the Americas, and more recently in West and East Africa, raise the question whether elimination of this debilitating disease is underway after long-term treatment of the population at risk with ivermectin. The situation in Central Africa has not yet been clearly assessed.
Entomologic data from two former endemic river basins in North Cameroon were generated over a period of 43 and 48 months to follow-up transmission levels in areas under prolonged ivermectin control. Moreover, epidemiologic parameters of animal-borne Onchocerca spp. transmitted by the same local black fly vectors of the Simulium damnosum complex were recorded and their impact on O. volvulus transmission success evaluated. With mitochondrial DNA markers we unambiguously confirmed the presence of infective O. volvulus larvae in vectors from the Sudan savannah region (mean Annual Transmission Potential 2009–2012: 98, range 47–221), but not from the Adamawa highland region. Transmission rates of O. ochengi, a parasite of Zebu cattle, were high in both foci.
The high cattle livestock density in conjunction with the high transmission rates of the bovine filaria O. ochengi prevents the transmission of O. volvulus on the Adamawa plateau, whereas transmission in a former hyperendemic focus was markedly reduced, but not completely interrupted after 25 years of ivermectin control. This study may be helpful to gauge the impact of the presence of animal-filariae for O. volvulus transmission in terms of the growing human and livestock populations in sub-Saharan countries.
| Over the past decades the Fight against river blindness, a tropical disease caused by a nematode worm, has been relatively successful, and a number of countries have been reported to be free of parasite transmission. In North Cameroon, we checked the occurrence of infective stages of Onchocerca volvulus in the transmitting black fly populations for more than three years and were able to confirm that the transmission there is low, but not yet interrupted. In a second location on a highland plateau, however, no infective stages of the human parasite were found. Instead, a closely-related parasite of cattle was present in both places. Given that the areas are not far away from each other and the biting frequencies of the black fly populations are similar, the historically earlier and higher density of cattle herds in one of the regions would explain why it is now free of the parasite due to the effects called zooprophylaxis and cross-reacting premunition. Changes in the socio-economic environment, especially the increase of human and cattle populations have a strong influence on the spread of river blindness in Africa.
| The interruption of transmission of Onchocerca volvulus, the causing agent of river blindness or onchocerciasis, has been confirmed for a growing number of endemic foci on the American continent [1,2,3] and in West [4] and East Africa [5,6,7]. The recent success in onchocerciasis control can be mainly attributed to the extensive and sustained mass treatment programs with the microfilaricide ivermectin, governed by institutions of the World Health Organization, like the African Programme for Onchocerciasis Control [8]. Long treatment rounds are necessary because the drug is only lethal to the larval stage and not the adult worm.
There are thus good prospects that elimination of onchocerciasis is well underway in the Americas [9] and may also have begun at different foci on the African continent [6,10,11]. However, currently there is a paucity of information on the actual situation in Central and Southern Africa, in particular with respect to vector transmission, albeit a significant proportion of these regions have been hyperendemic. Recent studies on the effects of ivermectin treatment on the epidemiology of O. volvulus in humans and the black fly vector Simulium damnosum sensu lato have been done in North and West Cameroon [12,13,14,15,16]. The caveat of the most recent studies is that the filarial species in the vector were not always correctly identified, and the prevalence of infective O. volvulus larvae and thus the transmission potential remains unknown. Local S. damnosum s.l. populations are vectors of at least two other species from the Onchocerca genus: O. ochengi, a common parasite of Zebu cattle Bos indicus [17] and O. ramachandrini, a filaria from the warthog Phacochoerus africanus [18]. The proportion of animal-filariae in the vector has direct and indirect consequences for parasite transmission to humans [19,20,21] rendering it an important factor to understand the epidemiology of river blindness. Furthermore, filariae closely-related to O. volvulus might repopulate the human host [22] posing a potential risk of infection, or they might transfer genes to O. volvulus which negatively affect the effectiveness of ivermectin, presently the sole drug intervention in use [23]. For this reason we combine microscopic differentiation of infective larvae with a PCR-based molecular approach which allows the detection of yet unknown filarial species and strains in addition to already known Onchocerca spp. [22].
This study presents the latest entomologic data of a longitudinal study in the Vina du Nord valley, North Cameroon, dating back to 1976 when ivermectin mass treatment had not yet commenced [24,25]. The impact on O. volvulus transmission after 25 years of annual community-directed treatment with ivermectin (CDTI) is demonstrated here. Furthermore, a second site endemic for onchocerciasis in an economically important cattle livestock region has been monitored since 1985. The epidemiologic data is also complemented with Onchocerca spp. transmitted by the same local vectors of the S. damnosum complex and discussed in light of their impact on transmission success of O. volvulus. We have not studied onchocerciasis transmission in regions where ivermectin treatment is contraindicated, such as co-endemic foci of Loa loa in the Central African rainforest, although they remain potential source-areas for reinvasion.
Two S. damnosum fly catching sites at two foci in Northern Cameroon were visited between two to four times per month (Fig 1). One former hyperendemic onchocerciasis focus is the village Soramboum close to the Vina du Nord river in the Sudan savannah (500 m altitude): 7°47'14"N; 15°0'22"E where ivermectin mass treatments have been conducted since 1987. Here we present entomological data collected from September 2009 till March 2013. The village has approximately 1000 inhabitants today, and between 1000 and 2000 cattle are located in the vicinity (personal observation). The Vina du Nord river is flowing perennially with an average annual water discharge of 150 m3 per second, with highest values between July and October [26]. The other formerly hypo- to mesoendemic focus monitored is the village Galim located 15 km south of Ngaoundéré (population: 500 inhabitants, approximately 5000 cattle in the vicinity, personal observation) at the Vina du Sud river (mean annual water discharge: 37 m3/s, 1050 m altitude): 7°12'2"N; 13°34'56"E, where CDTI has started in 1997. The entomological data was collected from April 2009 till March 2013. The area belongs to the Guinea-grassland of the Adamawa plateau, located in an important area for cattle livestock production in Cameroon [27]. Baseline and follow-up data of O. volvulus transmission to man before ivermectin mass treatments started is available for both foci [15,16,19,25,28], and publicly available via the project website www.riverblindness.eu (http://riverblindness.eu/epidemiology/fly-catching-sites-data/). Whereas S. damnosum sensu stricto and S. sirbanum are the predominant vector species at the Vina du Nord river, it is S. squamosum at the Vina du Sud river [29].
Fly catching was performed according to Duke et al. [30] with the following modifications. Blood meal-searching female Simulium flies were attracted on man by exposing the fly catcher’s legs and trapped with sucking tubes as soon as they settled before starting to probe. Usually the catching period was from 7 am to 6 pm, interrupted by a one hour break at noon. Afterwards, the catches were transported to the research station in Ngaoundéré and stored at -15°C until dissection.
Flies were sorted, counted, and a maximum of 100 female S. damnosum s.l. flies per site and day were dissected with needles under a stereomicroscope (Wild M5, Switzerland). The parous rate was determined by examination of the ovaries in the abdomen [31]. From parous flies infested with filarial worms the location (head, thorax or abdomen), molting stage and quantity was noted. Following the identification key of Wahl et al. [15], infective third-stage larvae (L3) were classified to species according to body length, measured by an ocular eye-micrometer attached to the stereomicroscope at 50x magnification, and shape of the anterior and posterior ends. For a subdivision of the L3 taken between February 2010 and February 2012, a molecular investigation of their mitochondrial DNA was conducted according to Eisenbarth et al. [22]. Briefly, single L3’s were digested with 1 to 2 μl proteinase K (20 μg/μl stock) in 75 μl DirectPCR lysis reagent (Viagen Biotech, USA) at 55°C. Two μl of the crude extract was used for each 25 μl PCR reaction. Primer pairs of three mitochondrial loci (12S rRNA, 16S rRNA and coxI) that allow for the discrimination of filarial species were used. The amplified PCR products were sequenced on an ABI Prism 3100 genetic analyzer (Applied Biosystems, USA) following the manufacturer’s protocol. For the comparison of the body lengths of L3's identified by molecular markers, a larger sample size was taken from flies caught in the same period at the Vina du Sud river about one kilometer downstream of the site near Galim. These flies were collected both from man and cattle.
The Annual Biting Rate (ABR) and Annual Transmission Potential (ATP) of Simulium damnosum flies were determined according to the literature [24,25,32]. First, the monthly biting rates (MBR) were calculated by the sum of the flies caught per month divided by the number of catching days and multiplied by the number of days per month. No correction was made for the missing hours due to rain, sandstorm, or any other reasons. The ABR is the sum of 12 MBRs per hydrological year, measured from April (beginning of rainy season) till March next year (end of dry season). For months during which no catches were attempted, the mean MBR value for the corresponding month and site over the respective decade was estimated by interpolation. The monthly infection rate was the sum of the infective L3 of O. volvulus, O. ochengi and O. ramachandrini from the head, thorax and abdomen found in all parous flies, divided by the sum of dissected flies. By multiplying the monthly infection rate with the respective MBR, the Monthly Transmission Potential (MTP) was determined. The ATP is the sum of 12 MTPs for one year, starting from April till March next year. Missing data points were extrapolated by the sum of all MTP divided by the number of months with data, and multiplied by factor 12. If the MTP data per year was below 3, the mean annual infection rate of proximate years multiplied with the respective ABR was used instead for the ATP calculation.
The statistical program Python version 3.4.1 was used for statistical analysis employing student t-tests. Results were considered statistically significant when the p-value was below 0.05. P-values were corrected for multiple testing by multiplying with the number of tests done. The effect size was calculated according to Cohen [33]. For depicting the distribution of the L3 body length from a random sample, violin plots, i.e. box plots with a rotated kernel density plot on each side, were used.
During the study period, a total of 39,082 flies were caught on human fly catchers, and 21,897 (55.6%) of them were dissected: a total of 2096 L3 were found (Table 1). Depending on the catching site, a mean of 1.96 ±0.53 L3 (max. 20) were harvested per infective fly in Soramboum and a mean of 3.49 ±0.13 L3 (max. 23) in Galim. Near Soramboum at the Vina du Nord site the infection rate (flies carrying L1, L2 and L3) of parous flies in the rainy season was significantly higher than in the dry season (mean: 10.9 vs. 5.9%, t-value = -4.91, p < 0.001), as well as the infection rate with infective L3 stages (mean: 7.8 vs. 3.3%, t-value = -4.88, p < 0.001). The opposite was true at the Vina du Sud site near Galim (mean infection rate: 11.1 vs. 16.6%, t-value = 2.36, p > 0.05; mean L3 infection rate: 4.3 vs. 7.1%, t-value = 1.89, p > 0.05). Moreover, the parous rate, which is a parameter of age structure, was on average 9.6% higher in the wet season than in the dry season in Soramboum (range: -18.3–19.3%, t-value = 5.31, p < 0.001), but in Galim on average 11.8% lower in the wet season compared to the dry season (range: 6.7–24.3%, t-value = 3.83, p < 0.001). The higher proportion of infective L3 to developing larvae during the rainy season in Soramboum (64.0% vs. 53.5%, t-value = -1.88, p > 0.05), but not in Galim (33.9% vs. 35.3%, t-value = 0.88, p > 0.05) is lacking statistical support. A generally higher proportion of L3 in Soramboum can be explained by a longer storage time of the caught flies at ambient temperature during the time (often days) until they were brought to the laboratory, 225 km away by public transport, so that more larval stages developed further.
Fig 2 shows the ABR for the two study sites starting prior to the distribution of ivermectin. With the exception of 1995 the ABR in Galim was higher than in Soramboum, on average by a factor of 4.26 (SD ±3.30; range: 0.42–14.67; n = 17). The yearly fluctuations were more pronounced in the Vina du Nord valley and followed a cyclical pattern (Fig 2A). In contrast, the ABR at the Vina du Sud fluctuated only mildly apart from intermittent dips, which reached previous levels in the following year (Fig 2B). An ongoing trend of lower biting frequencies was evident in Soramboum since 2002 (mean: 19,700 flies per person and year vs. 35,348 before) and in Galim since 2006 (mean: 39,628 flies per person and year vs. 103,564 before). In Soramboum the decline in biting rate occurred mainly in the dry season from October till March with only little changes during the rainy season (Fig 3A), whereas in Galim the highest decline was within the peak of the dry season and the peak of the rainy season from February till August (Fig 3B). In the same period of declining ABRs, the monthly infection rate of all L3-harboring flies of all Onchocerca spp. increased in Soramboum from 2.25% (1987–2001: 95%-CI: 0.45; n = 90) to 3.26% (2002–2012: 95%-CI: 0.53; n = 89), while it remained stable in Galim (1989–2005: 3.34%, 95%-CI: 0.57, n = 100 vs. 2006–2012: 3.19%, 95%-CI: 0.62, n = 65).
A historic summary of the Annual Transmission Potentials over the last 36 years in Soramboum (Fig 4A) and 27 years in Galim (Fig 4B) illustrates the alterations in the ratio of animal-filariae and the human filaria O. volvulus in the vector. In Galim, annual filarial transmission rates remained high till 2006 (mean: 13,525 L3 per person and year, SD ±5334), when it dropped to 32.5% of previous levels (mean: 4395 L3 per person and year, SD ±2348; Fig 3B). In contrast, Soramboum experienced an increase of L3 transmission after the early years of ivermectin mass treatments, from an average ATP of 1045 ±438 L3 per person and year in 1987–88 to 2286 ±1338 in 1993–98, which later returned to former levels, i.e. an ATP of 1242 ±741 in 1999–2012 (Fig 4A), although the pre-ivermectin control ATP from the adjacent Touboro site was much higher (4140 L3 per person and year, Fig 3A).
According to morphological classification the species composition of the L3 population in Soramboum from 2009 till 2012 was 23.9% O. volvulus, 65.9% O. ochengi and 10.2% O. ramachandrini (Fig 5A). In previous years the species composition of O. volvulus—O. ochengi—O. ramachandrini fluctuated from 60.7%–12.3%–27.0% (1987–88) to 22.3%–65.3%–10.2% (1993–99) and 40.5%–50.8%–8.7% (2000–06, Fig 4A). Correspondingly, the species were composed as follows in Galim: 11.3% O. volvulus, 88.7% O. ochengi (1995–96), 19.2%, 80.8% (2000–05) and 17.4%, 82.6% (2006–12, Fig 4B). No O. ramachandrini L3 were not found at all.
At the Vina du Nord site 96 isolated L3 (10.3% of all found) from 52 infected flies (7.2% of all dissected) were subjected to molecular identification, of which 76 (79.2%) PCR products were successfully sequenced. At Galim from the Vina du Sud site, 40 L3 (3.5% of all found) from 22 infected flies (2.8% of all dissected) provided 28 (70%) amplicons of Onchocerca spec. which could be successfully sequenced. The species composition of these L3 from Soramboum was 6.6% O. volvulus, 76.3% O. ochengi and 17.1% O. ramachandrini (Fig 5B), whereas in Galim only O. ochengi was found (Fig 5D). A recently discovered O. ochengi genotype called 'Siisa' [22,23,34] contributed to 8.6% and 10.7%, respectively, of the local O. ochengi L3 population in Soramboum and Galim (Fig 4B and 4D, respectively). In comparison with morphological classification (n = 71), 72.2% (13/18) of so-called O. volvulus in Soramboum were in fact O. ochengi, and 2.4% (1/41) of O. ochengi were O. ramachandrini. All examined O. ramachandrini (n = 12) were correctly identified. Hence, the respective effective ATP in Soramboum for the years 2009 to 2012 was 68, 221, 58 and 47 for O. volvulus (mean: 98): 773, 2503, 1388 and 475 for O. ochengi (mean: 1285) and 18, 392, 238 and 70 for O. ramachandrini (mean: 180). Accordingly, the adjusted species proportion of the L3 population for these years were on average 6.3% O. volvulus, 82.2% O. ochengi and 11.5% O. ramachandrini (Fig 4A, right side).
At the Vina du Sud site near Galim (n = 67) O. volvulus (0/46) and O. ramachandrini (0/0) have not been detected since the introduction of molecular methods for L3 species identification (Fig 5D), although there were morphologically identified specimens of O. volvulus (n = 149, 14.6% of total, Fig 5C). All morphologically classed O. ochengi (n = 21) were correctly identified. Hence, the whole L3 population in the observation period 2009–2012 consisted of O. ochengi (Fig 4B, right side) with an ATP of 5096, 4525, 6753 and 2046 (mean: 4605).
In order to evaluate the reliability of body length as a characteristic trait that can be used for species discrimination of infective larvae, the body lengths of occurring Onchocerca spp. L3 in S. damnosum s.l. were compared with morphological and molecular identification methods (Fig 6). Whereas the inter-specific differences according to morphological criteria are significant (p < 0.001), no within-species length difference has been detected between morphological and molecular identification of O. volvulus and O. ramachandrini. A significant (p < 0.01) within-species difference has been found in the common genotype of O. ochengi sensu stricto, but with a low effect size (d = 0.598, n = 95); a significant difference (p < 0.001, d = 1.471, n = 19) was also found for the genotype O. ochengi 'Siisa'. Interestingly, a more than 4-times higher proportion of morphologically misidentified O. volvulus were O. ochengi 'Siisa' (25.4%; 15/59) than in the morphologically identified O. ochengi group (6.3%; 4/64). For the genotype O. ochengi s.s., this was vice versa (74.6%; 44/59 of misidentified O. volvulus vs. 92.2%; 59/64 of O. ochengi group).
This study represents a comprehensive 4-year dataset of transmission from sites in two onchocerciasis-endemic river basins and re-evaluates data collected up to 36 years ago. Whereas we can observe a break of the transmission cycle on the Adamawa plateau, the decline of parasite transmission seems to be halted in the focus of the Sudan savannah despite ongoing treatment intervention. The treatment intervention in this focus has passed the estimated life expectancy of the worm (10 to 15 years) almost by factor two. Conjected that the residual transmission does not stem from invading infected flies from other endemic foci by wind drift, the elimination of O. volvulus in previous hyperendemic foci in North Cameroon by yearly-given CDTI appears to be difficult, even though the actual risk of skin-lesions and blindness due to onchocerciasis is presumably very low.
Despite the fact that the Mectizan treatment campaigns on the mesoendemic Adamawa plateau have started about 10 years later, the transmission of the parasites there seems to be disrupted. One reason might be that the transmission rates in the past were partly overestimated due to misidentification of infective larval stages derived from animals. However, several studies including this one emphasize the adequate discriminative power of morphological characters for species delimitation, in particular the body length (Fig 6) and shape of head and tail [15,19,35]. It is thus very likely that at least a fraction of the L3 found were correctly identified as O. volvulus. However, if the endemicity of a region reaches hypoendemic levels in an area of intensive transmission of filariae of non-human origin, such as in this case, the reliability of microscopic examination is limited. For confirmation of interruptions of O. volvulus transmission, molecular methods like PCR are necessary.
Since 2006 there is a steady trend of lower ABRs, in particular at the Vina du Sud river, where biting rates before were with only one exception above 60,000 per man and year (Fig 2B). The vector transmission of filarial stages have also dropped during this time (Fig 4B), but to a lesser extent in the Sudan savannah (Fig 4A) due to a concomitant gain of the infection rate by bovine filariae. The reason for this vector decline could be the result of decreased availability of breeding sites or food for the aquatic Simulium larvae, and hence a drop in population size. A distinct increase of endoparasitic mermithids in human-biting nulliparous flies was evident at the Vina du Sud breeding sites (S1 table). It is, however, unlikely that these mermithids or other Simulium parasites are the main drivers for the massive decline in biting rates of recent years. A reduced longevity of adult flies was not observed, when comparing the current parous rate with those of flies at baseline [24]. Furthermore, a continuous rise in the pool of potential blood hosts, both human and livestock, may also contribute to lower individual biting frequencies. The regional impact of climate change cannot be excluded, either, although the water delivery of the investigated rivers have not changed drastically until 1980 [26].
The longitudinal monitoring in the Vina du Nord valley indicates that the average transmission of O. volvulus remained around 500 L3 per man and year for 20 years after the onset of ivermectin mass treatments (Fig 4A). This seems to contradict the reduction of onchocerciasis-positive patients in the region as a result of control strategies with ivermectin [12]. One reason could be that there is a variable degree of misidentification of O. volvulus L3. In the most recent monitored years 2009–2012, when molecular detection methods were used, the degree of morphological misidentification was 72%. However, earlier epidemiological data from the Sudan savannah [25] showed ATP above 4000 L3 per man and year (Fig 4A, left side). Even though no differentiation of the species had been undertaken at that time, the proportion of animal-filariae in S. damnosum s.l. were likely low due to the lack of cattle as potential blood hosts [for explanation see 36]. Hence, only filariae from the warthog could have been co-transmitted, and the infection rate of the vector with O. ramachandrini has not changed considerably during the observation period. Another theory states that the regulation of parasite transmission may be density-dependent instead of linear. That means the effective reproductive ratio of filarial worms equals one even though the basic reproductive ratio is much higher. In the Vina du Nord river basin, Renz [25] compared the prevalence of onchocerciasis and burden of microfilariae with the L3 infection rate in the vector and found no linear correlation, but rather a dependency of fly infection rate with prevalence in the human population instead of the community's microfilarial load (mff/mg). A density-dependent mechanism has already been shown for the parasite acquisition in cattle when inoculated with infective larvae of O. ochengi [37,38].
The observed seasonal variations of the entomological parameters match well with baseline data from the Vina du Nord river [25], including the number of infective flies with L3. Nonetheless, the O. volvulus ATP has drastically reduced to 3.5% of the baseline value meaning that the majority of infective flies now carry filariae of animal origin. Additionally, the number of L3 per infective fly decreased moderately (from 3.2 to 1.8). The low but stable transmission level of O. volvulus could mean that the threshold for maintaining endemicity is perhaps lower than current mathematical models predict (ATP ≥ 100 in West Africa [39,40], but also ATP ≥ 54 in Central America [41]).
Even though molecular techniques of identification give higher accuracy, they are less suitable for high throughput analysis due to limitations of time, cost-effectiveness and local infrastructure. They are nonetheless very useful for the detection of unknown strains and species of filarial nematodes in vector and host, such as Onchocerca ochengi 'Siisa' [22,23,34]. Experimental infection studies from Togo [42] and Mali [43], where O. ochengi microfilariae were inoculated by the vector from infected cattle, revealed shorter L3 body lengths (Togo: 680 μm, 540–680; Mali: 647 μm, 540–810) than our observations (740 μm, 600–940) and previous studies from Cameroon [44]. These values, however, lie in proximity to the measurements for O. ochengi 'Siisa' (660 μm, 600–900) and may thus represent or morphologically resemble this strain. Ultimately the evolutionary relationships of Onchocerca parasites in humans, cattle and game animals can be compared and tested with genetic markers by generating phylogenetic trees [22,34].
Besides the climatic disparities of the two foci, which is mainly a result of different altitudes, they share similar conditions for their respective black fly populations. One major difference, however, is the disproportionately higher cattle stock density on the Adamawa plateau compared to the situation in the Sudan savannah (Fig 1). The cattle to human ratio around the Galim focus is approx. 10:1, whereas in the Soramboum focus it ranges between 1:1 and 2:1, and was even lower in previous years, because nomadic cattle were not allowed to enter the Vina du Nord basin until 1975, and the local villagers had not kept any livestock animals, either. Nowadays, an increasing number of vagrant Bororo herdsmen arrive with their herds of zebu cattle and become settled. The inherent difference in livestock density is both culturally inherited (migrating pastoralists of the North vs. settled cattle farmers of the South) and due to biologic conditions (water and food scarcity during the dry season in the Sudan savannah; absence of tsetse flies on the Adamawa plateau, which transmit bovine trypanosomiasis). Invading O. ochengi L3 elicit a humoral immune reaction in humans, which cross-reacts with O. volvulus L3 antigens, thereby reducing transmission success [19]. The protective effect of populations under repeated antigen exposure is called premunition and well known for malaria and other infectious diseases [45,46]. On the Adamawa plateau this effect seems to be strong enough to prevent or at least complicate the regional endemicity of O. volvulus. The advent of nomadic herdsmen and their cattle herds in the Vina du Nord valley is congruent with an increased transmission of animal-borne filariae, in particular O. ochengi (Fig 4A). This sudden jump of animal-filariae in the vector population implies the diversion of large quantities of local S. damnosum s.l. to take their blood meal from cattle, thereby reducing the vector pool for humans [16]. This phenomenon has been termed zooprophylaxis and acts also as a protective trait against onchocerciasis transmission [15,20].
The important question is how the low but stable rate of onchocerciasis transmission in the Sudan savannah can be further curbed or completely stopped. Altogether, five molecularly identified O. volvulus L3 from two infective flies (3.85% of total, 95% CI: 0.47–13.21%) were found in the dry season of 2010 (February) and the rainy season of 2011 (June). Since yearly CDTI application rounds are given at the end of July, the late time point of occurrence after ivermectin treatment may hint to an incipient reconstitution of skin microfilariae in humans infected with O. volvulus 12 months prior. This would be a strong argument for the continuation or even temporary intensification of the ivermectin control program [10,36,47]. However, current political instabilities in adjacent countries and the exclusion of certain patient groups in treatment intervention programs, like nomadic people, illegal immigrants and refugees, could impede the long-term success of such measures. Ongoing monitoring of vector transmission is therefore crucial for health policy in onchocerciasis-endemic countries.
|
10.1371/journal.pgen.1003719 | Histone Chaperone NAP1 Mediates Sister Chromatid Resolution by Counteracting Protein Phosphatase 2A | Chromosome duplication and transmission into daughter cells requires the precisely orchestrated binding and release of cohesin. We found that the Drosophila histone chaperone NAP1 is required for cohesin release and sister chromatid resolution during mitosis. Genome-wide surveys revealed that NAP1 and cohesin co-localize at multiple genomic loci. Proteomic and biochemical analysis established that NAP1 associates with the full cohesin complex, but it also forms a separate complex with the cohesin subunit stromalin (SA). NAP1 binding to cohesin is cell-cycle regulated and increases during G2/M phase. This causes the dissociation of protein phosphatase 2A (PP2A) from cohesin, increased phosphorylation of SA and cohesin removal in early mitosis. PP2A depletion led to a loss of centromeric cohesion. The distinct mitotic phenotypes caused by the loss of either PP2A or NAP1, were both rescued by their concomitant depletion. We conclude that the balanced antagonism between NAP1 and PP2A controls cohesin dissociation during mitosis.
| Eukaryotic DNA is assembled into a nucleo-protein structure called chromatin. Nucleosomes are the basic building blocks of chromatin, comprising 147 bp of DNA tightly wrapped around a histone protein core. Histone chaperones mediate nucleosome assembly by preventing non-productive aggregation of histones with DNA. Here, we describe an unexpected function for the canonical histone chaperone NAP1 in sister chromatid resolution. The precisely orchestrated binding and release of cohesin is crucial for proper chromosome segregation in mitosis. Cohesin holds newly replicated sister chromatids together till early mitosis. Then, it is first removed from the chromosome arms and lastly from the centromeres. This process ensures proper chromosome resolution and segregation into daughter cells. Cohesin removal from the arms is initiated by mitotic kinases in early mitosis, but can be counteracted by protein phosphatase 2A (PP2A). We found that NAP1 blocks protein phosphatase 2A binding to chromosomal cohesin, thereby allowing cohesin phosphorylation and dissociation from the chromosome arms. Functional in vivo experiments established that the antagonistic activities of NAP1 and PP2A control the chromosomal cohesin cycle and sister chromatid resolution. These results provide a novel chromatin-assembly-independent mitotic function for a histone chaperone.
| Histone chaperones perform crucial functions during the duplication of eukaryotic genomes [1]–[2]. They guide the posttranslational processing and trafficking of newly-synthesized histones to replication forks and mediate replication-coupled chromatin assembly [2]–[8]. Histone chaperones CAF1, ASF1 and HIRA bind histone H3/H4 tetramers, whereas NAP1 binds both H3/H4 tetramers and H2A/H2B dimers. Although originally identified as factors that prevent aggregation and direct the assembly of histones on DNA [9], it turned out that histone chaperones play a variety of regulatory roles in chromosome biology. In addition to replication-coupled chromatin assembly, histone chaperones function in gene-specific transcription control, DNA repair and direct specific histone modifications [10]–[15].
Histone chaperones achieve these diverse functions through cooperation with other factors, such as histone modifying enzymes and ATP-dependent chromatin remodelers [15]–[20]. For example, ASF1 and NAP1 cooperates with histone modifying factors to differentially modulate local chromatin during NOTCH signaling [15], [21]. NAP1 associates with RLAF (RPD3 and LID associated factors), an assemblage of the histone deacetylase RPD3, histone H3 lysine 4 demethylase LID/KDM5, SIN3A, PF1, EMSY and MRG15. NAP1 recruits RLAF to the (E)Spl NOTCH-regulated genes to generate a repressive chromatin structure and mediate transcriptional silencing [15].
A specific function for histone chaperones during mitosis has not been established. Suggestively, we noted the potential association between NAP1 and cohesin in a proteomic survey of histone chaperones [15]. Cohesin is the conserved protein complex that mediates cohesion between sister chromatids after replication, which is crucial for proper chromosome segregation in mitosis and meiosis. The core of cohesin is formed by Stromalin (SA/SCC3), and a tripartite ring comprising SMC1, SMC3 and RAD21/SCC1. The cohesin ring embraces and holds sister chromatids together [22]–[23]. For a comprehensive discussion of mitotic cohesin dynamics we refer to a number of excellent reviews [24]–[31].
Briefly, cohesin binds chromosomes prior to DNA replication, enabling the linkage of newly replicated sister chromatids from S- through G2 phase. By metaphase, juxtaposed chromatids are only connected at their centromeric regions and have separate chromosome arms. This process is referred to as sister chromatid resolution and requires cohesin release from the arms, but not from the centromeres. During prophase, Polo-like kinase and potentially other mitotic kinases, phosphorylate SA, which triggers the bulk dissociation of cohesin from the chromosome arms [32]–[34]. This step also requires the cohesin releasing complex WAPL-PDS5 that interacts transiently with cohesin at mitotic entry [35]–[38]. All this time, centromeric cohesin remains associated and is protected from phosphorylation by the Shugoshin(Sgo)/MeiS332 family of proteins, which act in conjunction with PP2A [39]–[45]. At anaphase, separase-mediated cleavage of RAD21 causes the dissociation of centromeric cohesin, allowing sister chromatid segregation [46]–[47].
Motivated by the potential interaction between NAP1 and cohesin [15], we wondered if NAP1 might function in the cohesin chromosome binding and release cycle. We established that NAP1 and cohesin interact functionally. Through counteracting PP2A access to cohesin in early mitosis, NAP1 is a crucial regulator of the chromosomal cohesin cycle. Loss of NAP1 severely compromised cohesin release from the chromosome arms and sister chromatid resolution. These results uncover a mitotic function for NAP1 that is separate from its role in nucleosome assembly.
To examine the role of NAP1, we analyzed mitotic chromosomes prepared from colchicine-treated S2 cells after RNAi-mediated depletion of NAP1. Loss of NAP1, but not loss of the histone chaperone CAF1, caused a striking increase in the number of unresolved sister chromatids (Figure 1A and Figure S1A). In addition, depletion of NAP1 caused reduced cell proliferation, mitotic defects, an accumulation of poly/aneuploid cells and an increased portion of cells in the G1-phase of the cell cycle (Figure S1B–D). We also analyzed mitotic chromosomes from Drosophila larvae homozygous for the NAP1 knockout allele nap1KO1 [48]. Nap1KO1 homozygous flies are semi-lethal, and male escapers are sterile. Immunoblotting confirmed that NAP1 was not detectable in larvae homozygous for nap1KO1, whereas cohesin levels were unaffected (Figure S1A). We prepared mitotic chromosomes from colchicine-treated larval brain cells. In cells lacking NAP1 we observed a dramatic increase in cohesively linked sister chromatids, compared to wild type cells (Figure 1B). Quantification of S2 cells and larval brain cells with either resolved or unresolved sister chromatids confirmed the crucial role of NAP1 in this process (Figure 1C). However, we observed no changes in mitotic chromosome morphology, immunostaining efficiency of histone H3, micrococcal nuclease (MNase) sensitivity or nucleosome spacing upon depletion of NAP1 (Figure 1D and Figure S1E). This suggests that there are no gross changes in chromatin organization due to loss of NAP1. We conclude that NAP1 is required for sister chromatid resolution and normal mitosis.
Resolution of sister chromatids is initiated at prophase by the bulk removal of cohesin from mitotic chromosome arms [24]–[26]. Therefore, we examined chromosomal cohesin binding after knockdown of NAP1. We used antibodies against SA and RAD21 to visualize cohesin on mitotic chromosomes that were isolated from colchicine-treated S2 cells. Depletion of NAP1 caused a striking accumulation of cohesin on the mitotic chromosome arms (Figure 2A–B). In mock-treated cells, we could only detect cohesin binding to the centromeric regions of mitotic chromosomes. Likewise, chromosomes from homozygous nap1KO1 larval brain cells were densely coated with cohesin, whereas on wild type chromosomes cohesin binding was limited to the centromers (Figure 2C–D). Thus, NAP1 is required for cohesin release during mitosis.
Interestingly, NAP1's sub-cellular localization is dynamic and changes during the cell cycle (Figure S2). During interphase NAP1 is distributed about equally between cytoplasm and nucleus in Drosophila embryos, but at prophase there is a strong increase in nuclear NAP1. By metaphase, NAP1 has dissociated from chromosomes, along with most SA, but by anaphase both re-associate. Thus, nuclear accumulation of NAP1 at prophase agrees well with its function in promoting cohesin release in early mitosis. Together, these results suggest that NAP1 promotes sister chromatid resolution by mediating cohesin release from the chromosome arms.
We wondered if NAP1 interacts with- and co-localizes with cohesin on chromatin. Immunostaining of interphase 3rd instar larval salivary gland polytene chromosomes with antibodies against NAP1 and SA revealed a substantial overlap in their genomic binding loci (Figure 3A). For a high resolution analysis, we performed chromatin immunoprecipitations (ChIPs) in asynchronously dividing S2 cells using antibodies against NAP1, SA and SMC1. Following ChIP, isolated DNA fragments were mapped back to the genome by hybridization to Drosophila tiling arrays (Figure 3B and Figure S3A–C). All ChIPs were performed using 2 independent biological replicates, which showed a high degree of correlation (Figure S3D). The averaged genomic binding profiles of NAP1 and cohesin were highly correlated (r>0.7), indicating significant co-occupancy (Figure S3E).
We selected ChIP-chip peaks at a false discovery rate (FDR)<0.01, based on random permutation, yielding roughly 10,000 assigned binding loci for each factor. Intersection of genomic binding sites revealed a substantial overlap between cohesin and NAP1 loci (Figure 3C). Shared target loci include the NOTCH-regulated cut and E(spl) genes, and the ecdysone-controlled BrC and Eip75B genes (Figure 3B and Figure S3A–C). Binding of NAP1 and SA to these loci was confirmed independently by ChIP followed by quantitative PCR (Figure 3D–E). Previously, it was found that NAP1 and cohesin are required for the repression of NOTCH-target genes [15], [49], indicating that these factors might also cooperate in gene regulation.
To establish whether NAP1 interacts with cohesin, we immunopurified NAP1, SA and SMC1 from Drosophila embryo nuclear extracts (NE). Following extensive washes with a buffer containing 600 mM KCl and 0.1% NP40, immuno-purified proteins were resolved by sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) and visualized by Coomassie staining (Figure 4A–C). Protein identities were determined by mass spectrometry. Table S1 provides an overview of the proteins identified. In addition, we included the results of NAP1 purified from embryo NE using an independent antibody and NAP1 purified from S2 cells. All three independent NAP isolations yielded similar results. NAP1-associated proteins, such as the RLAF silencing complex [15], are mostly involved in transcription control. Importantly, the full cohesin complex was associated with NAP1 in all three independent purifications. Conversely, NAP1, but not RLAF, was identified alongside the cohesin subunits in both SA and SMC1 purifications. The majority of cohesin-bound proteins we identified have been implicated in cohesin biology, including the loading factors Nipped-B/SCC2 and MAU2/SCC4 [50]–[52]. Both NAP1 and cohesin purifications contained replication factor C, which is involved in DNA replication and cohesin loading [53]–[54]. We also noted the presence of protein phosphatase PP2A in the NAP1, SA and SMC1 purifications (Table S1). Immunopurification of PP2A from embryo NE followed by mass spectrometry revealed that PP2A is part of an extensive network of kinases and phosphatases (Figure 4D and Table S2). Most relevant for the present study, NAP1, the full cohesin complex, and cohesin loading factors Nipped B and Mau2 were all present in the PP2A purification.
We confirmed the specific association between NAP1, cohesin and PP2A by a series of co-immunoprecipitations (co-IPs) followed by immunoblotting. NAP1 was prominent in SA IPs, together with other cohesin subunits (Figure 4E). Conversely, the cohesin subunits SA, SMC1 and RAD21 were readily detected in NAP1 IPs (Figure 4F). Likewise, NAP1 and cohesin subunits were also detected in PP2A IPs under stringent conditions (600 mM KCl, 0.1% NP40). In addition, PP2A co-IPs under milder conditions (200 mM KCl, 0.1% NP40) with MeiS332, a Drosophila homolog of Sgo. Thus, the association between PP2A and Sgo/MeiS332 seems to be conserved from mammals to flies (Figure 4G). Other histone chaperones such as CAF1 or ASF1 did not bind cohesin, supporting the selectivity of NAP1 binding.
Protein-protein interaction assays using recombinant proteins confirmed the ability of NAP1 to bind cohesin subunits (Figure S4). NAP1 did not bind to the recombinant catalytic or regulatory subunits of PP2A, suggesting that they do not interact directly (Figure S4). In cells, the association between PP2A and NAP1 might be mediated through bridging by cohesin. Collectively, these results established a biochemical interaction between the histone chaperone NAP1 and cohesin.
To characterize the NAP1, cohesin and PP2A interaction further, we used Sephacryl S-300 size-exclusion chromatography (Figure 4H). NAP1 and cohesin subunits co-eluted in column fractions corresponding to an apparent molecular mass of ∼1.5 MDa. A substantial portion of NAP1 and SA, but not the other cohesin subunits, were also present in fractions corresponding to an average molecular mass of ∼300 kDa. We did not detect appreciable amounts of NAP1 or SA eluting at their predicted molecular weights, suggesting they are not present as free proteins. Co-IPs from pooled high molecular weight S-300 column fractions 12–14 confirmed the presence of a large assemblage harboring the full cohesin complex, NAP1 and PP2A (Figure 4I). In addition, co-IPs from the pooled lower molecular weight fractions 27–29 revealed a separate complex comprising NAP1 and SA, but devoid of PP2A or the other cohesin subunits (Figure 4J). A possible interpretation of these results is that NAP1 forms a separate module with SA, which interacts dynamically with the full cohesin complex.
The removal of cohesin from chromosome arms is triggered by SA phosphorylation in prophase, whereas centromeric cohesin is protected from phosphorylation by a complex of PP2A and Sgo/MeiS332 [32]–[34], [39]–[40]. The results from our interaction studies (Figure 4) suggest that NAP1, or the NAP1-SA module, may compete with PP2A for cohesin binding. Therefore, we considered the possibility that NAP1 promotes cohesin dissociation from chromosome arms by blocking PP2A binding to cohesin. To test this idea, we performed competition assays using recombinant NAP1 expressed in baculovirus (Figure 5A) and cohesin complex immunopurified from embryo nuclear extracts with antibodies against SA or SMC1 (Figure 4B–C). Recombinant NAP1 binds immunopurified cohesin and drives the dissociation of endogenously bound PP2A (Figure 5B–C). To test the effect of NAP1 on PP2A binding to cohesin in vivo, we immunopurified cohesin complex from S2 cells that were either mock treated or depleted for NAP1. Loss of NAP1 caused a strong increase in PP2A association with cohesin (Figure 5D). Accompanying the increase in cohesin-bound PP2A, immunoblotting with antibodies directed against phospho-Serine or SA revealed decreased levels of phosphorylated SA following NAP1 knockdown. Depletion of NAP1 affected neither the integrity, nor the stoichiometry of the cohesin complex (Figure 5E and Figure S5A). Likewise, cellular levels of PP2A were not affected by NAP1 depletion (Figure S5B).
Next, we investigated if NAP1 counteracts PP2A binding to cohesin during mitosis. Immunopurification of SA and SMC1 revealed increased association of NAP1 with cohesin in colchicine treated cells that are arrested in mitosis, compared to untreated cells (Figure 5F–G and Figure S6). Whereas NAP1 associates, PP2A dissociates from cohesin in mitosis allowing persistence of the phosphorylation of SA by mitotic kinases. After NAP1 knockdown, PP2A stayed bound to cohesin in mitotic cells and SA remained dephosphorylated suggesting that NAP1 drives PP2A dissociation (Figure 5G and Figure S6). To ensure the phosphorylation effects we observe are SA specific, we immunopurified SA from S2 cells under denaturing conditions. Immunoblotting with antibodies against phospho-Serine confirmed that we detect phosphorylated SA and not an associated factor such as SMC1, which dissociates from SA under these conditions (Figure 5H). Finally, the effect of colchicine treatment or NAP1 knockdown show that SA phosphorylation is cell-cycle regulated and depends on NAP1 (Figure 5I). We conclude that NAP1 counteracts PP2A binding to cohesin, thereby preventing SA dephosphorylation during G2/M transition, resulting in a net increase in phosphorylated SA.
To examine the role of NAP1 in the binding of PP2A to chromosomal cohesin loci, we performed ChIPs (Figure 6A). Knockdown of NAP1 resulted in a striking increase in PP2A association with the cohesin and NAP1 binding sites examined. In contrast, SA depletion caused a loss of PP2A binding to the genomic NAP1 and cohesin sites, suggesting that SA tethers PP2A to chromatin. Confirming the specificity of the assay, ChIP signals were strongly reduced after PP2A knockdown. Based on these ChIP results, we conclude that PP2A binding to chromosomal cohesin is attenuated by NAP1. This notion is supported by immunostaining of mitotic chromosomes. Knockdown of NAP1 caused a strong accumulation of PP2A onto the chromosome arms in ∼80% of cells (Figure 6B). Upon loss of NAP1, MeiS332 is no longer restricted to the centromers, but now coats the chromosome arms (Figure 6C). Thus, MeiS332 behaves similar to its centromeric partner PP2A. This result provides additional support for the notion that the role of PP2A-Sgo/MeiS332 in mitosis is conserved from flies to mammals. Collectively, our findings suggest that NAP1 regulates sister chromatid resolution by preventing PP2A binding to cohesin on chromosome arms. Blockage of PP2A allows phosphorylation of SA by mitotic kinases, which drives cohesin release and sister chromatid resolution.
If NAP1 mediates sister chromatid resolution by counteracting PP2A, concomitant loss of PP2A should reverse the effects of NAP1 depletion. To test this idea, we analyzed mitotic chromosomes after knockdown of either NAP1, PP2A or both factors (Figure 7A and Figure S7A). First, we observed that the effect of PP2A knockdown is the opposite of NAP1 depletion. Instead of unresolved sister chromatids, which are the hallmarks of NAP1 depletion, knockdown of PP2A caused diminished centromeric cohesion in ∼65% of mitotic cells (Figure 7A–B). This was accompanied by dissociation of centromeric SA and RAD21, but not MeiS332 in ∼60–65% of mitotic cells (Figure 7B–D). However, after concomitant depletion of NAP1 and PP2A the majority of mitotic chromosomes appeared normal. We observed no loss of centromeric cohesion and the majority of sister chromatids were resolved properly. In agreement with the rescued phenotype, cohesin and MeiS322 localization was largely normal in ∼70–80% of cells depleted for both NAP1 and PP2A.
Our earlier biochemical results indicated that NAP1 counteracts PP2A binding to SA, thereby preventing SA dephosphorylation. Therefore, we tested if the loss of phosphorylated SA after NAP1 knockdown was dependent on PP2A. Indeed, concomitant knockdown of PP2A and NAP1 restored the levels of phosphorylated SA, Whereas depletion of PP2A alone did not affect SA phosphorylation (Figure 7E). These results suggest that NAP1 and PP2A act antagonistically in the control of the cohesion cycle.
To complement the experiments in which we depleted endogenous proteins, we compared the effects of ectopic expression of NAP1, PP2A or the PP2AH59Q catalytic mutant [55] in S2 cells (Figure 7F and Figure S7B–D). Over-expression of NAP1 resulted in a mild increase of defective centromeric cohesion and a loss of cohesin binding to the centromers. Ectopic expression of PP2A gave the opposite phenotype, and mimicked the effect of NAP1 depletion, namely, cohesive linkage of the majority of mitotic chromosome arms (∼80%). Highlighting the importance of PP2A's catalytic activity, expression of PP2AH59Q did not lead to failed sister chromatid resolution, but a mild loss of centromeric cohesion. Thus, ectopic expression of a phosphatase-defective PP2A mutant yielded a similar phenotype as depletion of endogenous PP2A. When over-expressed together, NAP1 and PP2A cancelled each other out, resulting in wild type mitosis in most cells. The results of these ectopic expression assays are fully consistent with those of our depletion experiments. Collectively, they demonstrate the antagonistic function of NAP1 and PP2A in regulation of SA phosphorylation, cohesin release and sister chromatid resolution.
As reflected by their name, a major activity of histone chaperones is to prevent illicit liaisons and guide newly synthesized histones to sites of chromatin assembly. Here, we described a mitotic function for the canonical histone chaperone NAP1 that is unrelated to nucleosome assembly. We found that NAP1 binds cohesin and blocks dephosphorylation of SA by PP2A, thereby promoting cohesin dissociation from the chromosome arms. Consequently, chromosomal binding of cohesin during mitosis is controlled by the balance between the opposing activities of NAP1 and PP2A.
NAP1 is part of a large assemblage including the full cohesin complex and PP2A. In addition, NAP1 and SA form a subcomplex, which lacks the other cohesin subunits and PP2A. An attractive scenario is that the NAP1-SA module or NAP1 alone competes with PP2A-bound SA within the full cohesion complex. PP2A displacement by NAP1 allows stable phosphorylation of cohesin and its dissociation during early mitosis. NAP1 might also act as a direct inhibitor of PP2A catalytic activity, because a mammalian NAP1 homolog, SET, has been identified as a potent PP2A inhibitor [56], which promotes sister chromatid segregation during mouse oocyte miosis [57]–[58]. In addition, NAP1 might help cohesin phosphorylation by tethering Polo kinase to cohesin. In fact, we detected a potential association between NAP1 and Polo kinase (Figure S4). However, the dramatic chromosome condensation defects after Polo kinase depletion precluded a functional evaluation of a possible role of NAP1 in its function. Nevertheless, although we cannot exclude additional NAP1 activities, our functional experiments established that blockage of PP2A suffices to explain the crucial role of NAP1 during sister chromatid resolution.
NAP1 not only regulates the chromosomal distribution of cohesin and PP2A, but also that of MeiS332, a fly homolog of Sgo. The function of MeiS332 and PP2A appears to be largely conserved from mammals to flies because they bind each other and depletion of either factor causes a loss of centromeric cohesion [39]–[45]. Either knockdown of NAP1 or over-expression of PP2A (Figure S7E) caused spreading of MeiS332 onto the arms of mitotic chromosomes, accompanying the loss of sister chromatid resolution. Thus, the balanced antagonism between NAP1 and PP2A controls chromosomal association of both cohesin and MeiS332 during mitosis.
One level of regulation involves changes in NAP1's subcellular localization and chromatin binding through the cell cycle. At prophase there is a strong increase in the level of nuclear NAP1, but by metaphase, NAP1 and cohesin have dissociated from the chromosomes (Figure S2). Thus, the dynamic behavior of NAP1 correlates well with its function in promoting cohesin release at early mitosis. Regulation of NAP1 localization may involve cyclin B-cdc2/cdk1 kinase complexes. Previously it was found that yeast and vertebrate NAP1 are phosphorylated by cyclin B-cdc2 [59] and that yeast cyclin B requires NAP1 for its full range of mitotic functions [59].
We suggest that histone chaperones are at the hubs of specialized protein networks that perform a wide variety of tasks in chromosome biology. Through association with distinct partners, NAP1 is able to perform different functions. By acting as a histone chaperone, NAP1 mediates chromatin assembly [60]–[61]. Through recruitment of the histone H3 deacetylase and H3K4 demethylase complex RLAF, NAP1 controls gene-selective silencing at developmental loci [15]. Finally, by binding cohesin and blocking SA dephosphorylation by PP2A, NAP1 mediates sister chromatid resolution during mitosis. These results emphasize the surprisingly diverse- and specific regulatory functions of histone chaperones in chromosome biology.
Anti-SA antibodies were raised in guinea pigs and rabbits against a GST-fusion protein encoding SA amino acids 12–312, purified from E. Coli. Rabbit polyclonal antibodies against the NAP1 were previously described [15], anti-SMC1 and anti-RAD21 antibodies were a gift from D. Dorsett [49], anti-phospho-Serine (Abcam, ab6639), anti PP2A-C (BD Biosciences, 610555), anti-CID (Abcam, ab10887) and anti-α-Tubulin (Sigma, T5168).
Drosophila S2 cells were cultured in Schneider's media (Invitrogen 21720-024) supplied with 10% FBS. Double-stranded RNAs for NAP1, SA and PP2A/MTS were synthesized using an Ambion Megascript T7 kit according to the manufacturer's protocol with the following primers: 5′-TATTGAACAATGGACGCCC-3′ and 5′-TGAAACTCCAAGGTGTACG-3′ for NAP1; 5′- CAGTCAAATACATAAAATGATGGCG-3′ and 5′-GCTCAATCCATTGGTCAACA-3′ for SA; and 5′-GGCAGTCTTTCCCTTCGTATATC-3′ and 5′-CGAACTTGTGTCTCTGTCAACTG-3′ for the PP2A catalytic subunit encoded by the mts gene. Primers were flanked by T7-promoter sequence (5′-TTAATACGACTCACTATAGGGAGA-3′) at the 5′-end. For mock knockdowns, dsRNA against GFP was synthesized with the following primers (5′- CAAGAGTGCCATGCCCGAAGGT-3′ and 5′-TGTGGTCACGCTTTTCGTTGGG-3′) flanked by the T7 promoter sequence. For cytological or FACS analysis, S2 cells were incubated with dsRNA for 2 days as described [62]. Open reading frames (ORFs) of GFP, NAP1, PP2A and catalytically inactive PP2A - PP2AH59Q [55] were cloned into pENTR/DTOPO entry vector (K2400-20, Invitrogen) and subcloned into Drosophila expression vector pAHW (obtained from the Drosophila Genome Resource Center, DGRC) carrying the Actin 5C promoter and a sequence encoding the HA-tag by LR-clonase reaction. Polyethylenimine (PEI) ∼25000 Da (408727 Sigma-Aldrich) was used for transient transfection of S2 cells as described [63]. Protein expression and mitotic chromosomes were analyzed 48 hours post-transfection.
For indirect immunofluorescence analysis, S2 cells were fixed in phosphate buffered saline (PBS) buffer containing 3.7% formaldehyde for 5 min. Cells were incubated with antibodies against CID and α-Tubulin diluted at 1∶200 and 1∶1000, respectively. After washes with PBS, cells were incubated with fluorescently-labeled secondary antibodies (Molecular Probes) and analyzed with the Leica FW4000 imaging system. For mitotic chromosomes preparation from S2 cells and Drosophila larvae brain cells, cells were treated with 10 mM colhicine for 10 min, incubated with 0.5% sodium citrate for 5 min (10 min for brain cells) and fixed in 45% acetic acid containing 3.7% formaldehyde for 5 min. Next, cells were squashed on objective slides and covered by a cover-slip. Slides were washed with PBS, and incubated with anti-SA, anti-RAD21 and anti-MeiS332 antibodies, each diluted 1∶200, and anti-PP2A diluted 1∶20. After washes and incubation with fluorescent-labeled secondary antibodies, mitotic chromosomes were analyzed on a Leica FW4000 imaging system. To quantify the accumulation or loss of cohesin, PP2A or MeiS332 on mitotic chromosomes, chromosomes from at least 30 cells were analyzed. Fluorescent assisted cell sorting (FACS) of S2 cells was performed as described [64].
For chromatin immunoprecipitation experiments (ChIPs), S2 cells were fixed with 1% formaldehyde for 10 min. Fixation was stopped by addition of 125 mM glycine and cells were lysed in ice-cold L buffer (1% sodium dodecyl sulfate, 10 mM EDTA, 50 mM Tris-HCl pH 8.1, 0.5 mM phenylmethylsulfonyl fluoride - PMSF, and 100 ng/ml of leupeptin and aprotinin). Cross-linked chromatin was fragmented by sonication to an apparent length of ∼500 bp. Cross-linked chromatin (100 µg) was diluted with 9 volumes of buffer D (150 mM NaCl, 20 mM Tris-HCl pH 8.1, 2 mM EDTA pH 8.0, 1% Triton-X100, 0.5 mM PMSF, and 100 ng/ml of leupeptin and aprotinin) and pre-cleared with 10 µl of protein A agarose (16–157, Upstate). Pre-cleared chromatin was incubated at 4°C with appropriate antibodies overnight and precipitated with 20 µl of protein A agarose. For Mock ChIPs, chromatin was incubated with preimmune serum. Protein A agarose was washed extensively with buffer W (20 mM Tris-HCl pH 8.1, 2 mM EDTA pH 8.0, 0.1% SDS, 1% Triton X-100, 0.5 mM PMSF, and 100 ng/ml of leupeptin and aprotinin) containing 150 mM NaCl, and 1× with buffer W/500 mM NaCl. DNA retaining on the protein A agarose was eluted by incubating with 250 µl buffer E (1% SDS, 0.1 M NaHCO3, 500 µg/ml Proteinase K) for 2 hrs at 37°C and overnight at 65°C and extracted with QIAquick PCR purification kit (Qiagen Cat. 28106). DNA isolated from the ChIPs was analyzed by quantitative PCR (ChIP-qPCR) using the Bio-Rad CFX96 Real-Time System. Sequences of the primers used for qPCR are listed in Table S3.
For ChIP-chips, recovered DNA was amplified with REPLI-g (Qiagen Cat. 150025), digested with DNase, labeled and hybridized on Affymetrix Drosophila tiling 2.0R arrays. For each antibody we perform 2 independent biological replicates. ChIP-chip hybridization intensities were analyzed using R and R/Bioconductor packages as described [65]. In brief, log2(ChIP/Input) and log2(Mock/Input) ratios were quantile normalized in parallel. Then, averaged ratios were median scaled and log2(Mock/Input) ratio was subtracted from log2(ChIP/input) ratio resulting in ChIP-chip score. Finally, peaks were selected based on random permutations at false discovery rate (FDR)<0.01.
Nuclear extracts (NEs) from 0–12 hour old Drosophila embryos or S2 cells were prepared as described [66]. Immunopurification procedures were also performed essentially as described [66]. Briefly, for immunopurification of NAP1 and cohesin, extracts were incubated with affinity-purified antibodies cross-linked to protein A sepharose beads (GE Healthcare 17-0963-03) by dimethylpimelimidate. After incubation, beads were washed extensively with HEMG buffer: 25 mM HEPES-KOH pH 7.6, 0.1 mM EDTA, 12.5 mM MgCl2, 10% glycerol, a cocktail of protease inhibitors, and containing 600 mM KCl, 0.1% NP-40 (HEMG/600). Proteins retained on the beads were eluted with 100 mM NaCitrate buffer pH 2.5, resolved by SDS-PAGE and visualized by colloidal-blue staining or immunoblotting. For denaturing IPs of SA, cells were lysed in a buffer containing 6M Urea, 150 mM NaCl, 5 mM DTT and 50 mM Tris pH 7.6 followed by the step-wise dialysis to lower the Urea concentration to 2M. Finally, the extract were dialyzed against HEMG/400 and SA was immunoprecipitated as described above.
NE fractionation by (NH4)2SO4 precipitation, POROS-heparin and Sephacryl S-300 size-exclusion chromatography were performed as described [67]. In brief, NEs were concentrated by chromatography on a POROS-heparin (PerSeptive Biosystems) column equilibrated with HEMG/100 (pH 7.6) buffer followed by a step elution with HEMG/400 (pH 7.6) buffer (H0.4 fraction). The H0.4 fraction was loaded onto an 800-ml Sephacryl S-300 column (Pharmacia) equilibrated and developed with HEMG/100 (pH 7.6) buffer.
Mass spectrometry analysis of immunopurified protein complexes was performed on a LTQ-Orbitrap hybrid mass spectrometer (ThermoFischer) as described [68]. Detected peptides were matched against the FlyBase database (http://www.flybase.org/) using a Mascot search algorithm and identified proteins, Mascot scores and number of unique peptides are listed in Table S1 and Table S2. Details will be made available upon request.
Recombinant HA-tagged NAP1 (HA-NAP1) was expressed and purified using the baculovirus expression system. For competition assays, the cohesin complex bound to PP2A was immunopurified from embryo nuclear extracts with antibodies against SA or SMC1 as described above. Purified HA-NAP1 was incubated with the cohesin complex captured on Protein A beads followed by a series washes with HEMG/600. Proteins retained on the beads were resolved by SDS-PAGE and analyzed by immunoblotting. For interaction assays, HA-NAP1 captured on anti-HA coated beads was incubated with recombinant [35S]methionine-labeled cohesin subunits, PP2A subunits and Polo produced by in vitro transcription/translation system (Promega L1170). Proteins bound to NAP1 were resolved by SDS-PAGE and detected by autoradiography.
Data were deposited to Gene Expression Omnibus under accession number GSE30938 (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?token=xnuhnwkmqqmwcli&acc=GSE30938).
|
10.1371/journal.pgen.0030071 | Y Chromosome Lineage- and Village-Specific Genes on Chromosomes 1p22 and 6q27 Control Visceral Leishmaniasis in Sudan | Familial clustering and ethnic differences suggest that visceral leishmaniasis caused by Leishmania donovani is under genetic control. A recent genome scan provided evidence for a major susceptibility gene on Chromosome 22q12 in the Aringa ethnic group in Sudan. We now report a genome-wide scan using 69 families with 173 affected relatives from two villages occupied by the related Masalit ethnic group. A primary ten-centimorgan scan followed by refined mapping provided evidence for major loci at 1p22 (LOD score 5.65; nominal p = 1.72 × 10−7; empirical p < 1 × 10−5; λS = 5.1) and 6q27 (LOD score 3.74; nominal p = 1.68 × 10−5; empirical p < 1 × 10−4; λS = 2.3) that were Y chromosome–lineage and village-specific. Neither village supported a visceral leishmaniasis susceptibility gene on 22q12. The results suggest strong lineage-specific genes due to founder effect and consanguinity in these recently immigrant populations. These chance events in ethnically uniform African populations provide a powerful resource in the search for genes and mechanisms that regulate this complex disease.
| The parasitic disease kala-azar, or visceral leishmaniasis, is associated with liver, spleen, and lymph gland enlargement, as well as fever, weight loss, and anaemia. It is fatal unless treated. Three major foci of disease occur in India, Sudan, and Brazil. Importantly, 80%–90% of infections are asymptomatic. Understanding why two people with the same exposure to infection differ in susceptibility could provide important leads for improved therapies. We studied families with multiple cases of clinical disease from two villages in Sudan. After typing 300–400 genetic markers across the human genome, we determined which chromosomes carry susceptibility genes. We were surprised that our results differed from those published earlier for a village 100 kilometers from our site. All of these villages are occupied by people of the same ethnic group who migrated from western Sudan late last century following a major drought. We stratified our analysis by village, and used male Y chromosome markers to tag extended pedigrees. Our results suggest that recent immigration, in combination with consanguineal marriage in a strongly patriarchal society, has amplified founder effects resulting in different lineages within each village carrying different susceptibility loci. This demonstrates the importance of understanding population genetic substructure in studying genes that regulate complex disease.
| Ninety percent of clinical visceral leishmaniasis (VL) cases caused by protozoa of the L. donovani species complex (L. donovani, L. archibaldi, L. infantum, and L. chagasi) occur in three foci in India/Bangladesh/Nepal, Sudan, and Brazil. Skin-test data and lymphocyte proliferation assays [1–4] indicate that only one in 5–10 infected individuals develop clinical disease. Familial aggregation is a feature of VL in Brazil [5], providing a high relative risk (λ2S = 34) of disease in further siblings of affected sibling pairs [6]. In Sudan, familial clustering and marked differences in incidence of clinical disease and skin-test reactivity between villages inhabited by different ethnic groups that share environment and exposure [7,8] support a contribution of host genotype to susceptibility. In mice [9], different genes control innate versus adaptive immune responses, and we expect genetic control in humans to be complex. Nevertheless, understanding the genes/mechanisms that determine why two people with the same exposure differ in susceptibility to VL could provide important leads for improved therapies. A number of candidate gene studies have been undertaken that support roles for polymorphisms at SLC11A1, IL4, and IFNGR1 in controlling susceptibility to visceral leishmaniasis or post kala-azar dermal leishmaniasis in Sudan [10–12]. A genome-wide scan recently undertaken in eastern Sudan by Bucheton et al. [13] also reported a major gene (LOD score 3.5; p = 3 × 10−5) on Chromosome 22q12 controlling VL in the Aringa ethnic group. We now report on a second genome-wide scan undertaken in two villages occupied by the related Masalit ethnic group in eastern Sudan, in which we provide evidence for major loci at 1p22 and 6q27 that are Y chromosome–lineage and village-specific. Neither village provided evidence for a VL susceptibility gene on 22q12. The results suggest strong lineage-specific genes due to founder effect and consanguinity in these recently immigrant populations, and point to the potential power these chance events in ethnically uniform African populations provide in the search for genes and mechanisms that regulate this complex disease.
In their study, Bucheton et al. [13] used 63 multicase families from members of the Aringa ethnic group living in Barbar El Fugura, Gedaref State, eastern Sudan (Figure 1). The Aringa people in this village migrated from western Sudan/Chad to settle as subsistence farmers in the 1940s [13]. We worked [11,12] in two villages, El-Rugab and Um-Salala, located ~40 km apart in Galabat Province, Gadaref State, ~100 km south of Barbar El Fugura [13]. El-Rugab and Um-Salala are occupied by Nilosaharan speaking Masalit, who also migrated from western Sudan, starting in 1969, to occupy villages in the heart of the endemic area in eastern Sudan. The two villages have high rates of clinical VL. Using 38 pedigrees (48 nuclear families; Table 1), we performed a 360 microsatellite ~10-cM genome-wide scan. Lander and Kruglyak [14] proposed a classification for reporting the results of genome-wide scan data based on the number of times one would expect to see a result at random in a dense, complete genome scan. The thresholds they propose are: “suggestive linkage,” where statistical evidence would be expected to occur one time at random in a genome scan; “significant linkage,” 0.05 times; “highly significant linkage,” 0.001 times; and “confirmed linkage,” where significant linkage from an initial scan has been confirmed with a nominal p value of ≤0.01 in a second independent study. In the case of a sib-pair study, the first three categories correspond to point-wise significance levels of 7 × 10−4, 2 × 10−5, and 3 × 10−7 (LOD scores 2.2, 3.6, and 5.4). To analyze our genome-wide scan data we used nonparametric linkage analysis comparing identity-by-descent allele sharing across all relative pairs in the extended pedigrees. The thresholds proposed by Lander and Kruglyak vary marginally for other relative pairs combinations [14]. Although other authors consider these thresholds to be overly conservative [15], they serve as a guide to evaluate the significance associated with the nominal point-wise p values provided here. All p values reported here are nominal one-sided p values, except where we state that we have carried out simulations to calculate empirical p values. The nonparametric linkage analysis for the primary genome scan in Sudan provided evidence for linkage of VL to four regions on Chromosomes 1, 5, 6, and 13 (LOD scores 0.77 to 2.83; p values 1.5 × 10−4 < p < 0.029) across El-Rugab and Um-Salala (Figure 2). None of these matched regions that were identified in the earlier [13] primary scan of 38 families. As observed for the Masalit in our area [7,8], the Aringa ethnic group is highly susceptible to VL compared to the neighboring Hawsa and Fellata ethnic groups [13], and is closely related to the Masalit. Our failure to observe overlap with the earlier [13] study led us to consider whether founder effect and population substructure could influence genes regulating VL in different villages in eastern Sudan.
The Masalit ethnic group is polygamous; males have up to four wives at any time, depending on socio-economic status. The founding population for each village comprised 10–15 related males with their families. By 1997 population sizes were ~2,000 in El-Rugab and ~1,200 in Um-Salala. Using the program Pedigree Relationship Statistical Test (PREST) [16] to estimate relatedness between individuals we found that ~43% of marriages in El-Rugab and ~33% within Um-Salala were consanguineous, with varying degrees of relatedness between parents (Table 2). The combination of chance events in the founder population and consanguineal marriages could affect the genetic composition of each village. When the primary genome scan data for Sudan was stratified by village, the bimodal linkage on Chromosome 1 (Figure 2) resolved into two clear village-specific peaks: at 1p22 for Um-Salala (peak LOD score at D1S2868 = 2.81; p = 1.62 × 10−4) and 1q31.3 for El-Rugab (peak LOD score at D1S238 = 1.31; p = 0.007). Refined mapping provided genome-wide [14] evidence (LOD score 3.8; p = 1.45 × 10−5) for a major susceptibility gene at D1S2868 on 1p22 in Um-Salala, and improved the evidence (LOD score 1.59; p = 0.003) for linkage at D1S238 on 1q31.3 in El-Rugab (Figure 3A). Similarly, stratification and refined mapping on 6q27 (Figure 4A) provided evidence for a common susceptibility locus at D6S1719 (LOD score 2.13; p = 8.74 × 10−4) affecting both villages, with genome-wide [14] significance (LOD score 3.58; p = 2.47 × 10−5) for a susceptibility gene at D6S281 in El-Rugab alone.
To determine the contribution that loci at 1p22 and 6q27 make to genetic susceptibility to VL in each village, we estimated [17] the locus-specific λS from the ratio of the expected (0.25) proportion of affected sibpairs sharing zero alleles identical-by-descent under no linkage to the observed proportion. This ratio indicates the risk to siblings of affected individuals compared to the general population risk. At 1p22 in Um-Salala the λS was 2.90, and 1.63 at 6q27 in El-Rugab. Thus, these loci make an appreciable contribution to the total genetic component of disease susceptibility in each village, equivalent to or greater than that observed for the major genes for VL [13] at 22q12 (λS = 1.83 for all families; 2.11 for families affected towards the beginning of an outbreak), or for leprosy reported from genome scans at 10p13 in India (λS = 1.66) [18] and 6q25 in Vietnam (λS = 2.21) [19].
To understand more about population substructure within villages, we genotyped Y chromosome markers [20] for all male family members. In all cases, all available males within a pedigree carried the same Y chromosome haplotype. Two main Y haplotypes [20,21] were present, A3b2 from haplogroup A and E3b1 from haplogroup E, shown by pedigree in Table 1. Stratifying the linkage analysis by Y chromosome haplotype demonstrated strong lineage-specific effects within each village. The peak at 1p22 in Um-Salala was contributed to only by E3b1 lineages (Figure 3B); peak LOD score was 5.65 (p = 1.7 × 10−7), with locus-specific λS 5.07 at D1S2868. Consistent with the result obtained using stratification by village, neither E3b1 nor A3b2 lineages from El-Rugab village showed linkage to disease at 1p22 (Figure 3C). The peak at 6q27 (Figure 4B) for El-Rugab was also specific to E3b1 families; peak LOD score was 3.74 (p = 1.68 × 10−5), locus-specific λS 2.31 at D6S281. Again, neither E3b1 nor A3b2 lineages from Um-Salala village showed linkage to disease at the major peak at D6S281 (Genethon cM position 186.7) on 6q27 (Figure 4C). Although there were fewer A3b2 families (Table 1), and hence less power to observe positive linkage using these families on their own, their removal from the analysis within each village clearly enhances the LOD scores obtained for the E3b1 pedigrees indicating that lack of linkage in these families was reducing the LOD scores. To evaluate the significance of our results given the possibility of type I errors due to over-relatedness of parents in the pedigrees [22,23], simulations were performed after adding inbreeding loops to the pedigrees based on the relationships between parents determined by the PREST analysis (Table 2). For the peak at D1S2868 in Um-Salala, a LOD score of 5.65 was never observed in 100,000 simulations, providing a point-wise empirical p value <1 × 10−5 that retains genome-wide significance [14]. This is consistent with the LOD score 4.3 (p = 4.3 × 10−6) obtained when the real data were re-analysed with inbreeding loops added to the pedigrees. For the peak at 6q27 in El Rugab, the point-wise empirical p value was <1 × 10−4 (10,000 simulations performed), again broadly in line with the LOD score 2.84 (p = 1.5 × 10−4) obtained for real data re-analysed with inbreeding loops added to the pedigrees. Given the increased locus-specific λS values observed with the stratified analyses, the susceptibility loci at 1p22 in Um Salala and 6q27 in El Rugab make a substantial contribution to the genetic susceptibility to disease in E3b1-tagged lineages in these villages. In our analysis of population substructure we also sequenced the hypervariable HVS-I region [24] of the mitochondrial genome for all families (Text S1). A diverse range of mitochondrial haplogroups was observed (Table S1) that showed little overlap across Y chromosome lineages within villages.
One interpretation of the Y chromosome stratified data could be parent-of-origin effects. We looked at this in two ways: (i) by comparing identity-by-descent allele-sharing for maternally versus paternally derived alleles across all affected relative-pairs in the pedigrees (Table 3); and (ii) by comparing GENEHUNTER-MODSCORE results under trait models with or without imprinting (Table 4). Neither method provided evidence for parent-of-origin effects at either 1p22 or 6q27. This means that susceptibility alleles introduced into villages by the small number of founding males and their families are being transmitted by both male and female parents. The multipoint GENEHUNTER-MODSCORE analysis (Table 4), which is a parametric analysis that maximizes the LOD score with respect to penetrances and disease allele frequencies, also demonstrates recessive inheritance for susceptibility alleles at both 1p22 and 6q27, consistent with transmission of disease alleles through both male and female parents.
Here, we identified major lineage-specific genes at 1p22 and 6q27 controlling VL in adjacent Masalit villages in eastern Sudan. We hypothesized that chance events in the founding males and their families carried specific susceptibility alleles into each village, and that consanguineal marriage within patriarchal communities has enriched these alleles, leading to strong lineage-specific effects. This explains failure to replicate the major linkage peak on Chromosome 22q12 [13], although differences in disease phenotype and parasite strain could also contribute. In this setting in Sudan, Y chromosome haplotypes have provided tags for recently immigrant extended pedigrees that contribute to different linkages within, but not between, each village. This is consistent with the likelihood that Y chromosome haplotypes will only serve to mark the autosomes over a limited number of generations, in this case marking the autosomes carrying susceptibility alleles introduced by founders of extended pedigrees within each village. In this patriarchal society, mitochondrial haplotypes were more heterogeneous than Y chromosome haplotypes, and did not serve to mark the origin of the autosomes carrying the susceptibility alleles. Nevertheless, as expected for autosomal genes, susceptibility alleles introduced through the founding families tagged by Y chromosome haplotypes were inherited through both parents within the extended pedigrees. Our results demonstrate that an understanding of population substructure can contribute to the identification of disease susceptibility alleles in these recently immigrant African populations. In other settings, mitochondrial haplotypes, or a combination of Y and mitochondrial haplotypes, might both provide important markers of population substructure influencing the frequency and distribution of disease susceptibility alleles in different populations, as may a selection of autosomal markers.
Another important component of our study was that, in our analyses, we also took into account consanguinity and over-relatedness of parents within the pedigrees, which can lead to type I errors and inflated LOD scores [22,23]. It is possible that the LOD score of 3.5 (p = 3 × 10−5) reported by Bucheton et al. [13] for Chromosome 22q12 may also be inflated by consanguinity within the pedigrees used. To account for consanguinity, we employed a strategy similar to that used by Riaz et al. [25], who specifically selected highly inbred families to increase power in demonstrating genome-wide significance for linkage of stuttering to Chromosome 12 in 44 Pakistani families. In the absence of definitive information on the relationship between parents, the primary genome scan and additional markers used for refined mapping provide data from a large number of microsatellite markers that can be used to specify the level of inbreeding using PREST [16]. In addition, given the strong patriarchal society in this region of Sudan, we used Y chromosome haplotypes to specify families belonging to extended pedigrees within each village. Simulations performed after the addition of inbreeding loops to the pedigrees allowed us to determine a point-wise empirical p value <1 × 10−5 associated with the LOD score of 5.65, providing genome-wide significance for a visceral leishmaniasis susceptibility locus on Chromosome 1p22. Similarly, an empirical p value <1 × 10−4 was associated with the LOD score 3.74 for a locus at 6q27. Hence, we are confident that these two regions of the genome carry susceptibility genes for visceral leishmaniasis in this region of Sudan, and we have supporting evidence that a gene at 6q27 may also contribute to susceptibility to visceral leishmaniasis in Brazil [26].
Interesting candidate genes are located at 1p22, including DR1, which encodes the downregulator of transcription 1/TBP-binding negative cofactor 2, and inhibits transcription by binding to the TATA box binding protein [27] (TBP) located at 6q27. CIITA, the transactivator of major histocompatibility complex class II molecules in antigen presenting cells, requires the participation of, and is extremely sensitive to mutations in, TBP [28]. Other 1p22 candidates include glomulin (GLMN, also called FKBP-associated protein FAP48), which is antiproliferative for T cells[29], and growth factor-independent 1 (GFI1), which influences myeloid differentiation [30]. The Notch ligand delta-like 1 (Drosophila) (DLL1) and proteasome subunit beta-type 1 (PSMB1) are also at 6q27. Proteasome function is important in degradation of proteins by antigen-processing cells, which use the Notch pathway to instruct T cell differentiation [31]. Specifically, DLL1 induces T helper 1 cells to release interferon-γ, which is crucial for immune control of L. donovani [32]. Identification of the etiological genetic variants at 1p22 and 6q27 will contribute to our understanding of the complex interaction of genes and mechanisms associated with susceptibility to this important protozoan disease in humans.
The study was carried out in two villages, El-Rugab and Um-Salala, on the eastern bank of the River Rahad (Galabat Province, Gadaref State) in the heart of the endemic area in eastern Sudan (Figure 1). They are occupied by members of the Nilosaharan-speaking Masalit ethnic group who migrated to the area from El-Geneina in Darfur Province, western Sudan, principally following the great drought of the early 1980s, although a small number of founders (two brothers carrying the E3b1 haplotype and one sister and their respective families) first established Um-Salala village in 1969 [33]. Um-Salala village lies 40 km to the north of El-Rugab. At the December 1997 census the population of Um-Salala was 1,225 (566 males; 659 females) and that of El-Rugab was 2,084 (1039 males; 1045 females). These villages have been under annual surveillance for VL by the Institute of Endemic Diseases since the mid-1980s. In 1996 a treatment centre was established in the area by Médecins Sans Frontières. Epidemiological and further demographic details relating to the study site are described in detail elsewhere [7,33,34]. The annual incidence rate of VL cases is ~38.5/1,000 persons/y. The reported [33] mean ± SD age at diagnosis for clinical VL is 7.5 ± 5.1 y; all except one patient were <17 y of age. Isoenzyme typing of parasites isolated from cases during the 1995/1996 season revealed the presence of three zymodemes in this area, corresponding to L. donovani s.s., L. infantum and L. archibaldi [33]. Parasite isolates were not made from all cases used in this study. For this study the noninvasive buccal swab technique was used, and DNA extracted from the buccal swab buffer or amplified directly from the buffer using multiple displacement amplification (MDA, Molecular Staging; http://www.qiagen.com). Ethical approval for the study was obtained from the Institutional Review Board of the University of Khartoum, Khartoum, Sudan. Informed consent for sample collection was obtained from adults, and from the parents of children <18 y old.
Multicase families with VL were ascertained from epidemiological and medical records of the Institute of Endemic Diseases and Médecins Sans Frontières. Families were pursued where records indicated two or more cases in the family. Households generally comprised a senior male with his sons, together with their (multiple) wives and families. When available, both fathers and mothers of families with VL cases were interviewed to determine pedigree structure, with DNA collected from those parts of the pedigree that were informative for linkage. A total of 69 nuclear families from 57 pedigrees were used at different stages of the study (Table 1). Of these, 97 individuals from 13 pedigrees with 20 nuclear families from Um-Salala village, and 123 individuals from 25 pedigrees with 28 nuclear families from El-Rugab village, were successfully genotyped and integrity of their families verified using the computer program PedCheck [35] for 360 out of 400 microsatellite markers typed for the primary genome scan. Additional families were used to complement the data at later stages of the study, but these did not have sufficient power to independently replicate linkage. Table 1 shows the numbers of individuals in families available for analysis after checking for genetic integrity (PedCheck [35]) within all families. Nuclear families with one affected offspring were always part of a larger pedigree, and so contributed to the linkage analysis, which compared allele-sharing across all relative pairs within the pedigree (cf. below).
All of the individuals classified as affected in these families were diagnosed with clinical VL that responded positively to specific anti-leishmanial treatment. In this respect our affection status may be less severe than that used by Bucheton et al. [13], who reported that 2% to 5% of VL subjects died in spite of treatment. For our study, diagnosis of clinical VL was made on the basis of clinical, parasitological, and serological criteria as described [7,11,12,33]. At initial presentation, symptoms suggestive of VL included fever, often prolonged and not cyclical (differential diagnosis for malaria), pale continence of skin due to anaemia, weight loss, hepatosplenomegaly, and generalized lymphadenopathy. Specific antibody was measured using the direct agglutination test (DAT). Parasitological confirmation of cases was made by examination of Giemsa-stained lymph node aspirates. All clinical examinations were carried out by experienced clinical staff from the Institute of Endemic Diseases Leishmaniasis Research Group (83% of cases) or Médecins Sans Frontières (17% of cases). Only cases that were DAT positive, parasitologically confirmed by aspirates, and responded to specific anti-leishmanial treatment were included in the study. Subclinical forms of disease [33] were not included. The 57 families (Table 1) included 84 affected males (mean ± SD age at diagnosis 8.56 ± 4.02; range 2–16) and 87 affected females (mean ± SD age at diagnosis 8.86 ± 4.77; range 2–20). The affected parents reported in Table 1 were all historical cases >25 y of age at the time of sample collection.
For the primary genome scan, families were genotyped for the 400 markers that make up the Applied Biosystems medium density 10 cM resolution human index map. Data from 360 (90%) of these markers that were successfully genotyped (i.e., mean ± SD 79 ± 9% of individuals typed for every marker) and verified using PedCheck [35] were used for analysis of the primary genome scan. Refined mapping was performed on these families by successful genotyping and PedCheck [35] of 46 additional microsatellite markers in regions positive for linkage at p < 0.05. To complement data from the primary genome scan, 32 markers in positive regions from the ABI medium density map, and the 46 additional markers, were successfully genotyped and PedChecked [35] in additional families (Table 1). Tests for Hardy-Weinberg Equilibrium (HWE) were performed within Stata using genetically independent family members from the two villages. All markers used for linkage analysis were in HWE (data not shown).
SNaPshot assays using the ABI PRISM SNaPshot Multiplex System (http://www.appliedbiosystems.com) were designed to detect SNPs M02, M13, M40, M42, M78, M89, M118, M144, M145, M148, M171, M181, M215, and M224 as defined by Underhill et al. [20] and the Y Chromosome Consortium [21]. Carriage of the ancestral allele at M42, variant alleles at M144 and M13, and ancestral alleles at M118 and M171, classified individuals as belonging to haplotype A3b2 in haplogroup A [21]. Carriage of variant alleles at M42, M145, M40, M215, M78, and M224 classified individuals as belonging to haplotype E3b1 in haplogroup E [21]. Males from four families from El-Rugab carried variant alleles at M42, but were wild type at M145 and M181 which excluded them from haplogroups A, B, D, or E. On the basis of previously reported [20] haplogroups from Sudan, and the fact that haplogroup C does not occur in Africa, we placed these males into haplogroup J but we did not determine their specific haplotype. Lineages carrying the Y chromosome haplotypes A3b2 and E3b1 appeared equally susceptible to VL, as evidenced by the presence of 55.6% affected individuals in A3b2 families and 52.8% in E3b1 families for which phenotypic data were available for all sibs in the nuclear family. Ability to identify genes that control susceptibility to VL in the A3b2 haplogroup A families was limited since there were only four and two nuclear A3b2 families included in the primary genome scan families from El Rugab and Um-Salala, respectively. For refined mapping, there were eight and six A3b2 families, respectively (Table 1).
Nonparametric multipoint linkage analyses were performed in ALLEGRO [36], with results reported as allele sharing LOD scores [37] and plotted as sign(dhat)•LOD [36]. The Spairs scoring function with 0.5 weighting was used to take account of differences in pedigree size [36]. The Spairs scoring function compares allele-sharing identical-by-descent across all relative pairs within the extended pedigrees, including the singleton families within larger pedigrees. Unaffected members of pedigrees were included to assist ALLEGRO to infer missing parents' genotypes. Of the multicase nuclear families (N = 9 Um-Salala; N = 8 El-Rugab) where there was one missing parent, all had a minimum of two offspring and most (N = 6 Um-Salala; N = 5 El-Rugab) had three to six offspring. Two families with three (Um-Salala) and four (El-Rugab) offspring had both parents missing. Allele frequencies for the microsatellite markers were calculated separately for each stratified analysis in SPLINK [38], which uses unrelated individuals in the pedigrees to calculate frequencies. Information content for markers was estimated in ALLEGRO. All LOD scores reported are multipoint. Simulations (100) performed within ALLEGRO using data for a typical set of six linked polymorphic microsatellite markers (7–10 alleles; heterozygosity 0.73) showed that the primary genome scan family set (Table 1) across both villages had 100% power to detect a major gene at an allele-sharing LOD score = 3.00; p = 1.02 × 10−4, and >98% power to detect an allele-sharing LOD score = 3.95; p = 1.01 × 10−5. The separate El-Rugab and Um-Salala primary genome scan family sets had >94% and >79% power to detect a major gene at an allele-sharing LOD score = 2.07; p = 0.001, respectively. The primary scan plus additional families (refined map) had >96% and >97% power to detect a major gene at an allele-sharing LOD score = 2.07; p = 0.001 for El-Rugab and Um-Salala, respectively.
Parametric linkage analysis was performed using GENEHUNTER-MODSCORE [39], which maximizes the parametric LOD score with respect to penetrances and disease allele frequency, and is a further development of GENEHUNTER-IMPRINTING [40] based on the original GENEHUNTER version 2.1 [41]. Multipoint analysis was performed under trait models with and without imprinting. To allow for imprinting, two penetrance parameters (instead of one fHet in the analysis for no imprinting) were specified for individuals who are heterozygous, one for paternal origin and one for maternal origin of the disease allele. The four penetrance parameters are f+/+, fm/+, f+/m, and fm/m, where + specifies the wild-type allele and m the mutant allele, and the paternally inherited allele is listed first. In conjunction with the MOD score, the analysis yields the penetrances and disease allele frequency (p) of the best-fitting trait model. The estimate of the disease allele frequency obtained by MOD-score analysis has the largest variance of all trait-model parameters [42], and can be higher than the true value because specifying a higher frequency can compensate for a general model misspecification and hence lead to robustness in a multipoint analysis [43]. The MOD-score analysis transforms penetrances to provide a dominance index D, and an index of imprinting I. D is positive (dominant model) if both heterozygote penetrances equal fm/m, negative (recessive model) if both heterozygote penetrances equal f+/+, and zero (semi-dominant or additive) if the average of the two heterozygote penetrances is halfway between the two homozygote penetrances. I is positive (maternal imprinting) or negative (paternal imprinting) if one heterozygote penetrance equals f+/+ and the other fm/m, and zero (no imprinting) if both heterozygotes penetrances are equal. A comparison of the difference in peak MOD scores obtained with or without imprinting provides an additional test for presence of imprinting [42]. Simulations performed by Weeks et al. [44] and Hodge et al. [45] show that a critical value of 3, used for LOD scores, should be adjusted by some value in the range of 0.3 to 1.0 to maintain a similar type I error. To account for the additional parameter of imprinting, a further adjustment of the critical value is necessary. Strauch et al. [42] propose that MOD scores >3.5 obtained with the imprinting parameter provide at least suggestive [14] evidence for linkage. Data for MOD scores are provided only where this critical value is achieved or exceeded.
A second evaluation of parent-of-origin effects or imprinting was determined by comparing sharing of maternally derived alleles identical-by-descent with paternally derived alleles identical-by-descent in affected relative pairs using the program MERLIN [46] that allows rapid analysis of allele sharing across extended pedigrees.
To estimate the contribution that specific loci make to the genetic component of disease susceptibility, we estimated [17] the locus-specific λS from the ratio of the expected (0.25) proportion of affected sibpairs sharing zero alleles identical-by-descent under no linkage to the observed proportion across all pedigrees, as determined using affected sibpair linkage analysis in GENEHUNTER [41].
The program PREST [16] was used to estimate degree of relatedness between individuals in our families by estimating the probabilities P0, P1, and P2 of two individuals sharing 0, 1, and 2 alleles identical-by-descent over the 360 microsatellite markers successfully genotyped in the primary genome scan, or 78 markers successfully genotyped in all families during refined mapping. Unrelated individuals should have probabilities 1, 0, and 0, respectively. Full sibs have P0 sharing probabilities of 0.25; half sibs plus first cousin 0.375; grandparent-child, avuncular or half-sib 0.50; double first-cousins 0.5625; first cousins or half-avuncular 0.75; half first cousins 0.875; and second cousins 0.9375. We therefore considered any relationship between parents in the families with P0 sharing probabilities <0.95 to be indicative of a consanguineal marriage, with predicted relationships as outlined in Table 2. On the basis of these predicted relationships, we used a procedure similar to that adopted by Riaz et al. [25], adding appropriate inbreeding loops to the pedigrees and carrying out 100,000 simulations for E3b1 pedigrees from Um-Salala using the refined Chromosome 1 map, and 10,000 simulations for E3b1 pedigrees from El-Rugab using the refined Chromosome 6 map, to determine empirical p values associated with the observed LOD scores for linkage to VL susceptibility.
World Health Organization data on global distribution, prevalence, and incidence of visceral leishmaniasis can be found at http://www.who.int/tdr/diseases/leish/default.htm. Pregap and Gap4 programs are available at http://www-gap.mcs.st-and.ac.uk/Download/index.html. Information about the Applied Biosystems medium density map and microsatellite markers is available at https://products.appliedbiosystems.com/ab/en/US/adirect/ab?cmd=catNavigate2&catID=600770. Details of the SPLINK program are available at http://www-gene.cimr.cam.ac.uk/clayton/software/splink.txt.
The Swiss-Prot database (http://expasy.org/sprot) accession numbers for the proteins discussed in this paper are DLL1, O00548; DR1, Q01658; GFI1, Q99684; GLMN, Q92990; PSMB1, P20618; and TBP, P20226. |
10.1371/journal.pgen.1005388 | Calmodulin Methyltransferase Is Required for Growth, Muscle Strength, Somatosensory Development and Brain Function | Calmodulin lysine methyl transferase (CaM KMT) is ubiquitously expressed and highly conserved from plants to vertebrates. CaM is frequently trimethylated at Lys-115, however, the role of CaM methylation in vertebrates has not been studied. CaM KMT was found to be homozygously deleted in the 2P21 deletion syndrome that includes 4 genes. These patients present with cystinuria, severe intellectual disabilities, hypotonia, mitochondrial disease and facial dysmorphism. Two siblings with deletion of three of the genes included in the 2P21 deletion syndrome presented with cystinuria, hypotonia, a mild/moderate mental retardation and a respiratory chain complex IV deficiency. To be able to attribute the functional significance of the methylation of CaM in the mouse and the contribution of CaM KMT to the clinical presentation of the 2p21deletion patients, we produced a mouse model lacking only CaM KMT with deletion borders as in the human 2p21deletion syndrome. No compensatory activity for CaM methylation was found. Impairment of complexes I and IV, and less significantly III, of the mitochondrial respiratory chain was more pronounced in the brain than in muscle. CaM KMT is essential for normal body growth and somatosensory development, as well as for the proper functioning of the adult mouse brain. Developmental delay was demonstrated for somatosensory function and for complex behavior, which involved both basal motor function and motivation. The mutant mice also had deficits in motor learning, complex coordination and learning of aversive stimuli. The mouse model contributes to the evaluation of the role of methylated CaM. CaM methylation appears to have a role in growth, muscle strength, somatosensory development and brain function. The current study has clinical implications for human patients. Patients presenting slow growth and muscle weakness that could result from a mitochondrial impairment and mental retardation should be considered for sequence analysis of the CaM KMT gene.
| Calmodulin (CaM) is a highly abundant, ubiquitous, small protein, which plays a major role in the transmission of calcium signals to target proteins in eukaryotes. Hundreds of CaM targets are known, and their respective cellular functions include signaling, metabolism, cytoskeletal regulation, and ion channel regulation, to name but a few. CaM is frequently modified after translation, including frequently trimethylation at a single amino acid, however, the role of this methylation is not known. Human patients with a homozygous deletion of the gene that methylates CaM, CaM-KMT, are known, but they also have a deletion of additional genes. Thus, to study the role of CaM–KMT, we produced a mouse model in which CaM-KMT is the only deleted gene, with the deletion constructed as in the human patients. The model proved to reveal the function of methylation of CaM, since CaM was found to be non-methylated and the methylation of CaM found to be important in growth, muscle strength, somatosensory development and brain function. The current study also has clinical implications for human patients. Patients presenting slow growth and muscle weakness that could result from a mitochondrial impairment and mental retardation should be considered for sequence analysis of the CaM KMT gene.
| Calmodulin (CaM) is a highly abundant, ubiquitous, small, acidic protein, which plays a major role in the transmission of calcium signals to target proteins in eukaryotes. Hundreds of CaM targets are known, and their respective cellular functions include signaling, metabolism, cytoskeletal regulation, and ion channel regulation, to name but a few. CaM target proteins include, for example, several CaM-dependent protein kinases (CaMK), other enzymes, myosins, receptors, ion channels, and a number of other proteins. The affinity of CaM towards its protein targets varies between the nM and μM ranges [1].
CaM was found to be frequently trimethylated at Lys-115, a solvent-accessible residue (see PDB file 1UP5 5). A limited number of studies have shown that the methylation state of CaM can change in developmental and tissue-dependent manners [2–4], influence the activator properties of CaM with target enzymes [5] and cause phenotypic changes in growth and developmental processes at the level of a whole organism [4,6]. These observations suggest that CaM methylation could be a dynamic mechanism attenuating the interaction of CaM with target proteins influencing a plethora of eukaryotic cellular and developmental processes. [7] However, all observations regarding the role of methylation on CaM were in plants and insects, the effect in vertebrates was to the best of our knowledge not demonstrated. We recently identified the CaM lysine methyl transferase (CaM KMT) and found it to be highly conserved from plants to vertebrates [7]. The human form was previously known as c2orf34 residing at locus 2p21. We reported that deletions of this gene in homozygosity, in the contiguous gene syndrome named the 2P21 deletion syndrome (MIM #606407), are associated with cystinuria, intellectual disabilities, hypotonia, mitochondrial disease and facial dysmorphism [8]. Since the syndrome involves the deletion of 4 genes: PP2Cβ, SLC3A1, PREPL and CaM KMT [9–11], the specific contribution of CaM KMT to the clinical presentation of the patients could not be determined [12]. Further delineation of the contribution of the deleted genes to the patients' phenotype was revealed by the identification of patients with smaller deletions including only SLC3A1 and PREPL, these patients presented only with hypotonia and cystinuria (HCS) [13–16]. Two additional sibling patients with deletion in SLC3A1, PREPL and CaM KMT presented in addition to the features presented in classical HCS a mild/moderate mental retardation and a respiratory chain complex IV deficiency in one of the patients [17]. Two additional patients with deletion of just PREPL and CaM KMT presented hypotonia and a mild/moderate mental retardation, the mitochondrial activity was not verified in these patients [18]. Recently it was reported that deletion of PREPL causes growth impairment and hypotonia in mice [19], observed also in patients with smaller deletions in the 2p21 interval, not including CaM KMT [16]. Thus we set forth to produce a mouse model lacking only the CaM KMT gene with deletion borders as in the human 2p21deletion syndrome to be able to attribute the functional significance of the methylation of CaM in the mouse and the contribution of CaM KMT to the clinical presentation of the 2p21deletion patients.
The mouse homolog of the gene has the same gene structure and spliced forms as the human, and the mouse syntenic region is identical to the human. Thus we expected that by creating a mouse model, in which the coding sequence of the first exon and the adjacent 300 bp of the first intron are deleted, exactly like the border of the deletion of this gene in the 2p21 deletion syndrome patients, will reveal which components of the human phenotype are caused by the specific deletion of CaM KMT. We did not include the 5' untranslated sequence of the first intron in the deletion in order not to affect regulatory regions. The generation of the mouse model began with the design and production of a targeting construct designed to replace the coding sequence and the 5' of the first intron by β -Gal and the Neo gene flanked by floxed repeats. The scheme of the construct is presented in Fig 1A. The results of the PCR reaction used to determine the genotype of the mice is shown in Fig 1B. To ensure that CaM KMT is not expressed in the CaM KMT-/- mice we performed RT-PCR on kidney RNA and demonstrated the absence of a PCR product in the two mice that were verified (Fig 1C, RT-PCR of the CaM KMT-/- with primers for GAPDH that used as control was positive, not shown). We further demonstrated the absence of production of CaM KMT protein in the CaM KMT-/- mice in kidney and liver (Fig 1D). All tissues of the CaM KMT-/- mice appeared normal by histologic examination at 21 days and 9 weeks.
Since the CaM KMT-/- mouse was constructed with the β-Galactosidase expressed from the initiating AUG of the CaM KMT, the expression of the β-Galactosidase enzyme can be detected by the color production produced with the substrate ONPG (ortho-Nitrophenyl-β-galactoside) [20]. β-Galactosidase activity in different tissues of twelve CaM KMT-/+, three CaM KMT-/- and one CaM KMT+/+ adult SWJ129 male mice was determined. The highest levels of expression were found in kidney and liver, followed by brain and testis (Fig 2A). Indeed, we succeeded in demonstrating the absence of CaM KMT protein in CaM KMT-/- by Western blotting but only in kidney and liver (Fig 1D, the full blot is presented in S3 Fig), probably because the level of expression of CaM KMT is too low to be detected in other tissues. The deletion does not reduce the transcription of the adjacent gene PREPL (S4 Fig). We have also verified whether the deletion affects cystine excretion in urine since patients with the 2P21 syndrome have cystinuria and CaM KMT is highly expressed in kidney. Analysis of cystine in the urine of two CaM KMT-/- C57Bl6J male adult (4 and 10 month old) mice in comparison to comparable CaM KMT+/+ indicated no cystine in the urine of the CaM KMT-/- (S5 Fig). Thus, the deletion of CaM KMT does not appear to affect the activity encoded by the SLC3A1 gene. We have previously demonstrated that in the cells from 2p21 deletion patients the loss of CaM KMT expression resulted in accumulation of hypomethylated CaM compared to normal controls, suggesting that CaM KMT is essential for CaM methylation and there are no compensatory mechanisms for CaM methylation in humans [11]. To evaluate if CaM KMT is essential for calmodulin methylation also in the various tissues of mice we performed an in vitro methylation assay using lysates from CaM KMT-/- and CaM KMT+/+ controls as a source for CaM as a substrate. The lysates were incubated with purified HsCaM KMT and [3H-methyl] AdoMet as the methyl donor. A protein of the molecular size of CaM was radioactively labeled in CaM KMT-/- lysates, while this labeling was absent in CaM KMT+/+ controls for all tissues tested except for skeletal muscle (and partly in liver) where CaM KMT+/+ controls also appeared to be hypomethylated, though to a lesser extent compared to CaM KMT-/- (Fig 2B). The amount of CaM in heart tissues is too low for detection even in the CaM KMT-/- mice. CaM was purified from quadriceps of CaM KMT+/+ adult females and analyzed by MS/MS for the presence of methylation. Depending on the experiment, both forms of CaM, methylated and not methylated were detected in the samples (S6 Fig).
The pattern of expression of CaM KMT was also verified in brain by β-Gal staining (Fig 3). It was found to be ubiquitously expressed and mostly homogenous in all brain tissue Fig 3B and 3D and 3E. Examples of the expression in the cerebral cortex, striatum and cerebellum are presented in magnified photographs (Fig 3F–3I). This pattern of distribution is in agreement with the expression of CaM KMT as shown in the Allen Brain Atlas project (http://www.brain-map.org/).
Since the patients of the 2P21 deletion syndrome demonstrated mitochondrial defect in muscle biopsies, including ragged red fibers [8] we verified possible mitochondrial functional deficiency by Gomori-Trichrome staining of the quadricep muscles of two CaM KMT-/- C57Bl6J male adult (8 month old) mice in comparison to comparable CaM KMT+/+. Muscle pathology was consistent with variation in fiber size (myopathic feature), also observed in patients with deletions of SLC3A1, PREPL and CaM KMT (17) and occasionally peripheral sub- sarcolemmal accumulation of mitochondria reminiscent of ragged red fibers (Fig 4).
Since the human P21 deletion syndrome also had reduced respiratory chain deficiency of all complexes except complex II [8] and the patients with SLC3A1, PREPL and CaM KMT showed reduced respiratory chain deficiency of complex IV [17] we further analyzed the enzymatic activity of the respiratory chain complexes in brains and muscles of the CaM KMT-/- C57Bl6J five adult (8 month old) mice in comparison to comparable CaM KMT+/+ and observed a marked decrease of both complex I and IV activities in the brain (Fig 5). This defect was only partially reflected in the muscle as complex I was unaltered and complex IV activity was moderately but not significantly decreased. The combined activities of II+III were decreased in both tissues. Complex II, which is solely nuclear encoded, was not affected in either tissue in agreement with the findings in the 2P21 deletion syndrome patients [8].
All developmental and behavioral analyses at all ages were performed on mice that were backcrossed for 10 generations into the C57Bl6J strain and comparisons done between siblings of heterozygous parents. The effect of CaM KMT on the development of sensory and motor reflexes was addressed during the first month of mice life. During this period mice were daily weighed as control for gross growth.
Reduced weight was obtained in CaM KMT-/- newborns of both sex compared to CaM KMT+/+, while the CaM KMT+/- mice gained weight similar to the CaM KMT+/+ mice. Male CaM KMT-/- mice weight was in average 4gr lower compared to CaM KMT+/+ (P<0.001) and 2.5 gr lower in CaM KMT-/- female, compared to CaM KMT+/+ female mice (P<0.003). These differences were preserved until the age of 80 days as indicated in Fig 6A and 6B. Muscle strength was tested by the time newborns were able to hold to a vertical wire as described in the methods section. Male mice of all groups performed similarly in the first couple of weeks, while at P17 the CaM KMT-/- mice were capable to hold to the wire for significantly shorter time compare to the CaM KMT+/+ and CaM KMT+/- groups. In female mice a similar effect was evident already at age of 7 days. In both sexes this difference was still evident at age of 21 days when the CaM KMT-/- mice could hang to the wire for 13sec less than CaM KMT+/+ (Fig 6C and 6D and S1 and S2 Tables). ANOVA for repeated measurements in female: F2,64 = 13.05, P<0.001 and male F2,48 = 5.6 p<0.001. Mice attraction to auditory stimulus was observed in control mice already at day 8–9 and became prominent in most CaM KMT +/+ mice by the age of 11–12 days. The CaM KMT-/- mice of both sex were delayed to acquiring this function compared to the CaM KMT+/+ (ANOVA, F2,81 = 10.24, P<0.001 and F2,100 = 7.56, p<0.01, in male and female, respectively), however by the age of 15 days in male and 14 days in female all mice fully reacted to the auditory stimulus, as shown in Fig 6E and 6F and in S1 and S2 Tables. Nest finding testing depends on mice sensory identification of nest location, their motor ability and motivation to return to the cage. Mice of the CaM KMT-/- show reduced scores in the nest finding test compared to CaM KMT+/+ and CaM KMT+/- groups during days 8–10, (ANOVA, F2,64 = 6.35, P<0.003; F2,48 = 4.2 p<0.01), in male and female, respectively, while at later ages all groups reached maximal scores (Fig 6G and 6H). A summary of developmental evaluation statistical values can be seen in S1 Table: Developmental profile test and S2 Table: Developmental profile test statistics.
Characterization of mouse behavior was performed only in males older than 3 months. Due to similar performance of the CaM KMT+/+ and CaM KMT+/- mice the results of these groups were included in one group. Male weight among the mice that survived to this age did not differ significantly between groups (Fig 7A). Tail length was also measured as an additional index for body growth and showed similar values between the CaM KMT+/+ and CaM KMT-/- groups (7.3 +/-0.1 cm and 7.1 +/- 0.1 cm, respectively; Student t-test, t = 1.03, p = 0.3). All mice were first tested in the open field, general exploration evaluated by the distance walked. Walking velocity and mobility showed no difference between genotypes. No significant differences between groups were observed when the time in the center of the arena was analyzed as an index for mice anxiety. The majority of CaM KMT-/- mice were not able to cling to a grid for the duration of the test and held significantly shorter time compared to the mice of the CaM KMT+/+ group where most of the mice griped to the grid 60 sec, the maximal time tested (49.2 ±17.6 and 29.5±25.5, in CaM KMT+/+ and CaM KMT-/-, respectively, Student t-test, t = 2.3, p<0.03) as shown in Fig 7B. Motor performance and coordination was addressed on the balance beam by the analysis of time to fall from the beam and the time required reaching one of the two escape boxes located at the end of the beam. Mice of both groups improved their performance in the repeated trials and stayed longer on the beam compared to previous trial (time to fall, F2,19 = 6.7, p<0.01 and F2,10 = 2.7, p<0.05; in CaM KMT +/+ and CaM KMT-/-, respectively, ANOVA for repeated measurements) (Fig 7C), however only the CaM KMT+/+ group improved the time to reach the box (time to box F2,22 = 7.9, p<0.01 and F2,7 = 2.7, p<0.1 in CaM KMT+/+ and CaM KMT-/-, respectively) as depicted in 7D. Higher ability of motor function and coordination is required to hold to an accelerating rotating rod, when rotation speed is gradually elevated. Shorter time on the rotating rod and lower speed were characteristic of CaM KMT-/- mice, compared to the CaM KMT+/+ mice performance (F1,35 = 9.5, p<0.01 and F1,35 = 9.4, p<0.01, for duration and speed, respectively). In addition, while the CaM KMT+/+ mice improved their performance during the 3 cycles of the test (F1,22 = 6.8, p<0.01 and F1,35 = 6.1, p<0.01, for duration and speed, respectively, ANOVA for repeated measurements), the CaM KMT-/- mice did not improve during the training and maintained similar scores through the test. Finally, avoidance memory was tested by the passive avoidance test. Mice were trained 10 trials per day in the passive avoidance apparatus to avoid stepping down from the platform. The duration until mice placed 2 limbs on the chamber floor was recorded as an index for learning. The score at the last trial of each training day is presented for each individual mouse in Fig 7G. Larger proportion of mice of the CaM KMT+/+ group improve their score within the training days and did not step down from the platform at the last trial of the test; the average time on the platform in the CaM KMT+/+ and CaM KMT+/- group was 44.3±56.2 sec in the first day and 101.1±42.6 in the last day of training, F1,13 = 6.9, p<0.01, while the mice of the CaM KMT-/- groups failed to acquire learning of the task; mice of this group acquired an average of 42.4±51.9 sec on the platform in the first day and 67.7±62.1 at the last day of training, F1,7 = 1.37, p<0.3 where about a half of the mice step down from the platform (Fig 7H). It should be noted that the performance at the first day of the test was similar in both groups, thus the differences obtained in the following days may be due differences in the learning process. In addition to exclude the possibility that the differences obtained in the passive avoidance test were due to genotype dependent differences in the somatosensory response mice were tested in the hot plate test. Similar response time in the hot plate test, characterized by 12.43 +/- 1.0 sec; 13.04+/-0.8 sec in the CaM KMT-/- and the CaM KMT+/+ mice was obtained. A summary of the statistical analysis of adult mice behavior is presented in S3 Table. Adult mice behavior statistics.
To verify whether the reduction in muscle strength is accompanied by changes in the muscle structures we performed Hematoxilin and Eosin staining on triceps brachii (fore limb) and quadriceps (hind limb) muscles from CaM KMT+/+ adult mice. No abnormalities were revealed.
Overall, developmental delay was observed in mice carrying the CaM KMT-/- genotype compared to CaM KMT+/+ mice, while the heterozygote mice development was comparable to that of the CaM KMT+/+ newborns. Although some of the criteria were acquired with age, impaired motor function, limited motor learning and failure to learn the avoidance of aversive stimulus characterized the CaM KMT-/- mice at adulthood. Heterozygote CaM KMT+/- behave similarly to the CaM KMT+/+ mice in all tested tasks, indicating that one copy of the gene is sufficient to support neurodevelopment.
Based on interest in the importance of methylation of CaM and the underlying causes of the 2P21 deletion syndrome, we have generated a mouse model lacking only the CaM KMT gene with the borders of the deletion as in the human syndrome.
Here we show for the first time that the protein CaM KMT is essential for normal body growth and somatosensory development as well as for muscle strength and the proper functioning of the adult mouse brain. Developmental delay was demonstrated for somatosensory function (attraction to sensory stimulus) and muscle strength (clinging on a wire) and for more complex behavior which involved both basal motor function and motivation (nest finding). We exclude the possibility that newborns did not find the nest due to impaired smell sensation since these mice were similar to the CaM KMT+/+ in the odor test. Developmental delay was observed at the young age for the tasks involving somatosensory processing, namely the sensory attraction and nest finding. In contrast, the task that involves mostly motor function, namely hanging on a wire, lower performance of the CaM KMT-/- relative to the CaM KMT+/+ persists to adulthood. At adulthood CaM KMT-/- perform similarly to the CaM KMT+/+ mice in low demanding tasks such as exploration in the open field and balanced beam. However, in tasks which require higher levels of motor and coordination function, namely rota-rod, the CaM KMT-/- did not improve their performance with training and performed below the CaM KMT+/+ mice. Notably, by repeated training on the rota-rod test CaM KMT+/+ improve their performance, as reflected both by the time they stay on the apparatus and the speed they manage to adjust to (Fig 7E and 7F). This change in time, which is indicative of motor learning was not seen in the CaM KMT-/- mice. The complex coordination required for this task involved the action of cerebellum and other components of the motor control system [21,22], but also involve the function and plasticity of the neuromuscular junction [23] and muscular strength. Transmission on the neuromuscular junction was not tested in the current study; however muscle myopathy was observed by the Gomori Trichome staining of in the striated muscle and may suggest that lower function in the rota-rod is of peripheral source.
The expression of CaM KMT β-gal obtained in cells of all brain structures suggests that this protein is required for the function of all cells in the central nervous system. If indeed required for the proper function of all neurons we would expect a gross behavioral phenotype. However it is possible that CaM methylation is essential for the execution of particular tasks, as presented in the results, for example the abundance of CaM KMT distribution in all cells of the cerebellum and low functioning in the rota-rod test suggests that CaM lysine modification is essential for complex motor coordination and motor learning. An additional aspect of learning, where the CaM KMT-/- mice failed to improve with training, was shown in the passive avoidance test involving learning of an aversive stimulus. It is possible that CaM methylation at one or more of the brain regions supporting this behavior, in particular the amygdale and cortico–hippocampal network, is essential for this type of learning [24] and memory consolidation [25,26].
In a recent evaluation of three older 2P21 syndrome patients, 20–30 years of age, it was found that, although able to walk and eat they are entirely dependent on their parents. They demonstrated severe mental incapacity and lack of speech despite normal hearing. This presentation of the patients, which could be more severe than that observed in the mouse model, may represent the effect of the additional missing gene the phosphatase PP2Cβ (PP2Cβ) for which no mouse model is yet reported.
The impairment of complexes I, III and IV of the mitochondrial respiratory chain in brain was anticipated based on the 2P21 deletion patients. Surprisingly the muscle complexes were less affected, indicating possible tissue variability. This could possibly be due to factors with overlapping functions present in muscle but not in brain. Notably complexes I, III and IV are dependent on mitochondrial DNA encoded subunits while complex II, the only one with unaffected activity is synthesized from nuclear DNA. Thus in accord with the previous finding in patients we hypothesize that CaM KMT could possibly be involved in the production of proteins encoded by mitochondrial DNA genes.
The present mouse model contributes to the evaluation of the role of methylated CaM, since in the tissues where CaM is methylated the elimination of the CaM KMT activity left the CaM non-methylated (Fig 2B), indicating there is no compensatory mechanism and CaM KMT is the only enzyme that methylates CaM. This is in agreement with our previous finding in lymphoblastoid cells derived from the 2P21 deletion patients [11]. The finding of hypomethylation of CaM in muscle was confirmed by mass spectrometry analysis performed on CaM extracted from the adult female control mice. Analyses by MS/MS lack the ability to be quantitative and consequently it wasn`t possible to estimate the amount of CaM that was not methylated with this experiment. One previous article indicated that 68% of skeletal muscle CaM is not methylated in adult male rats (it is also reported that 12% of liver CaM is not methylated) [27]. Heart CaM is also indicated to be undermethylated but we didn`t observe the same results in our samples. An older article [28] seems to have indication of undermethylated muscle CaM in rabbit. The hypomethylation of CaM in muscle and the major effect observed in muscle is presently not understood and requires further study. It is possible that the methylation may be needed at a specific window during muscle development and is not necessary at the time it was tested in our study, which was adult mice. Indeed the muscle phenotype is observed already in the development of the newborn mice with the reduced ability of the CaM KMT-/- mice to hold on the wire. An additional possibility is that the muscle phenotype could be caused by a compromise in brain function. Indeed, the tasks requiring brain involvement were affected. These findings can also point to the importance of CaM KMT in proper brain function. Methylation of CaM was reported in six regions tested: caudate nucleus, cerebral cortex, superior colliculus, brain stem, diencephalon and pituitary gland [29]. Thus, the impairment in intellectual ability observed in human with deletion of CaM KMT with additional genes [8,17,18] could be attributed to CaM KMT.
The present study enables to better assess the contribution CaM KMT to the clinical presentations of patients with different deletions involving it and additional adjacent genes homozygously (Table 1). All patients with the deletion of CaM KMT present severe growth retardation, severe neonatal hypotonia and moderate mental retardation. When tested, a mitochondrial deficiency was observed (12,18). All these clinical presentations can now be attributed to the absence of CaM KMT. The absence of the PREPL gene in the smallest deletion could contribute to the growth retardation and reduced muscle function as demonstrated by the mouse model deleted of PREPL (19). Thus, CaM KMT could present a new developmental delay gene, whether its effect could be all attributed through its effect on the mitochondria remains to be determined.
The current study has clinical implications for human patients. Patients presenting slow growth, muscle weakness that could result from a mitochondrial impairment and mental retardation could have a mutation in the CaM KMT gene, should be considered for sequence analysis of the CaM KMT gene.
The study was approved by the Institutional animal care and use committee of the Ben-Gurion University of the Negev. Ben-Gurion University of the Negev's animal care and use program is supervised and fully assured by the Israeli Ministry of Health; it is operated according to Israel's Animal Welfare Act 1994 and follows the Guide for Care and Use of Laboratory Animals (NRC 2011). In addition, BGU is assured by the Office of Laboratory Animal Welfare, USA (OWLA) #A5060-01, and fully accredited by the Association for Assessment and Accreditation of Laboratory Animal Care International (AAALAC). Mice were sacrificed by CO2.
The construction of the knockout construct was done in a stepwise manner. A β-gal-Neo cassette flanked by 50bp homology regions matching the boundaries of the desired deletion was constructed in pBR322 (S1A Fig). Briefly, a nuclear localization signal containing β- galactosidase gene (pNlacf, a gift from Richard Palmiter) was amplified using the following primers, Forward-CCGCGGGAGAGCCGTGGCGGACGCAACCTCCGGGTCCAGTCAGCTCTGGAGATGGGGCCCAAGAAGAA Reverse–CCGCGGCAGCATGCCTGCTATTGTCT, the forward primer contained a 50bp homologous region to the 5’ UTR of the CaMKMT. The neomycin resistance (pEGFP-N1 Clontech) gene was amplified using the following primers, Forward- ATAACTTCGTATAATGTATGCTATACGAAGTTATGCACTTTTCGGGGAAATGTG, Reverse–GGTGACCAGGCTGGGAGTATAACTTCGTATAGCATACATTATACGAAGTTATGACGCTCAGTGGAACGAAAA, both primers contained a 35bp LoxP sequence whereas the reverse primer also contained a 50bp sequence homologous to position 300–350 of the first intron of CaMKMT. The two PCR products were cloned to the pGEM-t easy vector (T-A cloning, Promega) and further subcloned into pBR322 (S1B Fig). The above construct was then used to modify a BAC containing the CaMKMT region (S1B Fig) of the mouse strain 129S7/AB2.2 (a gift from the Sanger Institute, the AB2.2 BAC library: bMQ-132H2). The modification was performed using the recombineering method that uses phage λ proteins that facilitate homologous recombination in E.coli [30]. Briefly, the BAC was transformed into DY380 bacterial strain (kindly provided by Dr Neal Copeland, NCI-Frederic, MD, USA); resistant colonies were further transformed with the β-gal-Neo cassette (excised from pBR322 by restriction digestion). Resistant colonies were analyzed by PCR for proper integration at the 5’ and 3’ sites, for each site one primer was within the cassette and the other in the homologous region in the BAC. Primers for 5': TGGGCTTTTTCTCTTCCTGA, GTTTTCCCAGTCACGACGTT (primers A and B, S1C Fig); Primers for 3': CGTGAGTTTTCGTTCCACTG, GAGCGGTTTGGAAGACTGAA (primers C and D, S1C Fig). An example of PCR products matching correct integration can be seen in S1D Fig. To generate the knockout construct that will be transfected into ES cells, the entire pBR322 was amplified by PCR using two primers which had at the 5’ and 3’ ends a 50bp sequence homologous to regions 8120bp upstream and 3446bp downstream to the β-gal gene respectively. Transformation of DY380 containing the modified BAC with the pBR322 based PCR product generated a pBR322 plasmid that contained a 8120bp homology arm upstream to the β-gal gene and 3446bp homology arm downstream to the β-gal gene. The Knockout cassette including the homology arms was cut from the pBR322 backbone by digestion with SgsI and transfected into ES cells from mouse strain W4/129SvEv at the University of Iowa Knockout core. Similarly to the primer set used to verify correct integration into the BAC, the primer sets for verification of homologous recombination in ES cells included one within the cassette and the other outside of the homology region used for recombination. Primers for the 5' integration site: GTCTTTGGAACGAACCAAGCAGCA, TGTGCTGCAAGGCGATTAAGTTGG. Primers for the 3' integration site: CAATACGCCCGCGTTTCTTCCTTT, TAATGCCAGTTCTTGGGAGGCAGA. This would create PCR products of 3878bp for the 3' side and 8857 bp for the 5' side. Out of 192 clones resistant to G418, three showed homologous recombination (S2 Fig).
PCR genotyping of the mice was done using primers: Forward: 5' CCTGACAGGGAAGAAGTTGG 3' and Reverse: 5' CGCCTCTGCCTCAGTCTCT 3' to detect the wild type genotype and primers Forward: 5' GTGCACGGCAGATACACTTG 3' Reverse: 5' GATGGCTGGTTTCCATCAGT 3' to detect the knock out sequence. PCR conditions were:30 cycles at 94° 40sec., 62° 45 sec. and 72° 1 min. followed by 10 min at 72°, using DreamTaq (ThermoScientific)
RNA was extracted from kidney using DirectZol (ZymoResearch, USA). cDNA was produced from 3 microgram RNA with Superscript II (InVitrogene). The quality of the resulting cDNA has been tested by the amplification of GAPDH with the primers: forward-5′AAAACGTCCATGTTCCCATC3′; reverse-5′CCCCAGACACGATAAGCAGT3. The primers for PCR of CaM KMT were from the first exon: 5' TGCAGAGACTGAGGCAGAGG 3' and from the second exon: 5' TGTGGCTTCTGTGACTGAGAAC 3'. PCR was performed on 1 microliter of cDNA, conditions were:40 cycles at 94° 40sec., 63° 45 sec. and 72° 1 min. followed by 10 min at 72°, using DreamTaq (ThermoSCIENTIFIC).
Western analyses were performed as detailed in [11] using the polyclonal CaM KMT antibody described [11]. β-Gal activity was detected by ONPG analysis (Sigma) according to manufacturer instructions and [20].
Was done by Bactochem Ltd. Israel on 2 mice at each 21 days and 9 weeks that included: kidney, pancreas, testis, intestine, lung, brain, heart, thymus, striated muscle, salivary gland, and barderian gland, lymph node. The Hematoxilin Eosin staining on the adult mice was done by the standard procedure.
Analysis of free amino acids including Cystine in urine samples of mice was done according to Moore, S., and Stein, W. H. (1948) [31]. 100μl of the urine sample were transferred into an Eppendorf vial and 250μl of Acetonitrile were added into this vial and the solution was vortex. After centrifuge for 5 minutes at 5000rpm 300μl of the upper phase were transferred into another vial and evaporated to dryness, by a concentrating centrifuge. The dry extract was dissolved by 250μl amino acid sample buffer, vortex and filtered by using 0.45μm nylon filter. 25μl were injected into the Amino-Acid-Analyzer {AAA- Knauer model A200, Column & packing: High resolution Lithium column for AAA, 80 x 4.6 mm, Temp. 30–60°C; Mobile phase: Lithium buffers for physiological amino acid analysis (With variable pH and lithium concentrations). One of the samples was spiked with Cystine standard and the preparation and analysis was done as described above. The Cystine and other amino acids were separated on an ion exchange column and derivatized with Ninhydrin after their elution from the column (Post column derivatization). The Cystine and the amino acids were detected at 570nm and 440nm. The Cystine and the amino acids were identified and quantified against standard that was injected into the AAA. The concentration of each amino acid in the sample (mg/L) was calculated using an Excel program.
Mice were perfused with cold PBS pH 7.4, brains were excised and sliced at 0.3mm intervals using a vibrotome (Leica VT1200), Brains were washed in PBS for 5 min. and fixed with cold 1.5% formaldehyde, 0.2% glutaraldehyde, 5mM EGTA and 2mM MgCl2 in PBS for 1 hr. Two washes were done at 4 ° followed by an overnight wash at 37° with wash buffer: 2 mM MgCl2, 0.01% deoxycholate, 0.02% NP40 in PBS, before transfer to X-Gal staining buffer at 37° for 10hrs. The slices were washed 30 min. at 37° and twice again for 30 min at room temperature. The slices were embedded in glycergel containing half volume of glycerine, 5% gelatin and 1% phenol in water and kept at 4 ° until visualized.
Were done as detailed in [7] and [11]. Tissues and organs excised from WT or KO adult female mice were extracted in buffer 50mM Tris pH = 7.5, 150mM KCl, 2mM MgCl2, 2.5mM MnCl2, 0.01% Triton X-100, 2mM CaCl2 and 2mM TCEP. Protein content in the lysates was measured using the Bradford protein assay. One hundred ug of total proteins were incubated for 2h at 37C in presence of 5ug HsCaM KMT and 3H-methyl Adomet in 50ul reaction containing 50mM Tris pH = 7.5, 150mM KCl, 2mM MgCl2, 2.5mM MnCl2, 0.01% Triton X-100, 2mM CaCl2 and 2mM TCEP. Samples were precipitated in 10%TCA and resuspended in 30ul blue loading dye. Samples were run a 12.5% SDS-PAGE gels that were then treated with Enhance (Perkin Elemer), dried and exposed to X-ray film. Control samples contained only 0.67ug of CaM and CaM KMT.
Lysates from muscle tissues (both WT and KO) were loaded on a 100μl phenylsepharose 6FF column equilibrated in the presence of 5mM CaCl2. After extensive washing CaM was eluted by addition of 50mM Tris pH = 7.5 and 10mM EGTA. The whole elution was TCA precipitated, loaded on SDS-PAGE gel and the band excised and submitted for MS/MS analysis.
The enzymatic activities of respiratory chain complexes in isolated mitochondria were measured at 37°C by standard spectrophotometric methods as described [32,33]. Briefly Complex I was measured as rotenone sensitive NADH-CoQ reductase monitoring the oxidation of NADH at 340nm in the presence of coenzyme Q1. Complex I+III was measured as rotenonene sensitive NADH- cytochrome c reductase at 550 nm. Complex II was measured as succinate dehydrogenase (SDH) based on the succinate mediated phenazine methosulfate reduction of dichloroindophenol at 600nm. Complex II+III was measured as succinate cytochrome c reductase and following the reduction of oxidized cytochrome c at 550 nm. Complex IV (COX, cytochrome c oxidase) was measured by following the oxidation of reduced cytochrome c at 550nm.
Citrate synthase (CS), an ubiquitous mitochondrial matrix enzyme, serving as a control, was measured in the presence of acetylCoA and oxaloacetate by monitoring the liberation of CoASH coupled to 5',5'-dithiobis (2-nitrobenzoic) acid at 412nm. Activities were calculated as μmol/min/mg protein (U/mg) and presented ratio (U/U) to CS.
Behavioral analyzes: Newborn behavior was examined as follows: the mother was removed from the home cage every day during postnatal days 4–21 to examine the newborns in the cage. Newborn mice were tested in sequential order. Each newborn was tested, marked gently by non-toxic paint, wiped lightly with nesting material with maternal scent, and returned immediately to the nest.
At age 3 months, mice were tested in a battery of behavioral tests addressing autistic like behavior in the adult mice. A week before the experiments began; the mice were separated and placed in individual cages to avoid the effect of social hierarchy on mice behavior. During that week, the mice were handled daily by the experimenter for 2 min to adapt the mice to the presence of the experimenter. All experiments were videotaped and analyzed offline using “EthoVision” software (Noldus, Netherlands). All mice were tested daily between 16:00 and 20:00. In all the behavioral tests, males and females were tested in separate sessions. The testing arena was cleaned between trials with 70% ethanol.
Statistical analyses were performed using SPSS 18.0 software. Univariate general linear model (GLM) analysis was used to test the effects of genotype. ANOVA for repeated measurements was used for variables that were repeatedly measured at different time points. In all tests equal variance of the data not assumed therefore Dunnett T3 Post-hoc was used for multi comparisons. 2-tails student T-test was used when 2 groups were compared. In all tests differences with a p-value < 0.05 were regarded as significant. Results are presented as the mean ± SD.
|
10.1371/journal.pcbi.1005572 | Persisting fetal clonotypes influence the structure and overlap of adult human T cell receptor repertoires | The diversity of T-cell receptors recognizing foreign pathogens is generated through a highly stochastic recombination process, making the independent production of the same sequence rare. Yet unrelated individuals do share receptors, which together constitute a “public” repertoire of abundant clonotypes. The TCR repertoire is initially formed prenatally, when the enzyme inserting random nucleotides is downregulated, producing a limited diversity subset. By statistically analyzing deep sequencing T-cell repertoire data from twins, unrelated individuals of various ages, and cord blood, we show that T-cell clones generated before birth persist and maintain high abundances in adult organisms for decades, slowly decaying with age. Our results suggest that large, low-diversity public clones are created during pre-natal life, and survive over long periods, providing the basis of the public repertoire.
| The enormous diversity of T-cell receptor (TCR) molecules allows our adaptive immune system to recognize and fight infections. TCRs are formed through the stochastic rearrangement of DNA. By analysing human repertoire sequences of identical twins using a statistical model for TCR formation, we identified T-cells that were exchanged between twin embryos during pregnancy. We exploited the slightly different recombination statistics between fetal clonotypes and mature ones to track their relative fractions in adult T-cell repertoires of different ages. We showed that the decay of fetal clonotypes with age is extremely slow, spanning several decades. Our findings suggest that an important part of our adaptive immune system is formed before birth.
| The adaptive immune system relies on the diversity of T-cell repertoires to protect us from many possible pathogenic threats. Each T-cell expresses on its surface many copies of a unique T-cell receptor (TCR), which engages with antigenic peptides—from self or foreign proteins—presented by other cells through their Major Histocompatibility Complex (MHC) molecules. The binding strength between the TCR and the peptide-MHC complex, which is typically weak for self peptides, and strong for some foreign peptides, is a major factor in determining the onset of an immune response. Since each TCR is only specific to a small fractions of the possible peptides, the body needs to maintain a very large diversity of TCRs to be able to recognize any possible foreign peptide from pathogens. Understanding how this diversity is generated, and how it develops and matures with age, is thus paramount to understanding adaptive immunity.
TCR diversity is produced by the V(D)J recombination machinery which generates the repertoire de novo in each individual. Repertoire diversity is encoded not only in the set of specific receptors expressed in a given individual, but also in their relative abundances—the number of T-cells expressing each unique TCR—which can differ by orders of magnitude. These differences are in part due to antigenic stimulation (infection, vaccination), implying that clones increase their sizes in response to common or recurring infections. Despite this great diversity, different individuals—regardless of their degree of relatedness—do express a subset of the exact same receptors, called the public repertoire [1]. This overlap is often interpreted as the convergence of individual repertoire evolutions in response to common antigenic challenges [2]. Indeed, some public TCRs are known to recognize common pathogens such as the cytomegalovirus (CMV) or the Epstein-Barr virus (EBV) [3]. However, this interpretation is challenged by the fact that these two properties—large differences in clone sizes and public repertoires—are also observed in naive repertoires, for which antigenic stimulation is not expected to be important [4, 5].
An alternative explanation for public clones, which does not invoke convergent repertoire evolution, is that both abundant and public receptors are more likely to be produced by rearrangement, and just occur by coincidence [1, 6]. This idea is backed by some compelling evidence. First, the amount of clonotype sharing between pairs of individuals can be accurately predicted in both naive and memory pools from statistical models of sequence generation [7]. Second, the likelihood that a clonotype sequence is shared by individuals has been reported to correlate with its abundance [6, 8]. However the origin of this correlation remains elusive. In addition, public clonotypes often have few or no randomly inserted N nucleotides, which limits their diversity [6]. Terminal deoxynucleotidyl transferase (TdT), the enzyme responsible for N insertions, is inactive in invariant T-cell subsets [9] and in some fetal T-cell clones. These subsets could contribute to the emergence of the public repertoire. Another confounding factor is the ageing of repertoires, and the concomitant loss of diversity, which is expected to affect the structure of clonal abundances as well as the repertoire’s sharing properties. How do all these effects shape the structure and diversity of TCR repertoires, and control their functional capabilities? Here we propose and test the hypothesis that a sizeable fraction of public clonotypes are created before birth. These clonotypes have low diversity because of reduced TdT activity, making them more likely to be shared among unrelated invididuals. Their large abundances, due to reduced homeostatic pressures in the early stages of repertoire development, allow them to survive over long periods.
We first examined in detail the question of clonotype sharing between individuals. Each TCR is a heterodimer made of two chains encoded by two distinct genes. Each gene is formed in the thymus by assembling together two or three gene templates from a finite set of germline segments—V and J segments for the α chain, and V, D and J segments for the β chain. In addition to the large diversity created by the combinatorial choice of germline segments, further diversity is produced by randomly deleting base pairs from the joining ends of the segments, and by inserting random non-templated (N) base-pairs at each junction. Each chain forms three loops, called Complementarity Determining Regions (CDR), which come in contact with the peptide-MHC complex during recognition. The first two loops, CDR1 and CDR2, are encoded in the germline V gene and are thought to interact mostly with the MHC. By contrast, the CDR3 concentrates most of the diversity, as it covers the junctions between the germline segments. The CDR3 interacts with the peptide directly, and is thus believed to play the biggest role in the recognition of foreign peptides.
After recombination, receptors are tested and selected for function and lack of auto-reactivity. The recombination mechanism frequently produces non-functional (also called nonproductive) receptor sequences, typically containing frameshifts or stop-codons. If the recombination result of the first chromosome is nonproductive, the second chromosome will recombine. In case this second recombination is successful, the cell will contain two recombined genes—one productive and one nonproductive. To avoid confounding effects due to convergent selection (both selection in thymus and clonal expansion in response to infection), we first focused on out-of-frame receptor sequences, which are nonproductive and hence must result from these first unsuccessful recombination events. Because the cells that contain them owe their selection and survival to the productive gene on the second chromosome, these out-of-frame sequences give us direct insight into the raw V(D)J recombination process [10, 11], free of clonal selection effects. The number of shared clonotypes between two sets of clonotypes, or clonesets, is approximately proportional to the product of the cloneset sizes [8, 11, 12]. We call the ratio of the two the normalized sharing number. In the regime of rare convergent recombination, this number is equal to the probability that two independent recombination events give the same sequence; it is thus independent of the cloneset sizes, and provides an appropriate measure of sharing for comparing different pairs of datasets with different sequencing depths. Under the assumption that sharing occurs by pure chance, only due to convergent recombination, this number can be predicted using data-driven generative probabilistic models of V(D)J recombination accounting for the frequencies of the assembled V, D, and J gene segments and the probabilities of insertions and deletions between them [7, 11, 13, 14]. We can estimate sharing either of the entire nucleotide chain (alpha or beta), or of the CDR3.
Genetically identical individuals may be expected to have more similar recombination statistics due to similar recombination enzyme biases [8, 15–19], and therefore share more sequences. To assess these genetic effects, we looked at the sharing of TCR alpha and beta-chain receptor repertoires between three pairs of monozygous twins (6 individuals). We synthesized cDNA libraries of TCR alpha and beta chains from the donors’ peripheral blood mononuclear cells and sequenced them on the Illumina HiSeq platform (see S1 Fig and S1 Text). For each pair of individuals, the normalized number of shared out-of-frame alpha sequences was compared to the prediction from the recombination model trained on the out-of-frame repertoire of each individual, as shown in Fig 1 (see also S2 Fig for similar results on sharing of CDR3 nucleotide sequences). Sharing in unrelated individuals (the 12 non-twin pairs among 6 individuals, black circles) was well predicted by the model (Pearson’s R = 0.976), up to a constant multiplicative factor of 2.07, probably due to differences in effective cloneset sizes. While twins did share more sequences than unrelated individuals (the 3 twin pairs, red circles), this excess could not be explained by their recombination process being more similar. The model prediction was obtained by generating nucleotide sequences from models inferred using each individual’s cloneset as input [13, 14], mirroring their specific recombination statistics (see S1 Text). The normalized sharing number departed significantly from the model prediction only in twins, calling for another explanation than coincidence in that case. The same result was obtained for beta out-of-frame CDR3 nucleotide sequences (S3 Fig), although less markedly because of a lower signal-to-noise ratio due to smaller numbers of shared sequences. Most of beta out-of-frame nucleotide sequences shared among the highest-sharing twin pair associated with CD8 CD45RO+ (memory) phenotype in both individuals. This observation is surprising, because the non-functionality of these sequences excludes convergent selection as an explanation for it (see S1 Text for details).
We then examined the sharing of in-frame nucleotide CDR3 sequences. Most of in-frame sequences are functional, and have passed thymic and peripheral selection. Since these selection steps involve genetically-encoded HLA types (the type of MHC that cells express) and are therefore expected to be similar in related individuals, we wondered whether the functional repertoires of twins also displayed excess sharing. Remarkably, we found some excess sharing in the in-frame beta repertoire (S4 Fig), but none in the in-frame alpha repertoire (S5 Fig). However, the failure to observe excess sharing in this last case can be explained by the much higher expected number of shared nucleotide sequences in the alpha in-frame repertoire (due to both in-frame sequences being more numerous than out-of-frame ones, and to the lower diversity of alpha chains compared to beta chains) which could mask this excess in twins (see S1 Text).
To investigate the origin of excess sharing between twins, we looked at the statistical properties of shared alpha out-of-frame nucleotide sequences from Fig 1. Shared clonotypes between non-twins, which happen by coincidence, should have a higher probability Pgen to have been produced by V(D)J rearrangement compared to non-shared clonotypes. Indeed, the distribution of Pgen among shared sequences, plotted in Fig 2, can be calculated from the probabilistic model of generation (blue curve), and the prediction agrees very well with the data between non-twins (red curves). By contrast, shared sequences between twins deviate from the prediction (green curve), especially in the tail of low-probability sequences, but are consistent with a mixture of 18 ± 3% of regular sequences (black curve), and the rest of coincidentally shared sequences (blue curve). These numbers agree well with the excess sharing in twins, which amounts to 17% ± 3% of non-coincidentally shared sequences, as estimated from Fig 1. Nucleotide sequences shared between twins also have higher numbers of insertions and are therefore longer than those shared between unrelated individuals or according to the model (S6 Fig, p = 2 ⋅ 10−8, two-sided t-test)—a trend that is even more pronounced in memory cells (S7 Fig, p < 10−16). Note these observations about recombination probabilities and the number of insertions are related: sequences with many insertions each have a low generation probability because of the multiplicity of inserted nucleotides.
Taken together, these observations support the existence of another source of shared sequences than coincidence in twins. Since the sharing of cord blood between twins is the only natural instance when the immune systems of two individuals share cells, we propose that the increased sharing of private TCRs between identical twins dates back to the sharing of cord blood cells, and that these shared clones persist into late age. This persistence of fetal clonotypes could be due to the long lifetime of the exchanged naive clones. Alternatively, long persistence could be achieved by the independent transition to memory of the shared clones in both twins.
To verify the hypothesis that clones formed during fetal life persist over long periods, we now turn to the analysis of data from unrelated individuals. We characterized the in-frame beta-chain repertoire of human cord blood and also three healthy non-twin adult donors of different ages (see Materials and methods and S1 Text). One feature of the rearranged chains is the number of insertions at the junctions between the gene segments (VD and DJ in the case of beta chains). We ranked beta TCR clonotypes from human cord blood data by decreasing abundances and plotted the mean number of insertions (inferred iteratively and averaged over groups of 3000 clonotypes, see S1 Text), as a function of this abundance rank (Fig 3A). The most abundant clones in cord blood had markedly smaller numbers of insertions (black line). The naive repertoire of a young adult (blue line) showed a much weaker dependence on abundance than the cord blood repertoire, but followed a similar trend. The dependence was even further reduced in older adults (purple and green lines). Interestingly, the number of insertions in the beta chains of the adult memory repertoire (red, orange and maroon lines) did not depend of the abundance of these cells. This observation can be explained by the resetting of the size of memory clones following an infection, erasing features of the abundance distribution inherited from fetal life. Looking more closely into the distribution of the number of insertions (Fig 3B) reveals that low mean numbers of insertions are associated with an enrichment in clonotypes with zero insertions. Accordingly, the fraction of naive zero-insertion sequences generally decreased with abundance rank (Fig 3C), with again a stronger dependency in cord blood and young adults. Fewer numbers of insertions in the cord blood are expected because TdT, the enzyme responsible for random insertions, is initially strongly downregulated in prenatal development [20, 21]. This enrichment in low-insertion sequences persists and shows weak signatures in the adult naive repertoire, suggesting long lifetimes of cord blood clonotypes (although not necessarily of individual cells).
The enrichment of zero-insertion sequences in large clonotypes of young people, relative to the baseline of zero-insertion clonotypes produced in adulthood, can be used to verify the hypothesis of long lived fetal clonotypes originating from the cord blood. Analysing publicly available TCR beta repertoire data from individuals of different ages [23, 24], we observed a slow decay of abundant zero-insertion clonotypes in the unpartitioned repertoire (memory plus naive) with age, with decay rate of 0.027 ± 0.009 yr−1, or a characteristic time of 37 years (Fig 4). However, the excess of abundant TdT- clonotypes of fetal origin only affects naive cells (Fig 3A), whose relative fraction in the repertoire is also known to decrease with time [23]. To assess the importance of this confounding effect, we fit an exponential decay model for the percentage of naive cells measured in same donors using flow cytometry (see S3 Table) and found a characteristic decay rate of 0.015 ± 0.002 yr−1, or a decay time of 67 years. The red curve in Fig 4, which shows the expected decay of zero-insertion clonotypes if it had been solely caused by the decay of the naive pool, does not agree with the data. Although the decay of naive cells within the top 2000 clonotypes could in principle be faster than in the overall T-cell population, we did not observed such an effect in the three individuals for which we have data partitioned into memory and naive clonotypes (see S1 Text I.G). Therefore, the attrition of the naive pool alone cannot explain the decrease of zero-insertion clonotypes, which we attribute instead to the progressive extinction of clones of fetal origin combined with their gradual replacement by newly generated naive cells. This is consistent with the hypothesis that excess clonotype sharing between twins is enabled by long-lived naive cells, but does not exclude the possibility that this excess sharing can be supported by memory cells as well.
We have shown that abundant clones are enriched with zero-insertion sequences, both in the cord blood and in the adult naive repertoire. Zero-insertion clonotypes (regardless of their origin) are most likely to be shared by convergent recombination than regular sequences, because they are more likely to be generated due to reduced diversity. What are the implications of this observation for sharing between unrelated individuals? Since zero-insertion sequences are overrepresented among abundant clonotypes (Fig 3), we predict that abundant out-of-frame clones are more likely to be shared.
To make our prediction quantitative, we built a mixture model of the out-of-frame alpha repertoire (see S1 Text for details). We assumed that clonotypes of a given abundance C are made up of a certain fraction F(C) of TdT-, zero-insertion clonotypes, and a complementary fraction 1 − F(C) of regular TdT+ clonotypes. Because TdT+ clonotypes may also have no insertions, the fraction of the TdT+ and TdT- sets had to be learned in a self-consistent manner. To learn these fractions, for each abundance class C we directly quantified the fraction F0(C) of sequences in the data that are consistent with zero insertions (i.e. can be entirely matched to the germline segments). Because non-templated nucleotides can coincide with the template, and also because TdT+ cells may have no insertions, F0(C) is not equal to F(C). However they are linearly related, so that it is enough for a model to agree with the data in terms of F0(C) to also guarantee agreement in terms of F(C). We generated a large number of nucleotide alpha out-of-frame sequences using our recombination model, and separated them into two groups: those that are consistent with no insertions (group A), and the others (group B). For each abundance class C, we created articifical datasets made of a fraction F0(C) of sequences from group A, and a fraction 1 − F0(C) from group B, where we recall that F0(C) is estimated from the data. We then repeated the sharing analysis in these artificial datasets in the same way as in the real datasets. The model accurately predicts the normalized sharing number of out-of-frame alpha-chain CDR3s as a function of clonotype abundance (Fig 5), up to the common multiplicative factor of 1.7 by which the non-mixture model generally underestimates CDR3 sharing (see S2 Fig). Thus, the enhanced sharing of high-abundance clonotypes is entirely attributable to their higher propensity to have no insertions, making them more likely to be shared by chance.
We found that adult twins present an interesting case of microchimerism in the adaptive immune system: shared rare TCR variants that recombined before birth survive for decades in their repertoires. We have also shown that adult naive repertoires, but not memory repertoires, have similar zero-insertion TCR clones distributions as cord blood repertoires. With age, the clone size distribution of naive adult repertoire becomes more similar to that of the memory repertoire. We hypothesize that this similarity between adult naive and cord blood repertoires is maintained by long lived fetal clones. Our results on the biological trafficking of T cells in twins are robust to possible experimental artefacts. First, our framework relies on the accurate counting of TCR cDNA sequences using unique molecular identifiers [25]. To exclude the possibility of contamination during the PCR and sequencing process, we double barcoded each cDNA library. To further exclude the possibility of early contamination of the blood samples, we performed replicate experiments at different times using different library preparation protocols. Comparison of repertoire overlaps from such replicate experiments for the same set of twins shows no difference and rules out experimental contamination as a confounding effect (see S1 Text). We also observed the same effects in previously and independently collected datasets [8], further excluding the possibility of experimental artefacts (S8 Fig). This reproducibility also suggests that the majority of out-of-frame sequences are not sequencing errors. Additional evidence for this fact comes from the different fractions of out-of-frame sequences observed in alpha and beta chains in TCR cDNA sequencing data, 13 and 3 percents respectively [8]– both of these fractions are much higher than the indel rate for the illumina platform [26, 27]. Our conclusions rely on a variety of data sources, and make extensive use of statistical analysis. As it is not yet possible to collect data from the same donors over many years, statistical evidence such as the amount of sharing in twins, or the amount of zero-insertion clonotypes versus age, is needed to investigate the evolution of repertoires over decades.
Cord blood sharing between twin embryos could have important implications on twin immunity: they could share and respond with private clonotypes, which would otherwise not be likely to be produced independently. This could possibly include sharing of malignant [28–30] or autoimmune clones, leading to disease in both individuals. In very rare cases such transfusion could also occur between dizigotic twins, leading to chimerism [31]. Anastomoses between monochorionic twin placentas are very common (more than 85 percent of uncomplicated pregnancies [32]), however the amount of exchanged blood may vary, and in some extreme cases it even leads to adverse outcomes such as twin-to-twin transfusion syndrome [33]. These effects could possibly affect the initial number of in utero shared clonotypes. This mechanism of sequence sharing is very different from sharing by convergent recombination [6], because it also implies the sharing of the second TCR chain and of the cell phenotype. Paired repertoires studies, which combine alpha and beta chains together [34, 35], could be used to track clones shared between twins more precisely, and distinguish them from convergently recombined ones.
Our results suggest two mechanisms with opposite effects on the sharing of clonotypes in twins as a function of the number of insertions. On the one hand, we have argued in Figs 1 and 2 that clonotypes shared through direct cell exchange should have a ‘normal’ number of insertions, because they are not due to random convergent recombination (which favors low numbers of insertions). On the other hand, we have shown in Fig 3 that cord blood cells are enriched in zero-insertion clonotypes, suggesting that clones shared in utero should be enriched in clonotypes with no of few insertions. Which one of these two effects dominate? TdT is suppressed in human embryos mostly in the first trimester of pregnancy [21]. Since TdT is active in the later trimesters the majority of the cord blood repertoire consists of clones with non-zero insertion numbers [22] similarly to the regular TdT+ post-natal clones. We show that the insertion distribution for non-abundant clones in cord blood closely resembles the insertion distribution observed in adults, with most clonotypes having insertions (see Fig 3B II). Such clonotypes could be exchanged in utero between twins, and easily identified as shared clonotypes with low Pgen. Our theory predicts that twins should also exchange zero-insertion clonotypes, which are abundant in cord blood. However these shared clonotypes are indistinguishable from clonotypes shared by convergent recombination, which are also likely to have zero insertions. Therefore, the higher abundance of zero-insertion clonotypes in cord blood relative to mature repertoires does not contradict the observed sharing of high-insertion clonotypes due to cord blood exchange.
We have also showed that some of the clonotypes transferred in utero have the CD45RO+ phenotype, typical of central memory cells. It is possible that the longevity of these clones is connected with their memory status acquired early in life. To test this hypothesis, one would need to perform deep sequencing of purely sorted naive T-cells from adult twins and repeat the analysis presented in this paper. The transition from naïve to memory is also associated with clonal expansion, so it is possible that, within the in utero transfer hypothesis, the most easily detectable clonotypes shared between twins come from the memory population simply due to sampling effects. At the same time, the results plotted in Fig 3 suggest that naïve clonotypes may also be long lived. Thus, clonotypes transferred in utero in twins could be either of naive or memory origin.
Our conclusion that fetal clonotypes are long-lived is based on the analysis of over-abundant zero-insertion clonotypes. Invariant T-cells, MAIT (Mucosal-Associated Invariant T-cells) and iNKT (Invariant Natural Killer T-cells) are intrinsically insertion-less, have restricted VJ usage for alpha chain, and are often abundant. These cells are produced in adulthood and could in principle constitute a substantial fraction of our zero-insertion dataset, confounding our analysis. Since our abundant zero-insertion clonotypes have a very diverse usage of VJ genes, we can exclude that the majority of them are from invariant T-cells, although we did identify a small number of such invariant TCR alpha chain clonotypes, see S1 Text. An alternative explanation of the skewed zero-insertion clone size distribution of naive repertoires (see Fig 3A) is the existence of previously unknown subset of insertionless T-cells characterized by large proliferation activity, which would be produced in adulthood and make up the most abundant clones of the naive repertoire. To support this hypothesis, one would need to further assume that the production of these cells decays with age, to be consistent with the observations of Fig 4. Another related possibility is that insertionless clonotypes are generally favored by thymic selection, again in a age-dependent manner. However, in-frame clonotypes have been reported to be only moderately enriched (by less than 20%) in zero-insertion sequences relative to out-of-frame sequences (see Ref. [7], Fig 3E and 3F), meaning that thymic selection does not substantially favor zero-insertion clonotypes on average.
Our current data clearly shows that clonotypes that originated in the cord blood tend to be among the most abundant in the naive repertoire, but we cannot unambiguously point to the source of this effect. One possibility is convergent recombination [6, 36]: high clonotypes abundances could be due to the accumulation of multiple convergent recombination events made more likely by the limited recombination diversity during fetal development. However, we observed clonotypes with low generative probabilities among the most abundant clones in the cord blood repertoire, and also clonotypes with high generation probability among the least abundant clones. We conclude that convergent recombination alone could not predict cord blood clone frequencies. An alternative explanation is that these clones have had more time to expand than others. Fetal cells come from different precursors, and mature in a different environment (the fetal liver), than post-natal cells [37]. In vitro experiments have shown that fetal T-cells have a different proliferation potential than post-fetal cells [38]. Additionally, a vacant ecological niche effect may play a role. When these clones first appeared, the repertoire had not reached its carrying capacity set by homeostatic regulation, leaving room for future expansion. These clones may have initially filled the repertoire, later to be gradually replaced by post-fetal clonotypes. Consequently, fetal clones, including those whose TCR was recombined with no TdT, would be expected to have larger sizes. Quantitative TCR repertoire profiling (preferably with the use of unique molecular identifiers for accurate data normalization and error correction), performed for species with no TdT activity in the embryo, such as mice, as well as novel cell lineage tracking techniques [39] could be used to investigate the detailed dynamics of fetal clones. This large initial expansion of fetal clones could protect them from later extinction. This would suggest that the estimated 37-year lifetime of zero-insertion fetal clonotypes could be longer than that of regular clones produced after birth.
Sharing of beta TCRs has previously been shown to decrease with age [23]. Depletion of fetal clonotypes, which are more likely to be shared, could contribute to this phenomenon. Our results also predict that the excess sharing of clonotypes between twins due to the trafficking of fetal cells should decrease with age. In general, the observed abundance of large zero-insertion clonotypes and their persistence through significant part of our life should have important consequences for the adaptive immunity regulation both in pre- and post-fetal period. Interestingly, transgenic mice with induced fetal TdT expression showed impaired antibody response to certain bacterial pathogens, suggesting an important functional role of the low-diversity fetal repertoire in immune competence [20]. We could speculate that the primary target of these cells might be common pathogens with a long history of coevolution with humans, such as CMV and EBV.
Lastly, our general framework for analyzing the overlap between different repertoires has far-reaching practical implications for the tracking of T-cell clonotypes in the clinic. In particular, the analysis of overlap between pre- and post-treatment repertoires using probabilistic characteristics of clonotypes sharing could help determine the host or donor origin of clonotypes after hematopoietic stem cell transplantation (HSCT), and also increase reliability of malignant clones identification in minimal residual disease follow-up.
For a more detailed description of experimental and data analysis procedures see S1 Text Materials and Methods.
RNA was isolated from the PBMC of healthy Caucasian donors: 3 pairs of female monozygotic twins (aged 23, 23 and 25 years old), 19 year old and 57 year old males, a 51 year old female and cord blood from a female newborn. CD4+ and CD8+ populations were isolated using CD4+ and CD8+ T-cell positive isolation kits (Invitrogen), CD45RO+ and naive cells were isolated from PBMC using CD45RO+ enrichment and human naive T-cell isolation kits (Myltenyi) respectively. cDNA of TCR alpha and beta chain was synthesized and sequenced on the Illumina HiSeq platform (see S1 Fig for library preparation technique, S1 Table for the oligonucleotides used, S2 Table for all samples and numbers of sequencing reads).
Raw data processing and data analysis were performed using published open-source software tools: MiGEC (https://github.com/mikessh/migec), MiXCR (https://github.com/milaboratory/mixcr/), tcR (https://github.com/imminfo/tcr) and repgenHHM (https://bitbucket.org/yuvalel/repgenhmm/downloads). We processed raw sequencing data with MiGEC [40] to extract unique molecular identifiers and we used MiXCR [41] to determine the CDR3 position. All raw data is available online on our server (see S1 Text Methods E. for the links) and also in Short Read Archive (SRP078490).
Recombination models for beta and alpha chains were inferred using an EM-algorithm as described in [11, 13, 14], using the repgenHHM [13] and IGoR [14] software tools, selection models were inferred as described in [7]. The shared clonotype analysis was performed using the tcR package [42] and R statistical programming language [43]. To predict the number of shared out-of-frame clonotypes we generated random sequences using the recombination model parameters inferred separately for each individual in the previous step. We then filtered out-of-frame clonotypes and calculated the number of shared sequences between these simulated datasets using the tcR package.
To predict the number of shared in-frame clonotypes we also generated random sequences with recombination model parameters, filtered in-frame sequences and calculated the Q selection factors for each CDR3 amino acid sequence using selection models inferred separately for each individual. The number of shared sequences in the simulated in-frame datasets was reweighted by the Q factors as:
1 | S 1 | · | S 2 | ∑ s ∈ S 1 ∩ S 2 Q ( 1 ) ( s ) Q ( 2 ) ( s ) , (1)
where S1, and S2 are two synthetic sequence samples drawn from two models P gen ( 1 ),P gen ( 2 ) learned separately from the out-of-frame sequences of the two individuals, and Q(1)(s), Q(2)(s) are selection factors learned separately from these two individuals’ in-frame sequences. |S1| and |S2| denote the size of the two samples. The sum runs over sequences s found in both samples.
To estimate the distribution of the number of inserted nucleotides for different subsets of the repertoire (Figs 3 and 4), we used the same EM-algorithm when inferring the full repertoire models. To minimize the noise due to small subset sizes, we only learned the insertion distribution and took all other model parameters to be the same as in the previously inferred model in [11].
To fit the exponent decay of the ageing data we used the nlm2 R package. The data used in these fits is given in S3 Table. Fitting an exponentially decaying curve to the fraction Z of zero-insertion clonotypes in the 2000 most abundant clones as a function of age T (Fig 4):
Z ≈ c + a exp ( - b T ) , (2)
we found c = 0.00363 ± 0.00154, b = 0.0272 ± 0.0091 yr−1, and a = 0.016696± = 0.00188.
Fitting an analogous model for the attrition of the naive T-cell pool, i.e. the fraction N of naive T-cells as identified using flow cytometry (see [23] for details),
N ≈ a ′ exp ( - b ′ T ) . (3)
we obtained a′ = 0.68 ± 0.054 and b′ = 0.01485 ± 0.0018 yr−1.
All blood samples were taken in authorised diagnostics lab. All donors signed informed consent document for scientific use of their blood and publishing the results. Study was approved by local ethics committee and conducted in accordance with the Declaration of Helsinki.
|
10.1371/journal.pgen.1000121 | Zebrafish Whole-Adult-Organism Chemogenomics for Large-Scale Predictive and Discovery Chemical Biology | The ability to perform large-scale, expression-based chemogenomics on whole adult organisms, as in invertebrate models (worm and fly), is highly desirable for a vertebrate model but its feasibility and potential has not been demonstrated. We performed expression-based chemogenomics on the whole adult organism of a vertebrate model, the zebrafish, and demonstrated its potential for large-scale predictive and discovery chemical biology. Focusing on two classes of compounds with wide implications to human health, polycyclic (halogenated) aromatic hydrocarbons [P(H)AHs] and estrogenic compounds (ECs), we generated robust prediction models that can discriminate compounds of the same class from those of different classes in two large independent experiments. The robust expression signatures led to the identification of biomarkers for potent aryl hydrocarbon receptor (AHR) and estrogen receptor (ER) agonists, respectively, and were validated in multiple targeted tissues. Knowledge-based data mining of human homologs of zebrafish genes revealed highly conserved chemical-induced biological responses/effects, health risks, and novel biological insights associated with AHR and ER that could be inferred to humans. Thus, our study presents an effective, high-throughput strategy of capturing molecular snapshots of chemical-induced biological states of a whole adult vertebrate that provides information on biomarkers of effects, deregulated signaling pathways, and possible affected biological functions, perturbed physiological systems, and increased health risks. These findings place zebrafish in a strategic position to bridge the wide gap between cell-based and rodent models in chemogenomics research and applications, especially in preclinical drug discovery and toxicology.
| To understand chemical-induced biological responses/effects, it is important to have large-scale and rapid capacity to investigate gene expression changes caused by chemical compounds at genome-wide scale in an adult vertebrate model; this capability is essential for drug development and toxicology. Small aquarium fish with vast genomic resources, such as zebrafish, will probably be the only vertebrate models that allow for cost-effective, large-scale, genome-wide determination of gene expression net changes in the entire adult organism in response to a chemical compound. Presently, such a whole adult organism approach is only feasible in invertebrate models such as the worm and fly, and not in rodent models, hence the usefulness of such an approach has not been demonstrated in a vertebrate. By using two classes of chemicals with wide implications to human health, we showed that capturing net changes of gene expression at a genome-wide scale in an entire adult zebrafish is useful for predicting toxicity and chemical classes, for discovering biomarkers and major signaling pathways, as well as for inferring human health risk and new biological insights. Our study provides a new approach for genome-wide investigation of chemical-induced biological responses/effects in a whole adult vertebrate that can benefit the drug discovery process and chemical toxicity testing for environmental health risk inference.
| Chemogenomics, application of genomic tools in pharmacology and toxicology, offers a promising approach that will enhance drug discovery (target identification/validation, lead identification, efficacy evaluation) and toxicity assessment [1],[2]. Presently, invertebrates such as the worm Caenorhabditis elegans and fly Drosophila melanogaster, are the only animal models that have benefited from whole-adult-organism expression chemogenomics [3]–[6]. The benefits of whole-adult-organism chemogenomics would usually translate into large-scale, high-throughput, high-content and cost-effective applications for chemical biology. It is highly desirable that the benefits of whole-adult-organism chemogenomics can be realized in a vertebrate model because of the many biological processes, health risks and diseases that are restricted to a mature vertebrate system including humans. The existing cell-, fly- and worm-based models, while suited for high-throughput chemogenomics, lacked the relevant physiological whole-organism setting of an adult vertebrate. This is especially important in the context of pharmacology and toxicology when many of the potentially targeted organ-systems such as the endocrine, digestive (liver in particular), immune, muscular-skeletal, vasculature, kidney are absent from the existing high-throughput models. In contrast, the rodent models, though providing in vivo adult vertebrate data, are not suited for high-throughput applications and are not cost-effective [7], thus creating a bottleneck situation when in vivo biological data, especially toxicology, is required for the high number of ‘hits’ generated from in vitro screenings or for the many newly emerging industrial compounds and waste that are coming into contact with the public and environment. We propose that whole-adult chemogenomics performed on a small vertebrate such as the zebrafish would be a strategy that is sufficiently high-throughput, cost-effective and would generate high content in vivo vertebrate data potentially useful for large-scale screening and toxicity testing purposes.
Conceptually, whole-adult-organism expression chemogenomics would capture the sum-total of the transcriptomic changes in an entire adult organism as a single biological entity responding to exogenous chemical cues. This, however, would have its inherent limitations such as loss of weak signals or signals from smaller tissues and loss of specific location of response, and they may be compounded further by the greater biological complexity in vertebrates compared to invertebrates. Thus, while whole-adult-organism chemogenomics had been shown to be useful in invertebrate models with regard to compound screening [3],[4] and identifying biological processes affected by specific compounds [5],[6], it is not known if chemogenomics data generated from a whole adult vertebrate will be useful. We hypothesized that since strong and well-represented expression signals are likely to be detected in whole-adult-organism chemogenomics, the expression signals that are captured would be robust for predictive chemical biology and for uncovering biology that is widely associated with the chemical-induced responses/effects in the adult vertebrate.
However, the idea of performing high-throughput whole-adult-organism chemogenomics on a vertebrate model was practically not feasible, if not unimaginable, until microarray technology was made available to small aquarium fish such as the zebrafish. The availability of the zebrafish in large numbers, its small size, low husbandry cost, vast genomic resources and its present use in disease modeling [8] and drug screening [9],[10], make the zebrafish ideal for high-throughput whole-adult-organism chemogenomics. Moreover, owing to their close physiological relationship with the environment, fish are highly sensitive to environmental changes particularly exogenous chemical cues; therefore the impact of chemical effects on fish system is more easily defined and readily studied than on terrestrial species [11]. Previously, we and others have shown that zebrafish responded biologically to chemicals, such as small molecules, drugs and environmental toxicants, in a similar manner as mammals [12]–[16]. In this study, we chose P(H)AHs [represented by Benzo[a]pyrene (BAP), 3-Methylcholanthrene (MC), 2,3,7,8-Tetrachlorodibenzodioxin (TCDD)] and ECs [represented by 17-beta estradiol (Es), Diethylstilbestrol (DES), Bisphenol A (Bis)] as model compounds because they represent two classes of chemicals with wide implications to human health. Both P(H)AHs and ECs are potent AHR and ER agonists, respectively, and these receptors are known to cross-talk and they are regulators of important cellular functions that are involved in various biological processes and have been associated with several patho-physiological conditions [17]–[19]. Some of these compounds have been used as drugs or investigated for therapeutic potential [20]–[22]. Moreover, both classes of compounds are also environmental carcinogens and endocrine disruptors that have generated considerable public health concern [23],[24]. By focusing on P(H)AHs and ECs in this study, we performed chemogenomics on whole adult zebrafish and demonstrated that it is good for large-scale predictive chemical biology, for discovering biomarkers and major signaling pathways, as well as useful for human health risk and biological insight inference. Our study placed zebrafish in a strategic position to bridge the gap between in vitro cell-based model and in vivo rodent model in chemogenomics.
We generated 159 samples/arrays involving 28 treatment groups (4–7 replicates in each treatment group) in two experiments (‘A’ and ‘B’) and grouped them into four Datasets: I and II from experiment ‘A’ while III and IV from experiment ‘B’ (See Methods, Table S1 and Figure S1). Experiments ‘A’ and ‘B’ were performed one year apart using different batches of fish, reference RNA, reagents, array prints and experimental designs, which will test the robustness of the prediction models and help in identifying robust biomarkers. We trained six prediction models using Dataset I (Figure 1A and 1B) and validated them independently on Datasets II, III and IV (Figure 1C–1E). The validations were ‘independent’ in the sense that Datasets II, III and IV were totally left out and not used in the training of the prediction models that were tested on. Next, we trained another six prediction models using Dataset III (Figure 2A and 2B) and performed similar independent validation on the ‘unseen’ Datasets I, II and IV (Figure 2C–2E). The prediction models were trained using two supervised learning classifiers, k-nearest neighbours (kNN) and support vector machine (SVM), on selected discriminatory gene sets from Dataset I or III. During the training phase, the supervised learning classifiers used the individual discriminatory gene sets together with their expression data in Dataset I or III to produce respective sets of rules or reference weights that will serve as standards/models for the prediction of ‘unseen’ or unknown samples. The discriminatory gene sets for P(H)AH and EC classes were selected using threshold criteria of Q-value, FDR-value or P-value coupled with fold-difference between treated [combined representative groups of P(H)AHs or ECs] versus control samples in Dataset I or Dataset III (Figure 1A and 2A; details of criteria in Table S2). The different statistical treatments and learning classifiers were used to examine if any idiosyncrasy associated with data processing affected the performances of the prediction models.
First, we evaluated the prediction models using ‘leave-one-out’ cross-validation approach to avoid the statistical problem of over-estimating prediction accuracy that occurs when a model is trained and evaluated with the same samples. In this procedure, one of the 30 or 25 samples from Datasets I or III, respectively, was withheld and the remaining 29 or 24 of the respective samples were used to train a prediction model based on a selected discriminatory gene set to predict the class of the withheld sample. The process was repeated until all 30 or 25 samples were predicted in turn, and all prediction models were tested similarly. We found that the ‘Predicted Results’ for all the samples matched the ‘Expected Results’, hence 100% of the samples were correctly classified by all the prediction models (Figure 1B and 2B). The excellent performances of the prediction models were non-random (Fisher's Exact Test P-value = 8.35×10−9−4.89×10−7; Table S3), indicating their predictive powers.
To test the robustness of their predictive powers, we sought to validate them independently. In this procedure, we used the 30 samples from Dataset I to train prediction models based on the selected discriminatory gene sets and predict each of the 25 ‘unseen’ samples from Dataset III. Remarkably, the prediction models for P(H)AHs and ECs trained by Dataset I classified 100% of the samples in Dataset III correctly (Figure 1C). We then reversed the order by training the prediction models with the 25 samples from Dataset III and testing them on each of the 30 samples from Dataset I. The prediction models for P(H)AHs trained by Dataset III classified all the samples in Dataset I correctly while the prediction models for ECs performed comparably well though with one or two false negatives (Figure 2C). The findings indicate that the prediction models were remarkably robust because both Datasets I and III were obtained through two separate experiments that contained substantial biological and technical variations.
Next, we tested the specificity (the ability to identify negative cases) of the above prediction models on 9 types of compounds from classes with acute mode-of-action and toxicity effects differing from that of P(H)AHs and ECs. Thus, we tested each of the prediction models generated using the 30 and 25 samples from the respective Datasets I and III, on each of the 46 ‘expected-negative’ samples from Dataset II, independently. As anticipated, with only one exception, all samples were predicted as ‘negative’ matching the ‘Expected Results’ (Figure 1D and 2D). The robust performance of the prediction models generated from Dataset III is noteworthy since both Datasets II and III were derived from different experiments.
We also tested the performance of the prediction models on Dataset IV consisting of 58 samples obtained from fish exposed to multiple chemical mixture of BAP, DES and arsenic (As) at different concentrations and combinations. Arsenic was introduced to increase the complexity of the mixtures and to test if the prediction models could still perform well on samples exposed to mixtures consisting of a chemical not used in the training of the models. The performance of the prediction models for ECs was outstanding as they could classify 100% of the samples correctly notwithstanding the increase complexity that could arise from the mixture of compounds and that Datasets I and IV are from separate experiments (Figure 1E and 2E). The prediction models for P(H)AHs displayed comparable performances in terms of specificity although the sensitivity (ability to identify positive cases) varied from 62.5%–97.5% (Table S3). Notably, prediction models of P(H)AHs trained from Dataset I performed poorer on Dataset IV compared to those trained from Dataset III, suggesting that when mixtures were involved, the performance of the P(H)AH models were affected by the inter-experimental variations. In this case, different statistical approaches can affect prediction performance, but with appropriate statistical tests, it is possible to generate prediction models that are sufficiently robust. This was observed in the case of SVM-trained P(H)AH models using discriminatory gene sets from Dataset I selected based on FDR-val (sensitivity = 92.5%) and P-val (sensitivity = 90.0%) which performed better compared to Q-val (sensitivity = 62.5%) (Table S3).
Taken together, with the exception of Q-val_P(H)AH performance on Dataset IV, all the prediction models performed comparably well, in particular for most of ECs which scored 100% for specificity and sensitivity. The robust ability of the prediction models to discriminate compounds of the same class from those of different classes, even in a mixture of compounds, suggests that zebrafish whole-adult-organism chemogenomics is capturing genes associated with biological functions that are strongly affected by a class of compound. This demonstrates its potential use for compound screening, predictive chemical biology and biomarker discovery. The ability of the prediction models to tolerate a reasonable level of biological, technical and data-processing variations, indicate high amenability to real-life compound screening and predictive applications as such variations will inevitably occur over time, in different laboratories and experimental settings.
Having demonstrated their predictive powers, we used the discriminatory gene sets to identify potential biomarker genes for P(H)AHs and ECs. To do so, we first consolidated the discriminatory gene sets into their corresponding two major groups P(H)AHs and ECs by combining the gene sets within their respective classes (including only genes with unique GenBank Identity and similar mean expression directionality). Then, we examined the consistency of their expression profiles throughout all the 28 treatment groups in this study and validated some of the responsive genes using quantitative real-time PCR. A two-way hierachical clustering showed that the consolidated gene sets were able to cluster tightly the respective treatment groups including those in the mixture groups from the non-P(H)AH (Figure 3A) or non-EC treatment groups (Figure 3B). However, the gene expression profiles formed in the EC gene set (Figure 3B) were more distinct compared to the P(H)AH gene set (Figure 3A). This is due to the large number of genes that are specifically affected by ECs compared to the presence of certain xenobiotic metabolism or stress response-associated genes shared between P(H)AHs and other compounds, as well as the presence of some genes whose expressions are affected by compound mixtures. A closer examination of the genes reveals that ahr2 and its known regulated/responsive genes [25] such as cyp1A1, NQO1 homolog, nfe2l2, TIPARP homolog, gstp1, cyp1C1 were among the genes found in a tight cluster that shows similar expression pattern across the P(H)AH treatment groups (Figure 3A). Likewise, esr1 and its known regulated/responsive genes [26],[27] such as vg1 and vg3, nots, XBP1 homolog, NUPRI/P8 homolog were among the genes found in a tight cluster that shows consistent expression pattern across the ECs (Figure 3B). The findings indicate that zebrafish whole-adult-organism chemogenomics is able to capture important genes associated with major signaling pathways such as AHR and ER that are deregulated by the compounds. More importantly, the presence of many unknown genes clustering together with the known P(H)AH- or estrogen-responsive genes, suggests that these are potential novel biomarkers for P(H)AHs and ECs, respectively. The consistent expression patterns across several treatment groups containing P(H)AHs or ECs from two different experiments highlights the robustness of these biomarker genes.
In addition, we have independently validated 27 and 28 genes that were significantly (P-value <0.05) deregulated in the BAP and DES groups, respectively, using quantitative real-time PCR. We later observed that 16 and 21 of these genes were in the consolidated discriminatory gene sets of P(H)AHs and ECs, respectively (Figure 3; Table S4). As expected, there are a large number of significantly deregulated genes that are excluded in the discriminatory gene sets due to non-fulfillment of the stringent selection criteria but they are nevertheless responsive to BAP and/or DES. The good concordance of the microarray and PCR results for these genes gave us the confidence to pin down these biomarkers at the tissue level.
To confirm these biomarkers and obtain insights in targeted tissues, we performed a third independent experiment ‘C’ (Table S1) by treating zebrafish with BAP or DES followed by quantitative real-time PCR assay for the previously validated genes (Table S4) in seven tissue types (brain, gills, liver, gut, skin, testis and eyes). Out of the 27 and 28 validated genes for BAP and DES respectively, a total of 81 and 99 positive hits [significant gene deregulation (T-Test P-value <0.05)] were detected among the seven tissues (Table 1 and 2; Figure S2). About 37.0% (10/27) and 53.6% (15/28) of the validated genes in BAP and DES, respectively, were deregulated in 4–7 (>50%) of the selected tissues making them excellent biomarker genes for the respective class of compounds (Table 1 and 2). Among these 10 and 15 biomarkers for BAP and DES, respectively, 3 are well-known P(H)AH-responsive genes (ahr2, cyp1A1, and TIPARP homolog) and another 3 are well-known EC-responsive genes (esr1, vg1, and vg3), while the remainder 19 are potentially novel as their responsiveness to these compounds are relatively unknown/unreported especially within multiple tissue types. The P(H)AH biomarker genes are mainly associated with xenobiotic metabolism while the EC biomarker genes are mainly associated with molecular transport, metabolism and blood factors. Notably, about 90% (22/25) of these biomarker genes were found in the consolidated discriminatory gene sets. The remaining genes that were deregulated in 1-3 tissues may be useful biomarkers for tissue-specific analysis. Less than 10% of the validated genes in the tissues were found to be inconsistent with the whole fish data and this could be due to having not selected the appropriate responsive tissues or biological variations and cumulative effects of several tissues in the whole fish.
Among the selected tissues, eye and skin had the most BAP-responsive genes (63.0% and 51.9% of the validated genes, respectively) followed by gill, liver and testis (40.7% each tissue); these tissues yielded 79.1% of the total positive hits (Table 1; Figure S2). The findings are consistent with mammalian data as these organs (eye, skin, lung, liver and testis) are also known P(H)AH-targeted tissue in mammals [24],[28],[29]. As for tissues with the most DES-responsive genes, liver (92.9%) followed by gut (67.9%) and skin (60.7%) contributed to 62.6% of the total positive hits (Figure S2). The identification of these non-classical estrogen-targeted tissues is consistent with our previous data [30] as well as mammalian data [19],[31]. Activation of AHR or ER signaling pathways, as suggested by up-regulation of known responsive genes, was observed in many of these tissues. Interestingly, cyp1A which was up-regulated by BAP in all 7 tissues, was down-regulated by DES in 5 tissues, suggesting occurrence of similar inhibitory cross-talk between AHR and ER reported in mammalian cells [17],[18],[32]. Taken together, the findings show that the zebrafish shares similar biological responses in terms of molecules, signaling pathways and targeted tissues with mammalian system and is therefore a useful model for inference of chemical biology and health-risk inference in humans.
To evaluate the potential for extracting biological insights and health-risk inferences in humans, we mapped the consolidated discriminatory gene sets for P(H)AHs and ECs to available corresponding human homologs as previously described [14] and used them for knowledge-based data mining via Ingenuity Pathway Analysis (IPA) software (Figure 4A and 5A). Remarkably, the analysis of the human homologs from the consolidated P(H)AH and EC gene sets listed many affected molecular and cellular functions, perturbed physiological systems and human diseases/disorders that are known to be associated with these compounds [19], [23], [24], [28]–[31] (Figures 4 and 5; Tables S5 and S6). These also include canonical signaling pathways such as xenobiotic mechanism signaling (AHR, CYP1A, CYP1B1, CYP2C19, GSTP1, HSP90A, NFE2L2, NOS2A, NQO1, SULT2B1), ERK-MAPK signaling (EFL3, RAC1, STAT1, RPS6KA1) and PPARa/RXRa activation (CYP2C19, HSP90A, LPL, FABP1, NOS2A, ALAS1)for P(H)AH gene set and lipid metabolism (ACSL4, CYP1A1, CYP2C19, CYP51A1), protein ubiquitination pathway (PSMB6, PSMC2, PSMC5/SUG1, PSMC6, PSMD13) and coagulation cascade (FGB, SERPINA1, SERPINC1) for EC gene set. A correlation can be observed between these biological associations suggesting that prolonged or substantial perturbation of these biological functions (Figures 4B and 5B) and physiological systems (Figures 4C and 5C) by a compound would increase the susceptibility/risk of certain diseases/disorders (Figures 4D and 5D). Significantly, cancer was listed among the top most (Fisher Exact Test P-value = 1.98×10−5−4.76×10−2) associated disease as most P(H)AHs and some ECs are potent carcinogens. In addition, reproductive system disease, inflammatory disease, hematological disease and neurological disease were significantly associated with both P(H)AHs and ECs as the two classes of compounds are known to affect molecules and functions involved with reproductive system, inflammation, blood and nervous system (Figures 4C–D and 5C–D). Interestingly, the association of psychological disorders with ECs were also significant (P-value = 3.34×10−4−3.72×10−2; Table S6). While it is well-known that estrogen affects mental health [33], the grouping of ESR1 with GPX4, PSMC6, DIABLO, FBXO9, XBP1 homologs which have been associated with bipolar disorder [34],[35] suggests that these molecules may also play a role in estrogen-related psychological disorders (Table S6). Several of these patho-physiological systems indicated to be affected by the compounds (Table S5 and S6), such as reproductive, respiratory, dermatological/connective tissue, digestive/metabolic, nervous/neurological and visual/ophthalmic, corroborated with our multiple targeted-tissue analysis which showed that many of the biomarkers were deregulated in these tissues (Tables 1 and 2; Figure S2), suggesting that they are good ‘biomarkers of effect’.
Thus, a whole-adult-organism representation fits well for human health-risk inferences as the biological information are obtained, not from single tissue but multiple tissues interacting in a complex biological system where diseases/disorders usually develop or off-target effects occur. Incidentally, while zebrafish has always been used to model after human diseases [8], here we demonstrate the potential of zebrafish for predicting disease susceptibility or health risk associated with the exposure to a compound, and this in turn can further help to develop chemical-induced zebrafish models of human diseases.
A closer examination of the top connected networks generated by IPA using P(H)AH and EC datasets revealed interesting biological insights (Figure 6). In the P(H)AH network (Figure 6A; P-value = 1.00×10−39) the well-established xenobiotic-responsive molecules (AHR, CYP1A1, CYP1B1, CYP2C19, HSP90AB1, NFE2L2, NQO1, TIPARP) were linked with key signaling molecules such as NR3C1/GR, STAT1 and IGF1 providing new insights into alternative mechanisms of how P(H)AHs could exert adverse effects resulting in various pathological conditions. Notably in Table S5, 6 categories of P(H)AH-related diseases/disorders were found to be associated with AHR, NR3C1/GR, STAT1 and IGF1, while all 24 categories of the related diseases/disorders were associated with at least one of the four molecules. Interestingly, it was only recently that stronger evidence of cross-talk between AHR and NR3C1/GR are emerging [36],[37]. Our analysis suggests that HSP90 [38] may be one of the mediators, as also observed in the relatively more studied AHR-ER cross-talk [39]. As both NR3C1/GR and STAT1 are known regulators of inflammatory and immune response [40], respectively, the AHR-NR3C1/GR cross-talk may be another pathway that could contribute to the known immuno-toxicity effects of P(H)AHs [41]. The opposing direction of expression for CYP1A1 and CYP2C19, as displayed in both P(H)AHs and ECs networks (Figure 6), are evidence of inhibitory cross-talk between AHR and ER. While CYP1A1 is a known targeted molecule of this inhibitory AHR-ER cross-talk [32], this is the first time a member of the CYP2 family that includes important drug metabolizing enzymes is implicated and its regulation appeared opposite to CYP1A1. Apart from the xenobiotic-responsive molecules, the ECs network (Figure 6B; P-value = 1.00×10−57) linked clusters of molecules associated with proteasome-mediated degradation (PSMC2, PSMC5/SUG1, PSMC6, PSMD13), endoplasmic-reticulum stress response (FKBP11, HM13, RRBP1, PDIA4, SRPRB, SSR1, SSR2, XBP1), and cell cycle/death (CISH, FKBP4, P8, PA2G4, PTCH1), providing insights into cellular homeostasis and pathology associated with ECs and ER [19],[42]. The linking of the 5 known estrogen-responsive transcription factors PSMC5/SUG1, XBP1, P8, PA2G4 and ESR1 suggests that, under the influence of ECs, these transcription factors play important roles in mediating endoplasmic-reticulum stress response, proteasome-mediated degradation and cell cycle/death, thus offering new insights into the regulation of ‘unfolded protein response’ which has been intensely studied due to its association with diseases, drug resistance and its potential as therapeutic targets [43]–[45]. While further investigation is warranted, the analysis demonstrate the discovery potential of zebrafish whole-adult-chemogenomics and serve the purpose of alerting the researcher of the potential molecular interactions and effects induced by the compounds at the early drug discovery stage.
In summary, we have demonstrated that zebrafish whole-adult-organism chemogenomics is practical and effective for large-scale predictive and discovery chemical biology. Specifically, we have generated robust prediction models, identified and validated biomarker genes in multiple targeted tissues, identified important signaling pathways and biological functions as well as inferred human health risks and biological insights for both P(H)AHs and ECs. As strong and well-represented expression signals are likely to be captured, this approach is valuable for acquiring a molecular snapshot of the chemical-induced biological state of an adult vertebrate, which includes biomarkers of effects, deregulated signaling pathways, as well as possible affected biological functions, perturbed physiological systems and increased health risks. Moreover, this approach allows for rapid sampling in large-scale experiments, abundant sample materials for assays (does not require pooling or amplification of samples) and easy scaling-up of experiments, hence affords greater statistical power for data analysis. These are essentials for successful high-throughput genomics applications and for building up large database for predictive chemical biology [2]. The zebrafish is more cost-effective than the rodent model for in vivo toxicology [12],[13] and there had been proposals and successful attempts of using chemical screens and toxicity testing in whole-adult zebrafish as there are biological processes and diseases that are mainly associated with adults [9],[46]. Therefore, with zebrafish, cost-effective in vivo adult vertebrate chemogenomics can be performed earlier in the drug discovery process and for industrial/environmental toxicology, where high resolution tissue-specific data is yet required but robust and informative in vivo toxicological data is deemed valuable. This will enable researchers to better understand the potential liabilities of new compounds before advancing them to clinical test and may help shift attrition upstream [2],[47],[48] or before allowing contact with the public or releasing them to the environment. This present study has provided a new strategy for genome-wide investigation of chemical-induced biological responses/effects in a whole-adult vertebrate model; where previously such whole-adult-organism chemogenomics approach were thought only feasible in invertebrate models. The realization of its potential can benefit the drug discovery process and toxicology, in particular chemical toxicity testing for environmental health-risk inference.
Three independent experiments (‘A’, ‘B’ and ‘C’) were performed in this study where different batches of adult male zebrafish were exposed to various chemical compounds at different concentrations (Table S1). The selected chemicals represent compounds with toxicological interests and/or environmental-health importance, and the concentrations used were based on available published data or our preliminary acute toxicity exposure experiments conducted for the compounds. Experimental procedures were performed within the guidelines of National University of Singapore's Institutional Animal Care and Use Committee (NUS-IACUC). The fish were immersed in the chemical solutions for 96 hours [experiment ‘A’] or 72 hours [experiment ‘B’ and ‘C’] at a density of 1 fish/200 ml at 27±2°C in a static condition. Control fish were kept in vehicle solution or water under similar condition. Chemical solutions and water were changed daily. At the end of experiment, individual whole fish were snap-frozen and pounded to powder in liquid nitrogen for subsequent total RNA extraction using Trizol reagent (Invitrogen, USA) protocol. For experiment ‘C’, specific tissues were snap-frozen in liquid nitrogen for total RNA extraction using RNeasy Mini Kit (QIAGEN, Germany) followed by DNAseI treatment and heat inactivation. Reference RNA was obtained by pooling total RNA extracted from male and female zebrafish. The integrity of RNA samples was verified by gel electrophoresis, and the concentrations were determined by UV spectrophotometer.
The oligonucleotide probes for this array were designed by Compugen (USA) and synthesized by Sigma Genesis (USA). The arrays contained 16,416 oligonucleotide probes. The probes were resuspended in 3× SSC at 20 µM concentration and spotted onto in-house poly-L-lysine-coated microscope slides using a custom-built DNA microarrayer in the Genome Institute of Singapore (GIS). The arrays were spotted and quality controlled essentially as described by Eisen and Brown [49].
For fluorescence labeling of cDNAs, 20 µg of total RNA from the reference and sample RNAs were reverse transcribed in the presence of dNTPs mixed with Aminoallyl-dUTP (Sigma, USA) followed by coupling with mono-functional NHS-ester Cy3 and Cy5 dyes (Amersham, USA), respectively. A common reference design is used where equal amount of RNA samples from control and chemical-treated group were labeled with Cy5 and the same amount of common pooled reference RNA is labeled with Cy3. For each array, the Cy5-labeled samples from either the control or chemical-treated group was co-hybridized with the Cy3-labeled common reference. Thus, the respective paired Cy5- and Cy3-labeled cDNAs were pooled, concentrated, and resuspended in DIG EasyHyb (Roche Applied Science) buffer for hybridization at 42°C for 16 h in a hybridization chamber (Gene Machines). After hybridization, the slides were washed in a series of washing solutions (2× SSC with 0.1% SDS, 1× SSC with 0.1% SDS, 0.2× SSC and 0.05× SSC; 30 sec each), dried using low-speed centrifugation, and scanned for fluorescence detection.
The arrays were scanned using the GenePix 4000B microarray scanner (Axon Instruments, USA) and the generated images with their fluorescence signal intensities were analyzed using GenePix Pro 4.0 image analysis software (Axon Instruments, USA). All the arrays gave a mean signal to background ratio more than 5 and had >90% of the gene features that gave a measurable signal. Only gene features that were not flagged were extracted and subjected to Lowess normalization for further analyses. The microarray raw data have been formatted to be compliant with MIAME standard.
Statistical comparison of the relative mean expression level for each gene between test groups [combined representative groups of P(H)AHs or ECs] and their respective control groups from Dataset I or III were performed using Student's T-test and Significance Analysis of Microarray [50] (SAM) yielding respective P- and Q-values for each gene. The resulting P-values were further adjusted for Benjamini and Hochberg False Discovery Rate (FDR). The discriminatory gene sets were selected based on statistical threshold indicated by Q-value, FDR-value and P-value coupled with 1.5 fold-difference between treated versus control samples (Figure S1 and Table S2). The discriminatory gene sets together with their expression data in Dataset I or III were used to train two supervised learning classifiers, kNN and SVM, which generated prediction models for P(H)AHs class and ECs class. The procedure includes a training phase and a testing phase. In the training phase, the discriminatory gene sets together with their expression data from Dataset I or III were used as inputs to produce a set of rules or reference weights as standards/models for the testing phase. A ‘leave-one-out’ cross-validation is usually incorporated to validate the goodness of the model to avoid ‘over-fitting’ it (i.e. only good for predicting the training dataset but not sufficiently generalize to work well on other new and unknown datasets). The testing phase uses the standards created during ‘training’ to assign a discriminator score to each unseen (not used in the training) or unseen sample. Based on this score each sample is placed ‘into’ or ‘out of’ the class and their performances in terms of prediction specificity and sensitivity were determined (Figure 1 and 2; Table S3). Fisher's Exact Test was further used to determine that the performances of the prediction models were non-random.
Equal amounts of total RNA samples from reference, control and test groups were reverse transcribed to cDNA. The cDNA samples were used for quantitative real-time PCR analysis, performed using the Lightcycler system (Roche Applied Science) with Lightcycler-FastStart DNA Master SYBR Green 1 (Roche Applied Science) according to the manufacturer's instructions. Statistical comparison of the relative mean expression level for each gene between test and control groups was performed using Student's T-test and P-value<0.05 is considered significant.
To evaluate the potential for human health-risk inference, the zebrafish genes were mapped to their corresponding human homologs using the GIS Zebrafish Microarray Annotation Database (http://giscompute.gis.a-star.edu.sg/~govind/unigene_db/) as previously described [14]. The human homologs of the zebrafish genes from the consolidated gene sets P(H)AHs and ECs were used to mine the human database via Ingenuity Pathways Knowledge Base software (www.ingenuity.com). The ‘Biological and Disease Function Analysis’ was performed to identify biological functions and systems as well as diseases/disorders that were significantly associated with the gene sets. Fisher's Exact test was used to calculate P-values in determining the probability that each biological function, system and disease assigned to that data set is due to chance alone. P-value<0.05 is considered significant by the algorithm. Networks are generated from the available human homologs mapped from the discriminatory gene sets, by maximizing the specific connectivity of the human homologs, which is their interconnectedness with each other relative to all molecules they are connected to in Ingenuity's Knowledge Database. Networks are limited to 35 molecules each to keep them to a functional size and a network score is generated based on the hypergeometric distribution and is calculated with the right-tailed Fisher's Exact Test. In this study, only the top network for the consolidated gene sets P(H)AHs (P-value = 1.00×10−39) and ECs (P-value = 1.00×10−57) were used for further analysis. |
10.1371/journal.pcbi.1002336 | A Model of Ant Route Navigation Driven by Scene Familiarity | In this paper we propose a model of visually guided route navigation in ants that captures the known properties of real behaviour whilst retaining mechanistic simplicity and thus biological plausibility. For an ant, the coupling of movement and viewing direction means that a familiar view specifies a familiar direction of movement. Since the views experienced along a habitual route will be more familiar, route navigation can be re-cast as a search for familiar views. This search can be performed with a simple scanning routine, a behaviour that ants have been observed to perform. We test this proposed route navigation strategy in simulation, by learning a series of routes through visually cluttered environments consisting of objects that are only distinguishable as silhouettes against the sky. In the first instance we determine view familiarity by exhaustive comparison with the set of views experienced during training. In further experiments we train an artificial neural network to perform familiarity discrimination using the training views. Our results indicate that, not only is the approach successful, but also that the routes that are learnt show many of the characteristics of the routes of desert ants. As such, we believe the model represents the only detailed and complete model of insect route guidance to date. What is more, the model provides a general demonstration that visually guided routes can be produced with parsimonious mechanisms that do not specify when or what to learn, nor separate routes into sequences of waypoints.
| The interest in insect navigation from diverse disciplines such as psychology and engineering is to a large extent because performance is achieved with such limited brain power. Desert ants are particularly impressive navigators, able to rapidly learn long, visually guided foraging routes. Their elegant behaviours provide inspiration to biomimetic engineers and for psychologists demonstrate the minimal mechanistic requirements for complex spatial behaviours. In this spirit, we have developed a parsimonious model of route navigation that captures many of the known properties of ants routes. Our model uses a neural network trained with the visual scenes experienced along a route to assess the familiarity of any view. Subsequent route navigation involves a simple behavioural routine, in which the simulated ant scans the world and moves in the most familiar direction, as determined by the network. The algorithm exhibits both place-search and route navigation using the same mechanism. Crucially, in our model it is not necessary to specify when or what to learn, nor separate routes into sequences of waypoints; thereby providing proof of concept that route navigation can be achieved without these elements. As such, we believe it represents the only detailed and complete model of insect route guidance to date.
| The impressive ability of social insects to learn long foraging routes guided by visual information [1]–[8] provides proof that robust spatial behaviour can be produced with limited neural resources [9]–[11]. As such, social insects have become an important model system for understanding the minimal cognitive requirements for navigation [12]. This is a goal shared by biomimetic engineers and those studying animal cognition using a bottom-up approach to the understanding of natural intelligence [13].
In this field, computational models have proved useful as proof of concept [14], [15] that a particular sensori-motor strategy [16] or memory organisation [17] can account for observed behaviour. Such models of visual navigation that have been successful in replicating place homing are dominated by snapshot-type models; where a single view of the world as memorized from the goal location is compared to the current view in order to drive a search for the goal [16], [18]–[26]. Snapshot approaches only allow for navigation in the immediate vicinity of the goal however, and do not achieve robust route navigation over longer distances [27], [28]. Here we present a parsimonious model of visually guided route learning that addresses this issue. By utilising the interaction of sensori-motor constraints and observed innate behaviours we show that it is possible to produce robust behaviour using a learnt holistic representation of a route. Furthermore, we show that the model captures the known properties of route navigation in desert ants. These include the ability to learn a route after a single training run and the ability to learn multiple idiosyncratic routes to a single goal. Importantly, navigation is independent of odometric or compass information, does not specify when or what to learn, nor separate the routes into sequences of waypoints, so providing proof of concept that route navigation can be achieved without these elements. The algorithm also exhibits both place-search and route navigation with the same mechanism.
Individual ant foragers show remarkable navigational ability, shuttling long distances between profitable foraging areas and their nest. Despite low resolution vision and the availability of odometric information, many ant species preferentially guide their foraging routes using learnt visual information [2], [29]–[31]. The robust extraction and learning of the visual information required for route guidance is a product of the interactions between innate behaviours and learning [12], [32]. We highlight these interplays by sketching out the career of an individual forager.
Upon first leaving the nest, a new forager performs a series of short learning walks where a carefully orchestrated series of loops and turns allow her to inspect the visual surroundings from close to the nest entrance [33]–[35]. The knowledge gained during these special manoeuvres means she will be able to use visual information to pin-point the nest entrance after future foraging trips. When she finally leaves the vicinity of the nest she is safely connected to it because of her path integration (PI) system [12], [36]. In order to perform path integration, odometry and compass information are continuously combined such that at all times during a foraging journey the ant has the direction and distance information required to take an approximately direct path home. However, PI is subject to cumulative error and cannot take account of passive displacements, such as by a gust of wind. To mitigate these risks and ensure robust navigation, ants therefore learn the visual information required to guide routes between the nest and their foraging grounds [Reviews: [12], [37]]. During the early stages of learning the ants are reliant on their PI system for homing. However, as they become more experienced they come to rely more and more on visual information for route guidance [31]. The use of PI also provides consistent route shapes thereby facilitating and simplifying the learning of appropriate visual information [32], [38].
Extensive behavioural experiments over many years have led to a knowledge base of properties and behavioural signatures of visually guided navigation in ants that can be summarised as follows:
Computational models of visual navigation in insects followed experimental findings where ants [42] and bees [16] had been shown to guide their return to a goal-location by matching retinotopic information as remembered from the goal. With their seminal snapshot model, Cartwright and Collett [16] showed that within a certain catchment area [46] subsequent search for a goal location can be driven by a comparison of the current view of the world and a view stored at that goal. This has inspired roboticists and biologists to develop homing models [19]–[26] where a single retinotopic view is used to get back to a location.
Snapshot style models represent elegant, but abstract, sensori-motor strategies for navigation yet there are two major directions where such models need developing. Firstly, although snapshot models are very useful for understanding the information that is available in a visual scene [21], [47], to fully understand visual navigation we must consider the constraints imposed by a particular motor system and means of locomotion. Secondly, we need to understand how visual knowledge can be applied to the guidance of longer distance journeys and not just to the pin-pointing of a single goal location.
A significant component to any view-based homing algorithm is the sensori-motor interaction. The original snapshot model was developed following extensive experiments with bees. In the final stages of locating an inconspicuous goal, bees and wasps are able to fix the orientation of their body axis, perhaps using compass information, and then translate in any direction [48]–[50]. Inspired by this, the original snapshot model relies on stored views and current views being aligned to an external frame of reference before a matching procedure is used to determine a homing direction [16]. This represents a significant challenge for ants, and also for bees and wasps when flying rapidly over longer distances, where translation is predominantly in the direction of the body axis. In the context of our proposed model, however, the tight coupling of sensation and action is used to simplify the problem of learning a route. For an ant with fixed eyes and a relatively immobile head a given view implicitly defines a direction of movement and therefore an action to take. This suggests the following approach:
Given the success of snapshot-type models in place-homing, it is natural to assume that navigation over larger scales, that is, along routes, could be achieved by internalizing a series of stored views linked together as a sequence. Route behaviour in this framework would entail homing from one stored view to the next in a fixed sequence. While it has been shown that the catchment areas of individual snapshots can be quite large [21]–[47], [53], attempts to model route navigation using linked view-based homing have shown it to be a nontrivial problem which requires the agent to both robustly determine at which point a waypoint should be set during route construction and when a waypoint has been reached during navigation [19], [27], [28]. Essentially, for robust route navigation using a sequence of snapshots, an agent needs place recognition to determine where along the route it is [54]. Here we propose a different model that develops and refines ideas that have been recently put forward as an alternative to such a scheme [55]. Instead of defining routes in terms of discrete waypoints all views experienced during training are used to learn a holistic route representation.
We test our proposed route navigation strategy in simulation, by learning a series of routes through visually cluttered environments consisting of objects that are only distinguishable as silhouettes against the sky. This represents a challenging task due to the paucity of information and the potential for visual aliasing, whereby two locations appear similar enough so as to be indistinguishable. Our results indicate that, not only is the approach successful, but also that the routes that are learnt show many of the features that characterise the routes of desert ants.
Our navigation algorithm consists of two phases. The ant first traverses the route using a combination of PI and obstacle avoidance (as specified in the Materials and Methods) during which the views used to learn the route are experienced. Subsequently, the ant navigates by visually scanning the world and moving in the direction which is deemed most familiar. In the later experiments, the route is learnt by a neural network and the familiarity of each view is the output of the trained network. However, to show the utility of the proposed scanning routine, without the added complication of learning a familiarity metric, we first explored the performance of a system with perfect memory. This was implemented by storing views experienced every 4 cm along a training route and using these to determine view familiarity directly. Following Zeil et al. [21] we calculate the sum squared difference in pixel intensities between rotated versions of the current view and each stored view. The minimum across all stored images and all viewing directions experienced during a scan of the world from the current location is deemed the most familiar view for that location and a 10 cm step is taken in the viewing direction associated with this minimum.
Figure 1 shows that by storing the views along a training path and using these to drive a subsequent recapitulation of the route, robust behaviour is achievable. We used our algorithm to learn three routes through an environment containing both small and large objects randomly distributed across the environment. Three subsequent navigation paths were attempted for each route. Of the nine paths, all but one successfully return to the nest location, with the one failure caused by the simulated ant being drawn out of the stable route corridor by the presence of a tussock that dominates the visual field and causes visual aliasing. Despite the noise added to the movements during the recapitulation, paths are idiosyncratic though inexact. Within a corridor centred on the original route both a good match and a sensible heading are recovered that will, in general, drive the simulated ant towards the goal (Figures 1B–D). Outside of this route corridor the best match becomes poorer and, particularly within areas containing a high degree of visual clutter (i.e. within a group of tussocks) the proposed direction of movement less reliably points towards the goal. This is seen most clearly in panel C where, very close to tussocks, a significant proportion of the homeward directions determined by the algorithm (white arrows in Figure 1B–D) point away from the goal location. Often these erroneous signals direct movements back into the route corridor, although this is clearly a matter of chance.
The routes that are generated show a distinct polarity meaning that they can only be traversed in a single direction as is evidenced by the coherence of the homeward directions (arrows in Figure 1B–D). Importantly, the actions that result from following this strategy are not tied to a coordinate system and are therefore completely independent from the PI system that provided the initial scaffold for learning. In addition, the resulting routes are not dependent on a chained sequence of actions; appropriate actions are taken at any location along the route corridor independent of how that location was reached.
One potential problem with this navigational strategy is that if the simulated ant overshoots the goal it will, in general, carry on heading in the same direction and move further and further away from the goal location (Figure 2A). This is because there are no training views that point back towards the nest once it has been passed. This problem can be mitigated by including an exploratory learning walk during the training phase, a behaviour seen in many species of ants [33]–[35]. These initial paths take the form of a series of loops centred on the nest as can be seen in Figure 2B which shows the learning walk of a Melophorus bagoti worker taken from a paper by Muser et al. [33]. Essentially, this process means that in the region around the nest there will always be some views stored which are oriented back towards the nest.
To explore the possible effects of these initial short learning walks, the views experienced along them were added to the set of inbound views used for route learning. Figure 2 shows the end section of a route navigated after training with and without a learning walk. In these tests the simulation was not stopped when the simulated ant reached the nest location, analogous to blocking the nest entrance in a behavioural experiment. With the addition of a learning walk (Figure 2B), as the simulated ant passes the nest, rather than the best match being from the training route and oriented upwards (as in Figure 2A), the best match comes from the learning walk. The simulated ant is drawn into the loop of the learning walk it first encounters, leading to the looped paths seen in Figure 2B. Close to the nest, the density of points from the learning walk increases and there are multiple views from nearby locations oriented in a variety of directions. The best match at subsequent points will then likely be from different learning walk loops and so the ant stops following a single loop and enters more of a search-type path around the nest. Thus, our algorithm demonstrates both route following and nest search with the same mechanism.
Here we have shown that by storing and using panoramic views as they were experienced and aligned during training, we can achieve visually guided route navigation through a scanning routine and without recourse to a compass. The model is of particular interest since the resulting paths show remarkable similarities to many of the features that we observe in the routes of ants. Specifically, independence from the PI system that is assumed to scaffold the original learning; distinct polarity of routes; formation of a route corridor; and procedural rules that can be accessed out of sequence. By including a learning walk we can also get visually driven search for the nest location from the same mechanism. This algorithm demonstrates the efficacy of using a simple scanning behaviour as a strategy for seeking familiar views. However, the algorithm relies on the unrealistic assumption of a perfect memory of views experienced along the training route. We next investigate a more realistic encoding of the visual information required for navigation by training an artificial neural network using the views experienced along a return journey and a learning walk.
Having shown that the proposed scanning routine can produce ant-like paths, we next addressed the problem of learning a familiarity metric to use in place of a perfect memory system. Instead of storing all of the views experienced on a training route, the views were used to train a two-layered artificial neural network to perform familiarity discrimination using an Infomax learning rule [59]. Each view was presented to the network in the order in which it was collected and then discarded. This means that the memory load does not scale with the length of route but remains constant. Once trained, the network takes a panoramic image as input and outputs a familiarity measure indicating the likelihood of the view from that location and orientation being part of the learnt route. The trained network was then used in conjunction with the scanning routine to drive route navigation by presenting the rotated views to the network, and choosing the most familiar direction as the direction in which to navigate. The only difference in the behavioural routine was that the scanning range was reduced from to a slightly more realistic scan centred on the direction of travel from the previous timestep.
In a first experiment using this approach we employed the Infomax system to learn the same three training paths as in the previous experiment using a perfect memory. As Figure 3 shows, in this instance all returns were completed successfully. We do not believe this indicates that the approach is more robust than the perfect memory system but simply that the noise added to the system did not happen to nudge the agent into an area of the environment where visual aliasing would occur. In other ways the results of this experiment are very similar to the results obtained using a perfect memory. As these routes were learned using a single exposure to each of the training views, we are thus able to fulfil another of the desiderata for ant-like visual mediated route navigation: that routes can be learnt rapidly, in this case following a single trial.
To show that learning was not environment specific we conducted further simulations. Environments with varying densities of tussocks were randomly generated and a simple algorithm that performed path integration with obstacle avoidance was used to generate paths through them. In all of the environments we provided a distant horizon consisting of bush-like and tree-like objects as would be present in the natural environment of Melophorus bagoti [33]. In these experiments we also included a simplified learning walk at the start of training to prevent the simulated ant overshooting the goal.
We first examined a low tussock density environment compared to the environment used previously. Performance was good, although the lack of nearby objects resulted in less consistent paths (Figure 4). Example views taken from the training route (Figure 4, right) show how the panorama of distant objects provide a stable frame of reference throughout the route. Despite the sparse visual information in this environment, the distant objects help to keep the return paths heading in the right direction. The effect that the structure of the learning walk has on the return paths can be clearly seen near the goal location. As the simulated ant nears the goal it gets drawn into a series of left and right sweeps that reflect the left and right inbound loops of the learning walk and are analogous to an ant's search for its often inconspicuous nest entrance.
We next used an environment with a more dense set of tussocks (Figure 5). In this more densely tussocked world the distant panorama is no longer visible at all points along the route. This clearly makes route learning more difficult as is evidenced by the failures in three of the four runs. Because noise is added to the simulated ant's heading during route recapitulation the simulated ant may stray into previously unexperienced parts of the environment which, even a short distance away from the learnt route, can look very different in this cluttered world. Two attempts fail early when noise added to the heading leads the simulated ant to go to the left of a small tussock taking it into a part of the world with which it is not familiar. The other two returns do reasonably well. They do show some circling of tussocks, driven by training views where the path goes very close to a tussock and dominates the visual field, however both paths make it very close to the nest. This is a challenging environment in which to navigate and was picked to be at the limit of the algorithm's learning power following just one training run; other runs using a similar density of tussocks were more successful. Performance also improved if we removed or reduced the noise that was added to the direction of movement at each timestep. Of course, ordinarily ants would incorporate knowledge from several foraging trips during which time their performance becomes more stable and robust. We investigate this in the next section.
Performance of our algorithm was often quite reliable following a single training run but there were still failures (Figure 5). Ants, however, do not just use a single training run but will continue to develop their knowledge of the surroundings during multiple runs. We therefore investigated the effect on performance if multiple subtly different training routes were combined. The path integration algorithm that we used allowed the generation of multiple paths that were similar but not identical. The views collected along a number of paths were used to train the network. The learning scheme did not need to be altered as each view collected was simply presented to the network in the order that it was experienced.
Performance is shown for a twelve metre route in one of the more challenging environments (tussock density 0.75 ) following 1, 2, 4 and 8 training runs (Figure 6). Using multiple training runs can be seen to aid robustness, and after 8 training runs (Figure 6, far right) the recapitulated routes are efficient and consistent, even in this high tussock density environment. With repeated training runs the network will be exposed to a more comprehensive set of views from the route than with a single training run. It should be noted that using, say, four training runs is not the same as sampling the views four times as often during a single training run. In the latter case, sets of four consecutive points are not independent of each other. Using multiple runs however, views from similar locations are coupled only through the environment and thus variation in the views reflects the variation that will be experienced during navigation. For instance, if the distribution of objects in the world means that the training routes are canalised down a narrow corridor, it is likely that the navigated route will also go down a narrow path and so it does not matter that the training views from each run are similar. However, if the route corridor is broader, or even allows multiple paths, then multiple training routes allow a wider set of views that might be experienced when navigating, to be captured. Multiple runs therefore allow a broader, more robust, route corridor to be learnt.
It has been shown that Melophorus bagoti are able to learn and maintain more than one route memory when forced to learn distinct return paths to their nest from a series of different feeders [6]. In the experiments performed by Sommer et al. [6], seven training runs along a first route were followed by a control run to test whether the ant had learnt the route. This training schedule was repeated for a further two routes that each led back to the same location - the nest. Finally, the ants were tested on the first two routes to see if they had retained the original route memories. Here we attempt to replicate this experiment using our route learning algorithm to learn three 10 m routes performed in an environment with a tussock density of 0.75 .
To do this we train a network using the first route. The network is then tested before we continue to train the network using views from the second learning route. The network is then again tested before the final training session using views from the third route, before finally being tested on all three routes. The performance can be seen in Figure 7. The network is able to learn and navigate multiple routes without forgetting the earlier ones. It is interesting to note that when the third route is recapitulated following learning, the paths tend to get drawn back onto the previously learnt route 2, representing a possible confabulation of these two memories within the network. The individual route memories are not held separately and the return paths for route 3 are drawn back to route 2 as, at that point in the world, views from routes 2 and 3 are similar. This is not wrong per se, as the important thing is that the routes lead safely back to the nest. Also this property of routes can be seen in the original paper [6].
We have presented a parsimonious model of visual route guidance which replicates the properties and characteristics of ant navigation. We believe this model represents the only detailed and complete model of insect route guidance to date. However, for us, the major value of the model lies in it being a proof of concept that simple architectures and mechanisms can underpin complex cognitive behaviours such as visually guided routes. Visual navigation requires a cognitive toolkit capable of learning appropriate information, organising memories robustly and also a way of converting those memories into spatial decisions. By considering the way that sensory and motor systems are tightly coupled through behaviour, and utilising familiarity measures to drive route recapitulation, we have produced a minimal cognitive architecture that demonstrates visual route guidance and visually guided search for a goal.
There are three key aspects of this work that we would like to discuss further: (i) Using a familiarity measure to guide routes; (ii) the holistic nature of the route representation; (iii) how learning walks allow route following and nest search with the same mechanism.
Our own experience tells us that the human capacity for visual recognition is remarkable and clearly outstrips our capacity for recall. For instance, our ability to decide whether we have met somebody before, runs to many more people than those we can explicitly recall specific facts about. Theoretical investigations of abstract neural network models back up this intuition, with familiarity discrimination or recognition models having far greater capacity than associative models with the same number of processing units and weights [60]. Given the limited neural resources available to an ant and the need for rapid learning it makes sense to develop a navigational strategy that relies on recognition, as building either a cognitive map or employing some other form of associative learning are both harder tasks.
The fact that, in our experiments, sensible behaviour can be generated following a single traversal of a route indicates that a form of recognition memory may be sufficient for route navigation in the real world. In fact we would expect that in many ways the problem would be easier for an ant operating in the real world where there would be more information available to disambiguate different views and thereby reduce visual aliasing. The current model presupposes that the only information available to guide behaviour is provided by the high contrast silhouettes of objects against the sky. While we know that ants are able to use skylines to orient themselves [7], any additional visual information, for example colour, texture or celestial cues, information from other modalities [57], [61], [62], or internal motivational cues, would only help to reduce aliasing and improve reliability.
Whether insects have the appropriate brain architecture for storing visual information in this way is not known, though the mushroom bodies would be the obvious candidate neural structure. These higher brain centres, that are enlarged and elaborated in central place foraging insects, have been implicated in a number of cognitive functions including olfactory processing and associative learning [63]–[65], attention [66], sensory integration , sensory filtering [67], [68] and spatial learning [69], [70]. Farris and Schulmeister [71] present compelling evidence that large mushroom bodies receiving visual input are associated with a behavioural ecology that relies heavily on spatial learning. Furthermore, recent research by Stieb et al. [72] implicates the mushroom bodies in the behavioural transition from working inside the nest to foraging outside. In light of our model it would be interesting to evaluate the potential of the mushroom bodies for familiarity discrimination or recognition memory.
The pseudocolour plots in Figure 3 indicate how familiarity could provide another source of information for making routes more robust. If an agent was able to follow a combination of the gradient of the familiarity and the heading of the most familiar direction this would have the effect of drawing the recapitulated paths back onto the habitual route. While this gradient is apparent in the plots that are obtained by sampling from a dense grid of points, it is less obvious how an ant might extract this information, since it would be necessary to sample at least three non-collinear points whilst maintaining the most familiar heading. For a flying insect this would be much less of a problem. The familiarity gradient alone will only serve to draw paths back onto the route and will therefore not produce route following behaviour. However, preliminary results indicate that performance is more robust when the direction indicated by the most familiar view is combined with the familiarity gradient.
We have shown how a familiarity metric could in principle be used to guide successful route navigation; the proposed motor program is however not realistic. Although ants have been observed performing scanning behaviours such as we have used, in general they proceed in a far more purposeful manner when on or near their habitual routes. One issue that we need to address therefore is how familiarity of views could be used in a way that is more consistent with the fine-grained movements that ants actually perform. In order to do this we will need to simulate an environment in which behavioural experiments have been conducted and record in fine detail the movements of ants during their foraging career.
In our second set of experiments we train a network with the views experienced during a learning walk and along a route. There is no requirement for specific views to be selected and following training the network provides a holistic representation of visual information rather than a set of discrete views. The network in fact holds a holistic representation of all the visual information needed for the agent to get to a particular goal, as shown by our replication of multiple route learning. We have previously shown that other neural network models are also able to holistically encode this information [55], [73]. However the particular elegance of the Infomax procedure is that each view is presented to the network once and then discarded.
The consensus view amongst biologists is that ants do not hold spatial information in a unitary cognitive map [12], [74]–[76]. Indeed experiments have shown that the memories required to get to one goal (e.g. the nest) are insulated from the memories required to get to a second goal (e.g. a regular feeding site) [4], [77]. Indeed, if food-bound and nest-ward routes do not overlap then ants captured as they try to get to their nest are effectively lost if they are placed on their familiar food-bound route [4]. Our model could account for this if the motivational state of the animal formed part of the input to the familiarity network. In this way, views would appear familiar only within the correct motivational context.
One of the key properties of this model is that route guidance and place search come from the same mechanism. This comes from incorporating the views experienced during a learning walk into the overall task. Learning walks (and flights in bees and wasps) are a form of active vision where the insect shapes its own perception in a way that is beneficial for learning. This principle is demonstrated by our design of an artificial learning walk. If the views on the outbound sections of the learning walk are made to be more variable than those on the inbound sections, then the inbound views will be learned preferentially. A simple way to achieve this is to have curved outbound routes and straight inbound routes (see the Materials and Methods), a learning walk scheme that performed well. We imagine that when we have an understanding of how real learning walks are structured by the environment, performance will be improved and search paths will more closely resemble those that have been observed in ants.
Another more complex way to modulate learning would be to turn-off learning when not heading towards the nest. This would require some sort of input from the PI system and interestingly, recent detailed descriptions of learning walks in Ocymyrmex [34] highlight that PI is likely to be used to ensure ants look at the nest at discrete points during their learning walks. However, these learning walks are still compatible with either behavioural or cognitive modulation of learning. The use of PI might only be used to structure the learning walks and allow the ant to accurately face its nest thereby facilitating behavioural modulation of learning [52].
We have presented a parsimonious model of visual navigation that uses the interaction of sensori-motor constraints with a holistic route memory, to drive visual navigation. The model captures many of the observed properties of ant navigation and importantly visual navigation is independent of odometric or compass information. Additionally, in the model one does not need to specify when or what to learn, nor separate routes into sequences of waypoints, thus the model is a proof of concept that navigation in complex visual environments can be achieved without those processes. Our principal goal in this research project is to understand the likely and viable mechanisms underpinning insect navigation. Therefore our next step will be to evaluate the model using fine-grained recordings of ants learning and performing routes in their natural habitat.
To create the environments used in our experiments, a distant panorama of trees and bushes was generated and uniformly distributed densities of tussock-like objects were created over a central region. While the placement of the tussocks was performed by sampling from a uniform distribution, environments that did not contain many tussocks in the vicinity of the training paths were rejected. In some of the experiments additional 3D objects such as large trees and a building were added within the central region. The environment is intended to produce panoramic views that are typical of the natural environment of the Australian desert ant Melophorus bagoti (See [7], [8], [37], [78] for example images of this environment). Figure 8 shows an overview of a typical environment together with a series of views along a route. Notice how variable the views are and also how insignificant the large tree and the building (the solid black objects in Figure 8B) can be from the perspective of an ant. This is easiest to see in Figures 8D and E taken from the middle section of the route where the house, which is NE of the ant (i.e. just over half way along the image; notice the triangular roof in the high-resolution image) blends in to a tussock.
The simulated environment, programmed in MATLAB, consists of objects formed from flat black triangular patches as described below and rendered at a high resolution (Figure 8D), prior to being re-sampled at the low resolution of the simulated visual system (Figure 8C,E). This allows for subtle changes to be registered in response to small movements as would be the case for an ant with a low resolution visual system acting in the real world. This means the resultant view is composed of grey-scale values when a pixel is neither completely covered by sky nor completely covered by an object. In our simulated environment nearby objects are rendered in three dimensions whereas objects at a distance greater than 20 m from the route are flat but oriented so as to be maximally visible.
Once an environment consisting of triangular patches has been created, a panoramic view from any position within the environment can be generated as follows. We first change the origin of the world to coincide with the position of the simulated ant by subtracting the current x, y and z coordinates of the ant from the set of vertices, X, Y and Z that define the triangular patches. The set of vertices [X, Y, Z] are then converted into spherical coordinates [] that represent the azimuth, elevation and radial distance. The radial information is discarded and the patches re-plotted in 2D giving the required binary panoramic view (Figure 8D) which is stored as a high-resolution [size ] binary matrix. The final step is to reduce the resolution of this image to , which represents the approximate sampling resolution of the compound eyes of Melophorus bagoti workers [78]. The binary matrix is resized to [] using the imresize function in MATLAB and the average value of each [] block is then used as the value of the corresponding pixel in the low resolution representation. This averaging results in values in the range [0,1], with values between the two extremes indicating the fraction of sky and objects covered by a pixel in the original high resolution image [Figure 8E].
The routes shown in Figure 8 and used in the first sets of experiments (reported in Figures 1 and 3) are return paths taken from a paper by Muser et al. [33] that describes the foraging ecology of Melophorus bagoti. While we have no knowledge of the real environment from which these paths were recorded we assume that the overall straightness of the paths is somewhat typical and that they therefore represent a reasonable example of the sort of paths that these ants must learn.
In subsequent experiments, paths were generated iteratively starting from the end point of the outbound route using a combination of path integration and obstacle avoidance. Path integration was approximated by centring a Gaussian distribution with a standard deviation of on the correct homeward direction and sampling from this distribution. Obstacle avoidance was incorporated into the path generation scheme by modulating the Gaussian distribution used for path integration by multiplying it by the proportion of sky visible in each direction, (effectively the inverse of the height of the skyline; Figure 11A,B), raised to the power of 4, (Figure 11C). The resulting modulated Gaussian (Figure 11E) was renormalized and sampled from to determine a movement direction and a 4 cm step was taken in this direction. Training images are collected after every step. The obstacle avoidance modulation has the effect of biasing movements towards lower portions of the horizon while preventing completely movements towards objects that fill the entire visual field in the vertical direction. Due to the sampling involved in this process, individual paths between two locations will vary slightly allowing the collection of subtly different sets of images describing a route. A return path was considered complete when the distance to the nest was less than 4 cm.
Where learning walks were added to the training routes, we sampled views from pre-specified paths around the nest. Ants generally walk slower during their learning walks and so samples were taken every 2 cm along the paths as opposed to the 4 cm sampling that was used for generating the route data. These views are added to the start of the set of route views used to train the network in the order that they appear, beginning at the nest. The learning walk in Figure 2B is taken from [33] but see also [34] for another route shape that could have similar properties. The artificial learning walks were generated using a circular path with a radius of 0.5 m for the outbound section and a straight path for the inbound section (Figure 12). This is inspired by data from the learning flights of bumblebees whose early learning flights contain many loops with an inward portion oriented directly at the nest [A. Philippides, personal observation].
In the experiments that we report using a perfect memory system, route recapitulation is performed using a complete scan of the environment (in steps of ) at each timestep. Normally distributed noise with a standard deviation of is added to the preferred direction of movement and a 10 cm step is made in this direction (Figure 13).
In the experiments that we report using the Infomax model we employ a slightly more realistic scanning routine during route recapitulation and instead of performing full scans we limit the scans to the frontal in steps of relative to the current heading. We did this to make the scans more similar to those that real ants produce which are rarely as large as . This had a negligible effect on performance except that it made it impossible to follow a path that had any turns greater than as were present in the Muser et al. learning walk in Figure 2B. As before, normally distributed noise with a standard deviation of is added to the preferred direction of movement and a 10 cm step is made in this direction. However, when generating the pseudocolor plots in Figures 4 and 5, we did not have a current heading and so performed a full scan to generate an assumed movement direction.
For the perfect memory system each of the views experienced along a training path was stored. we then calculated a familiarity metric as minus the minimum of the sum squared difference in pixel values between the current view and each of the stored views, .(1)
The maximum familiarity score across all rotated versions of the current view will be obtained for the most similar stored view and the direction from which this maximum was attained determines the next movement to make. In this setting, if the simulated ant does not stray from the training path then it is guaranteed to choose the correct direction to move at each timestep. This is because the most similar view will always be the one that was stored at that location while facing in the direction required to recapitulate the route.
In order to perform familiarity discrimination we chose to use a neural network model that was specifically designed to perform this task [59]. The architecture consists of an input layer and a novelty layer with activation functions (Figure 14). The number of input units is equal to the dimensionality of the input which in our case is , the number of pixels in a down-sampled view of the world. The number of novelty units is arbitrary and here we follow [59] and use the same number of novelty units as inputs. We found that using as few as 200 novelty units can work well in many instances. We did not explore this aspect of the problem in any detail since we were more interested in the behavioural consequences of a familiarity driven approach. The network is fully connected by feedforward connections . Weights are initialised randomly from a uniform distribution in the range and then normalised so that the mean of the weights feeding into each novelty unit is 0 and the standard deviation is 1. The network is trained using the Infomax principle [79] adjusting the weights so as to maximise the information that the novelty units provide about the input, by following the gradient of the mutual information. The core update equation (4) in our learning scheme performs gradient ascent using the natural gradient [80] of the mutual information over the weights [81] (use of the natural gradient avoids the computationally expensive calculation of the inverse of the entire weight matrix). Since two novelty units that are correlated carry the same information, adjusting weights to maximise information will tend to decorrelate the activities of the novelty units and the algorithm can thus be used to extract independent components from the training data [81]. We choose to use this approach mainly because it only requires a single pass through the data. This means that each view is experienced just once and then discarded. While with a limited amount of data the algorithm is unlikely to converge to a particularly good set of independent components, it is enough that the components that are extracted provide a more suitable decomposition of the training data than of an arbitrary input. During learning the activation of each of the novelty units is computed as:(2)where is the value of the input and is the number of input units. The output of the novelty units is then given by:(3)The weights are adjusted using the following learning rule:(4)where is the learning rate and is set as 0.01 for this paper. Finally, the response of the network to the presentation of an unseen N-dimensional input is computed as(5)where denotes the absolute value. The network response could be viewed as an output layer but as it is a function of the activations of the novelty units, we follow [59] and do not represent it with another layer (Figure 14). As noted above, in this paper we set and the network is trained with each training view presented just once to the network in the order in which it is experienced in training. In [59] the authors use together with a threshold that must be determined empirically to determine whether the input is novel or familiar. For our purposes it is not necessary to determine a threshold as we only need to choose the most familiar input from a limited number of possibilities i.e. the views experienced during a single scan of the environment.
The difference between the way an image difference function and a neural network trained using an Infomax principle represent familiarity will be subtle. In essence, the difference is manifest in the way the information is stored. For image differences, each stored view defines a single point in an n-dimensional space, with n equal to the dimension of the images (n = 90×17 = 1530) and the image difference function gives the squared Euclidean distance of an input image from one of these stored points. This requires all of the views to be stored and so memory load increases as more views are experienced. The Infomax approach instead decomposes each view into a fixed number of components (determined by the number of hidden units in the network) which remains constant, independent of the number of views experienced. The Infomax measure is more abstract and reflects whether a test input is well described in terms of the learned components that the hidden units represent. By decomposing the input in this way it is possible compress redundant data resulting in more efficient memory storage.
|
10.1371/journal.pgen.1000474 | Ancient mtDNA Genetic Variants Modulate mtDNA Transcription and Replication | Although the functional consequences of mitochondrial DNA (mtDNA) genetic backgrounds (haplotypes, haplogroups) have been demonstrated by both disease association studies and cell culture experiments, it is not clear which of the mutations within the haplogroup carry functional implications and which are “evolutionary silent hitchhikers”. We set forth to study the functionality of haplogroup-defining mutations within the mtDNA transcription/replication regulatory region by in vitro transcription, hypothesizing that haplogroup-defining mutations occurring within regulatory motifs of mtDNA could affect these processes. We thus screened >2500 complete human mtDNAs representing all major populations worldwide for natural variation in experimentally established protein binding sites and regulatory regions comprising a total of 241 bp in each mtDNA. Our screen revealed 77/241 sites showing point mutations that could be divided into non-fixed (57/77, 74%) and haplogroup/sub-haplogroup-defining changes (i.e., population fixed changes, 20/77, 26%). The variant defining Caucasian haplogroup J (C295T) increased the binding of TFAM (Electro Mobility Shift Assay) and the capacity of in vitro L-strand transcription, especially of a shorter transcript that maps immediately upstream of conserved sequence block 1 (CSB1), a region associated with RNA priming of mtDNA replication. Consistent with this finding, cybrids (i.e., cells sharing the same nuclear genetic background but differing in their mtDNA backgrounds) harboring haplogroup J mtDNA had a >2 fold increase in mtDNA copy number, as compared to cybrids containing haplogroup H, with no apparent differences in steady state levels of mtDNA-encoded transcripts. Hence, a haplogroup J regulatory region mutation affects mtDNA replication or stability, which may partially account for the phenotypic impact of this haplogroup. Our analysis thus demonstrates, for the first time, the functional impact of particular mtDNA haplogroup-defining control region mutations, paving the path towards assessing the functionality of both fixed and un-fixed genetic variants in the mitochondrial genome.
| Mitochondria, the ‘power plant’ of the cell, have their own distinct genome (mtDNA), whose sequence varies among individuals around the globe. This variation, which was formed by the accumulation of mutations (variants) during the course of evolution, appears to alter the susceptibility to common complex diseases (such as Parkinson's disease and diabetes). However, since the accumulation of mtDNA mutations over time results in the formation of new combinations (genetic backgrounds), it is not clear which of the mutations are functional and which are “evolutionary silent hitchhikers”. Thus we aimed at assessing the functionality of mtDNA genetic variants, focusing on variants within the mtDNA regulatory region, hypothesizing that they could affect mtDNA activity and maintenance. We found that a variant defining mtDNA genetic background ‘J’ significantly increased the transcriptional efficiency and elevated mtDNA copy numbers in cells, as compared to other genetic backgrounds. Hence, mtDNA regulatory region variants can affect mtDNA maintenance, which may partially account for the involvement of this genetic background in disease susceptibility. Our analysis demonstrates, for the first time, the functional impact of a particular mtDNA variant that was fixed during evolution. Moreover, our findings underline the functionality of mtDNA variants in the evolutionary variable regulatory region.
| Mitochondria are the major sources for cellular energy, through the process of oxidative phosphorylation (OXPHOS), and thus play a central role in cell life and death. Since mitochondrial DNA (mtDNA) encodes 13 essential proteins of the energy production apparatus and 24 key factors of their translation machinery (i.e. 22 tRNAs and 2 rRNAs) it is not surprising that numerous association studies have demonstrated the involvement of mtDNA genetic backgrounds (haplotypes, haplogroups) in complex human disorders as well as in selective events that transpired during human evolution [1]. Such implied functional potential of natural mtDNA variants has gained recent experimental support in human and murine cytoplasmic hybrids (cybrids) [2],[3]. Further support was provided by back-cross experiments in Drosophila and rat as well as by inter-populations crosses in Tigriopus Americana revealing that the interaction of mtDNA with the nuclear genetic background is under selective constraint [4],[5],[6]. This implies that mtDNA variants underlying population divergence have phenotypic consequences. Nevertheless, all the above mentioned studies analyzed complete haplotypes, thus masking the effects of single nucleotide changes within haplogroups.
How thus can one isolate the functional effect of specific mutations from their linked genetic background? Previously, we considered evolutionary sequence conservation to assess the functional importance of all common genetic variants in coding mtDNA, following the logic, that the more conserved the nucleotide position among species, the higher the functional importance [7],[8],[9]. This approach revealed that certain mutations that define human mtDNA lineages alter nucleotide positions to an equivalent degree of conservation as disease-causing mutations. Although predicting functional potential for coding region mutations, this observation cannot currently be tested experimentally, largely due to technical difficulties hampering site-directed mutagenesis in human mtDNA. This obstacle could, however, be partially overcome by cloning mtDNA-encoded genes, changing their genetic code so as to permit their translation on cytoplasmic ribosomes and then re-direct the nascent protein products to the mitochondria by exploiting introduced mitochondrial targeting sequences [10],[11],[12]. The success of this approach is limited to only certain mtDNA-encoded proteins. By contrast, allotopic expression of some mtDNA-encoded proteins was toxic to cells [11]. As such, this approach was not used to study the functionality of common variants. Moreover, haplogroup-defining mtDNA variants are mapped to protein and RNA-coding genes, as well as to non-coding regions, thus calling for alternative approaches to assess the functionality of non-coding mtDNA haplogroup-defining mutations.
In contrast to the mtDNA coding region, most mitochondrial regulatory elements (excluding the three conserved sequence blocks, CSBs) are subjected to many back-mutations [13],[14]. Moreover, sequences of mtDNA regulatory elements are much more flexible over the evolutionary time scale. Nevertheless, alterations of certain nucleotide positions with relatively poor evolutionary conservation were associated with phenotypes [15] suggesting that attributes other than evolutionary conservation should be considered to assess the functionality of mtDNA control region changes.
We hypothesized that certain sequence elements across the mitochondrial control region, not necessarily those with high inter-species conservation, are subjected to natural selection within populations. To test this hypothesis we chose to assess the functional potential of genetic variants in the mtDNA control region within sites harboring accepted functionality. mtDNA harbors sequence motifs that are recognized by members of the mtDNA transcription and replication machineries, all of which are encoded by the nuclear genome [16]. Such proteins bind particular sequences in the mtDNA promoters and origins of replication (Figure 1). Specifically, mitochondrial transcription factor A (TFAM) binds 4 mtDNA sites, two of which lay within the promoter region and two which lay down-stream to the L-strand promoter. The OXBOX-REBOX transcription factor binds a motif in CSB3 [14], while mitochondrial transcription termination factor (MTERF) recognizes a motif within tRNA-Leu (nucleotide positions 3232–3256) [17]. The termination-associated sequence (TAS) is bound by a specific protein that has yet to be isolated [18] and, finally, mtDNA origin of the light strand (Ori-L) is bound by specific factors (Joseph Shlomai, personal communication).
Here, we demonstrate that some single base changes that were fixed within TFAM binding sites during human evolution alter protein binding efficiency. One of these changes not only altered TFAM binding efficiency but also increased the efficiency of in vitro transcription by ∼2.5 fold and was associated with more than 2 fold increase in mtDNA copy numbers in mtDNA-less cell lines (rho-0 cells) repopulated with the mtDNA from haplogroup J. The interpretations of these results within the framework of understanding the implications of the phenotypic effects of haplogroup J are discussed.
We have screened more than 2500 whole mtDNA sequences from various human populations worldwide (http://www.genpat.uu.se/mtDB/) for natural variants in experimentally established protein binding sites and regulatory regions in the mtDNA control region (Table 1). These include recognition sites for mitochondrial transcription factor A (TFAM) and transcription termination factor (mTERF), the conserved sequence blocks (CSBs), and mitochondrial recognition sites for proteins of the replication machinery (TAS and origin of replication of the light strand - OL), comprising a total of 241 bp in each mtDNA. This screen identified that 77/241 of the total screened sites harbored polymorphic point mutations which can be divided into non-fixed changes (57/77, 74%) and haplogroups/sub-haplogroup-defining changes (i.e. population-fixed changes), comprising 20/77 (26%) of the variable nucleotide positions.
To assess the evolutionary conservation of the tested control region elements we have aligned the mitochondrial DNA (mtDNA) D-loop sequences of primates and other available mammalian sequences in each such sites (see Materials and Methods). Only four protein-binding sites described in human mtDNA gave trustworthy alignment (i.e. CSB1, CSB3, MTERF and OriL) and were, therefore, termed ‘highly conserved sites’ (HSS). The rest of our studied sites were either aligned only among humans and the great apes (i.e. chimpanzee, gorilla and orangutan) or within some primates, and were therefore regarded as ‘low conserved sites’ (LSS). It is important to note that CSB2 was excluded from the analysis as it is mapped to human mtDNA nucleotide positions 299–313 which mainly comprise a cytosine-tract prone to frequent insertions and deletions in humans, the frequency of which is hard to compute. We next documented the minimum number of changes that had occurred at each nucleotide position during mammalian evolution. Accordingly, nucleotide positions at HSS sites (n = 100) were divided into highly conserved positions, i.e. completely conserved positions (n = 54) and changeable positions (n = 46). While screening for human variants, we noticed a highly significant under-representation of variants within the highly conserved nucleotide positions (4/54), as compared to the evolutionary changed positions (16/46) (Fisher exact test, p = 0.0009). Thus, the more conserved the nucleotide position the less variable it is.
The LSS sites also showed a pattern of variability, with most nucleotides being invariable and most of the un-conserved nucleotide positions undergoing small number of changes in humans (Table 1). We believe that this pattern is, in fact, a reflection of variability in the functional importance of LSS positions, i.e. some nucleotide positions in these sites are functionally more important than others.
As the first step towards experimentally assessing the functional potential of haplogroup-defining mutations (fixed mutations), we employed Electrophoresis Mobility Shift Analysis (EMSA). Specifically, we tested for the effect of transition variants defining Caucasian haplogroup J (C295T) and a sub-haplogroup of J, J1b2 (C242T), on TFAM binding. These variants occur within two of the four TFAM-binding sites previously mapped by footprinting human mtDNA [19],[20]. EMSA analysis of double-stranded oligonucleotides spanning each of the two different TFAM binding sites, revealed increased TFAM binding to the C295T variant (Figure 2), but no significant difference with C242T variant (data not shown). In the case of the C295T variant, this increase in binding was observed at multiple different TFAM to DNA ratios (10/1 and 25/1 molar ratios are shown in Figure 2). These results raised the question of whether the C295T variant could affect mtDNA transcription. To test for this possibility, an mtDNA fragment spanning nucleotide positions 184–628 was PCR amplified using DNA samples corresponding to haplogroup J (harboring the C295T variant), haplogroup J1b2 (harboring both the C295T and C242T variants) and haplogroup H (harboring neither of these variants) as templates. The resulting PCR products were cloned and then utilized as templates in in vitro mitochondrial run-off transcription assays using partially purified mitochondrial lysates as a source of POLRMT and transcription factors [21]. The efficiency of light-strand promoter-driven transcription was ∼2.5 fold greater from the two haplogroup J templates (J, J1b2) than that of the haplogroup H fragment (H) (Figure 3), providing direct evidence for the functional potential of naturally occurring genetic variants in the human mtDNA regulatory region. Notably, the J1b2 construct showed increased transcription activity, although its additional mutation (C242T) did not significantly affect TFAM binding (data not shown), suggesting that the more functional variant is C295T that defines haplogroup J as a whole. Interestingly, the transcription assay produced two major products, namely a ∼230-nt, full-length transcript and a shorter ∼160-nt transcript that maps to just upstream of CSB1.
We reasoned that our in vitro transcription results may not only reflect an effect of haplogroup J mutations on mtDNA transcription but also on replication, as these two processes are coupled in the mitochondria (reviewed in: [16]). Thus, in order to provide more clues to the physiological importance of our in vitro transcription findings, we assessed the effects of haplogroup J mutations on both mtDNA transcript levels (Figure 4) and mtDNA copy numbers (Figure 5) by real time PCR. To reduce variance in our measurements due to the expected effect of nuclear genetic factors, we performed our experiments in cytoplasmic hybrids (cybrids) carrying mtDNAs of haplogroup J (2 independent cybrids) or haplogroup H (5 independent cybrids). Two genes (i.e. ND1 and ND4L) exhibited reduced steady state levels in only one of the haplogroup J cybrids (designated 3861), whereas the other H- and L-strand transcripts showed no significance difference among the tested cybrids (Figure 4). Since steady state levels of mRNA reflect a combined effect of transcription and post-transcriptional regulation and since ND1 and ND4L are co-transcribed along with 10 H- strand genes (namely ND2-ND5, CO1-3, ATP6,8), the levels of which did not differ, we interpreted the lower levels of ND1 and ND4L in the 3861 cybrid as reflecting a post-transcriptional effect specific to this cybrid.
However, when the mtDNA copy number was assessed in the panel of cybrids, both haplogroup J cybrids had more than twice as many mtDNAs than did any of the haplogroup H cybrids (Figure 5). Since all cybrids were generated using the same Rho-0 cells (Wal-2A) and were grown under the same conditions, differences in the mtDNA copy numbers should reflect differences in the replication capacity of the different mtDNA haplotypes.
Our results reveal that some genetic variants that became fixed during human evolution affect protein binding to mtDNA. Moreover, a point variant in a TFAM binding-site (C295T) increased in vitro transcription ∼2.5 fold. These findings support the view that some evolutionarily fixed mtDNA variants lead to functional consequences. The C295T TFAM binding-site variant did not clearly align with other mammalian sequences, other than those of the great apes (humans, chimpanzees, gorillas and orangutans). Nevertheless, the C295T nucleotide position underwent only a single fixation event during human phylogeny in the branch leading to haplogroup J, suggesting that, although poorly conserved, the low human variability at this nucleotide position is in line with its functional potential. We thus suggest that the degree of variability of a nucleotide position within humans may serve as a clue for functionality. To experimentally test for the generality of this prediction the functionality of additional variable nucleotide positions should be tested.
Four TFAM-binding sites have been mapped in the human mitochondrial genome by DNA footprinting, two of which are important for transcription and are located between the heavy and light strand promoters. The importance of the other two binding sites, localized downstream to the light strand promoter is not known (Figure 1). The altered TFAM binding, as well as the altered in vitro transcription capacity of the C295T variant occurring in one of the downstream TFAM-binding sites, support the role of this site in mitochondrial transcription. To our knowledge, this is the first indication that TFAM binding downstream of the mtDNA promoter might influence transcription, possibly through a more complex configuration of the promoter or by modulating mtDNA wrapping by TFAM, as previously suggested [16].
The tested variants define known mtDNA haplogroups, namely Caucasian haplogroup J and a sub-group of haplogroup J (J1b2). Haplogroup J was previously associated with several complex phenotypes such as Parkinson's disease and longevity (reviewed in: [1] as well as with the tendency of type II diabetes patients to develop complications [22]. In addition, haplogroup J increases the penetrance of some mtDNA mutations causing the eye disorder, LHON (Leber Hereditary Optic Neuropathy). Since the C295T variant defines haplogroup J and was not fixed in any other human mtDNA lineage, it is possible that the observed functional effect can be partially attributed to its phenotypic effect. Since other mtDNA haplogroups have been associated with various health conditions, further analysis of haplogroup-defining variants will be revealing with regard to their impact on mitochondrial function and their potential health consequences.
In addition to the ∼230-nt, full-length transcription product expected in the run-off transcription assay, we observed a shorter ∼160-nt transcript (Figure 3) that extends to between CSB2 and CSB1, but closer to CSB1. The nature of this product is unclear, but it could be due to pausing or premature termination of transcription, processing of the transcript by RNAse [23] or due to a physiologically irrelevant nuclease activity in the extract used. Less than full-length products have been observed previously in human mitochondrial in vitro transcription assays using recombinant transcription components [24], however the main products observed in that study cluster near CSB2 not CSB1. Since the region surrounding CSB1 and CSB2 has been implicated in various aspects of transcriptional priming of mtDNA replication [24],[25],[26], and given the involvement of TFAM in mitochondrial transcription and replication [16], it is quite possible that, in addition to its effects on transcription documented herein, the C295T variant also could impact mtDNA replication efficiency. Using T7-based in vitro transcription experiment, a length polymorphism in mtDNA positions 303–315 was suggested to affect human mitochondrial transcription [27]. The elevated mtDNA copy number in haplogroups J versus H cybrids support the possibly increased replication capacity of haplogroups J. Since haplogroup J was associated with successful longevity in some European populations, and since mtDNA copy numbers are reduced over age, one can speculate that the elevated mtDNA copy numbers in haplogroup J may be one of the factors mediating the phenotypic effects observed for this haplogroup.
The increase in mtDNA copy number but not transcript levels in haplogroup J, as opposed to haplogroup H cybrids, raise the question of why differences in transcript levels were only observed cell-free in vitro transcription. Measurements of mtDNA transcript levels by real time PCR reflect the steady state level of mRNA products, which is affected by both transcriptional and post-transcriptional regulation. Therefore, it is still possible that direct measurement of transcription level in organello will reflect in vitro transcription differences between haplogroups J and H. Alternatively, it has recently been shown that increased copy numbers do not necessarily lead to increased transcription [28], suggesting that our in vitro transcription results reflect differences in replication rather than transcription, as discussed above.
Our experiments demonstrate the effects of specific point mutations in the mtDNA in an in vitro transcription assay. A major obstacle interfering with the analysis of specific mtDNA mutations is the current inability to perform site-directed mutagenesis within mtDNA. This could be partially overcome by mitochondrial import of mtDNA-encoded genes genetically engineering to be coded according to the cytoplasmic genetic code. Such re-coded mtDNA genes can be generated with or without the desired mutation. Indeed, this approach was successfully adopted for some mtDNA-encoded proteins (e.g. ATP6, ATP8, ND4 and ND2) but less successful with most proteins, implying limitations for this approach [10],[11],[12]. Such experiments are further complicated by the need to express the desired mutated mtDNA gene in cells not expressing the endogenous gene, a scenario currently available solely with natural disease-associated mutants. In contrast to coding region mutations, our experiments demonstrate that the effects of non-coding mtDNA mutations can be tested individually in cell-free systems.
In summary, our analysis sheds light on the functional effect of naturally occurring variants in the mtDNA control region and suggests that such variants contribute to the phenotypic impact of mitochondrial DNA haplogroups.
DNA from healthy individuals previously assigned to haplogroups H, J1 and J1b2 [29], one sample of each, was used as a template for PCR amplification of an mtDNA fragment encompassing nucleotide positions 184–628 (using a forward primer: 5′GGCGAACATACTTACTAAAGTG 3′ and a reverse primer: 5′GCCCGTCTAAACATTTTCAG 3′). The PCR products were purified using the Wizard SV Gel and PCR Clean-up system (Promega) according to the manufacturer's protocol and then separately cloned into the pGEM T-vector (Invitrogen) using a ligation kit (Roche). The ligated products were transformed into E. coli strain DH5α by heat shock and ampicilin-resistant clones were isolated. Plasmid DNA harboring each of the cloned fragments was isolated and served as templates in an in vitro mitochondrial run-off transcription assay (see below).
Oligonucleotides corresponding to wild-type or mutant TFAM binding sites (Table 2) were annealed, followed by 5′ end labeling with Polynucleotide Kinase (NEB). Labeled DNA probes (3 nM) were incubated with varying amounts of TFAM protein (75 nM and 30 nM) for 30 minutes at room temperature in 10 mM Tris-HCL (pH 7.5), 10 mM KCl, 1 mM DTT, 1 mM EDTA, and 6% Glycerol. Free and TFAM-bound probes were then immediately separated on a 5% non-denaturing polyacrylamide gel at 150 V for 75 minutes in 0.5× TBE at room temperature. Gels were then dried, exposed to x-ray film and bands quantified using Quantity One 4.5.0 (BioRad) imaging software.
Mitochondrial run-off transcription reactions were performed [21] using different mtDNA promoter-containing templates. PCR products corresponding to nucleotides 184–628 of human mtDNA and harboring mutations defining either haplogroup J, J1b2 or the Cambridge Reference Sequence (CRS) [30] were cloned into the plasmid pGEMT-EZ (Promega). Digestion of this plasmid with EcoR1 resulted in the formation of a linear transcription template from which specific initiation from the LSP promoter resulted in transcripts 223 nucleotides in length. Individual reaction mixtures (25 µl) contained 36 µg of EcoRI-digested template, 10 mM Tris-Cl, pH 8.0, 10 mM MgCl2, 1 mM dithiothreitol, 100 µg/ml bovine serum albumin, 400 µM ATP, 150 µM CTP and GTP, 10 µM UTP, 0.2 µM α P32UTP (3,000 Ci/mmol), 40 units of RNase OUTTM (Invitrogen) and 2.5 µl of a transcription-competent mitochondrial extract from HeLa cells that was prepared as described previously [21]. After 30 min at 30°C, reactions were stopped by adding 200 µl of stop buffer (10 mM Tris-Cl, pH 8.0, 0.2 M NaCl, and 1 mM EDTA). Samples were treated with 0.5% SDS and 100 µg/ml proteinase K for 45 min at 42°C and precipitated by adding 0.6 ml of ice-cold ethanol and 1 µg of yeast tRNA (Sigma). The resulting RNA pellets were dissolved in 20 µl of gel-loading buffer (98% formamide, 10 mM EDTA, pH 8.0, 0.025% xylene cyanol, 0.025% bromphenol blue), heated to 95°C for 5 min, and then separated on 6% polyacrylamide/7 M urea gels in 1× TBE buffer. Radiolabeled 10-bp ladder DNA (Invitrogen) was run in parallel as a marker to estimate RNA transcript sizes. Gels were dried and exposed to x-ray film at −80°C. The experiments were repeated 3 times.
Human genetic variants were screened within experimentally established protein binding sites and regulatory regions in the mtDNA control region (Table 1). Specific nucleotide positions of the variants were retrieved from the “Human Mitochondrial Genome Database” (mtDB) (www.genpat.uu.se/mtDB), which includes more than 2500 whole mtDNA sequences, representing human populations worldwide. To assign control region variants to specific branches in the human mtDNA phylogeny, sequences containing genetic variants in the screened regulatory elements were compared to the revised Cambridge Reference Sequence [30]. Coding region variants in each of the retrieved sequences were used to screen the global human mtDNA phylogenetic tree that harbors worldwide human nucleotide variations, classified into haplogroups and sub-haplogroups [31]. Since mtDNA is transferred only through the maternal lineage and hence recombination has (virtually) no effect, this procedure enabled us to provide haplogroup assignments for mtDNAs harboring specific control region variants of interest that were linked to haplogroup-defining coding region mutations.
We used a phylogenetic approach relying on haplogroup classification to determine the number of times that a mutation event at a given nucleotide position had occurred during human evolution. Such evaluation assumed that a group of sequences belonging to the same phylogenetic branch (i.e. lineage or haplogroup) represent one mutation event during human evolution. For fixed mutations, we set a cut-off value. Changes that were shared in 5 or more different mtDNA sequences belonging to the same phylogenetic branch were considered fixed. The reason for this cutoff value was, first, to avoid relatively recent fixation events that have yet to be statistically established as the number of whole mtDNA sequences increases. Secondly, branches harboring 2–4 sequences gave relatively low bootstrap values in the global mtDNA phylogenetic tree (data not shown). The conservation of mitochondrial regulatory elements in mammals was estimated using a multiple sequence alignment program [(MEME), http://meme.sdsc.edu/meme/intro.html]. The number of times that a mutation event occurred during the mammalian evolution was assessed using available phylogenetic trees (one for mammals and another tree for primates) [32],[33], assuming that mutations shared by members of the same tree branch had only occurred once in the ancestor of this branch.
The Wal2A rho zero nuclear donor cell line, which had been cured of its resident mtDNAs with ethidium bromide treatment [34], was fused to chemically enucleated lymphoblasts containing either haplogroup J or H mtDNA [35]. Donor lymphoblasts,1.5×106 , were pre-treated with 0.5 µg/ml actinomycin D for 20 hours, the cells combined with 1.5×106 Wal2A rho zero cells, and fused using a pH-balanced PEG-1450 mixture for 60 seconds. Cells were allowed a 5 day post-fusion recovery period in rho zero growth medium and then selected in a 1 µg/ml 6-thioguanine-supplemented DMEM medium containing 10% dialyzed serum solution for selection. Media were replaced every 4 days or when the media looked depleted until such time as the cells achieved normal growth rates, as determined by periodic cell counts. The cybrids were haplotyped post-fusion to determine successful transfer of the mtDNA. The presence of Wal2a nuclear background was confirmed by microsatellite testing and comparison to the original lymphoblast and Wal2A microsatellite identities. The haplogroup assignment of the cybrid clones was determined using haplogroup specific polymorphisms and by sequencing of the mtDNA control region [31].
mtDNA transcript levels were determined for each of the cybrids by PCR incorporation of SYBR Green dye into real time PCR products. The rate of dye incorporation was monitored using the Roche LightCycler 480 real-time PCR system. Target genes were normalized according to three reference gene products, namely GAPDH, β-Actin and β2-Microglobulin [36],[37]. The geometric mean of the reference genes CTs was used for normalization of transcript expression levels. Primers for the real time PCR reactions were designed using the Primer 3 software (Table 3). Four micro-liters of 5× SYBR Green PCR Master Mix (Roche) was mixed with 20 ng purified total RNA, 0.4 µM of gene-specific primers and ultra pure H2O in a 20 µl reaction volume within 384-well plates (Genesee Scientific and Roche). For amplification, reaction mixtures were incubated for 2 min at 50°C and 10 min at 95°C, followed by 45 cycles of three-steps consisting of 15 seconds at 95°C, 20 seconds at 60°C and 15 seconds at 72°C. Following the PCR step, samples were heated from 60°C to 95°C with a ramp time of 20 min to construct dissociation curves and verify that single PCR products were obtained. PCR products were also analyzed by gel electrophoresis to confirm primer specificity. Serial cDNA dilutions were used for primer validation experiments to demonstrate that both target and reference genes had equal amplification efficiencies, according to the standard curve method. The comparative CT method was used for relative quantification of gene expression as described by the real time PCR machine manual. Differences in the CT values (dCT) of the transcript of interest and reference genes were used to determine the relative expression of the gene in each sample. The dCT method was used to calculate fold expression. Experiments were carried out in triplicate for each data point. Roche analysis software versions 2.1 (Applied Biosystems) and Microsoft Excel were used for data analysis.
mtDNA copy number calculations were performed by comparing the ratio of the mtDNA-encoded ND2 gene levels to the nuclear DNA-encoded 18S rRNA gene using real time quantitative PCR amplification (Roche LC480) in three independent experiments per tested DNA. To quantify each gene product, serial 10× dilutions of known cloned PCR fragments were used to create a standard curve. CT values of the ND2 and 18S rRNA genes were plotted against this standard curve to provide an absolute copy number level for each gene. The ND2 to 18S rRNA ratios were then calculated for each of the cybrid DNAs. Primers used for amplification of the ND2 and 18S rRNA genes are listed in Table 3.
|
10.1371/journal.pmed.1002859 | Mild-to-moderate renal pelvis dilatation identified during pregnancy and hospital admissions in childhood: An electronic birth cohort study in Wales, UK | Chronic kidney disease (CKD) is a growing contributor to the global burden of noncommunicable diseases. Early diagnosis and treatment can reduce the severity of kidney damage and the need for dialysis or transplantation. It is not known whether mild-to-moderate renal pelvis dilatation (RPD) identified at 18–20 weeks gestation is an early indicator of renal pathology. The aim of this follow-up to the Welsh Study of Mothers and Babies was to assess the risk of hospital admission in children with mild-to-moderate antenatal RPD compared with children without this finding. We also examined how the natural history of the RPD (whether the dilatation persists in later pregnancy or postpartum) or its characteristics (unilateral versus bilateral) changed the risk of hospital admission.
This population-based cohort study included singleton babies born in Wales between January 1, 2009, and December 31, 2011 (n = 22,045). We linked ultrasound scan data to routinely available data on hospital admissions from the Patient Episode Database for Wales (PEDW). The outcome was a hospital admission for urinary tract causes (defined by an expert study steering group) in the first three years of life. We used Cox regression to model time to first hospital admission, according to whether there was evidence of RPD at the fetal anomaly scan (FAS) and/or evidence of dilatation in later investigations, adjusting for other predictors of admission. We used multiple imputation with chained equations to impute values for missing data. We included 21,239 children in the analysis. The risk of at least one hospital admission was seven times greater in those with RPD (n = 138) compared with those without (n = 21,101, conditional hazard ratio [cHR] 7.23, 95% confidence interval [CI] 4.31–12.15, p < 0.001). The risk of hospital admission was higher in children with RPD at the FAS and later dilatation (cHR 25.13, 95% CI 13.26–47.64, p < 0.001) and in children without RPD at the FAS who had later dilatation (cHR 62.06, 95% CI 41.10–93.71, p < 0.001) than in children without RPD (n = 21,057). Among children with RPD at the FAS but no dilatation in later pregnancy or postpartum, we did not find an association with hospital admissions (cHR 2.16, 95% CI 0.69–6.75, p = 0.185), except when the initial dilatation was bilateral (cHR 4.77, 95% CI 1.17–19.47, p = 0.029). Limitations of the study include small numbers in subgroups (meaning that these results should be interpreted with caution), that less severe outcomes (such as urinary tract infections [UTIs] managed in the community or in outpatients) could not be included in our analysis, and that obtaining records of radiological investigations later in pregnancy and postpartum was challenging. Our conclusions were consistent after conducting sensitivity analyses to account for some of these limitations.
In this large population-based study, children with RPD at the FAS had higher rates of hospital admissions when there was persistent dilatation in later pregnancy or postpartum. Our results can be used to improve counselling of parents and develop care pathways for antenatal screening programmes, including protocols for reporting and further investigation of RPD.
| Pregnant women usually have an ultrasound scan when they are about 20 weeks pregnant. The purpose of this anomaly scan is to look for structural abnormalities, such as abnormal growth or development of organs.
It is not known if some of the findings at this scan are indicators of health problems.
For example, fluid-filled areas in the baby’s kidneys (called ‘renal pelvis dilatation’) are sometimes seen, but it is not known whether these are associated with adverse outcomes in childhood.
In this study, children with renal pelvis dilatation, and a group of children who did not have this finding, were followed until they were three years old.
Babies who had this dilatation were more likely to be admitted to hospital with problems in the kidneys and urinary tract than children without dilatation, especially if the finding was still present later in pregnancy or after the birth (HR 25.13, 95% CI 13.26–47.64, p < 0.001). In children with dilatation at the anomaly scan, but no dilatation in later pregnancy or postpartum, we did not find an association with hospital admissions (HR 2.16, 95% CI 0.69–6.75, p = 0.185), except when the initial dilatation was in both kidneys (HR 4.77, 95% CI 1.17–19.47, p = 0.029).
These results can be used to improve counselling of parents and to develop care pathways for antenatal screening programmes, including protocols for reporting and further investigation of renal pelvis dilatation.
| Chronic kidney disease (CKD) is a growing contributor to the global burden of noncommunicable diseases [1,2]. Although relatively rare in children [1], management of CKD in paediatric patients is complex and costly [1,2], and the condition has a profound impact on children and their families [3]. Approximately half of all cases of CKD in children in high income countries are due to congenital anomalies of the kidney and urinary tract (CAKUT) [4,5]. Early diagnosis and treatment can reduce the severity of kidney damage and the need for dialysis or transplantation [6,7].
Routine investigations during pregnancy are an opportunity to screen for CAKUT. In the United Kingdom, all women are offered an ultrasound scan at 18 to 20 weeks gestation to detect major anomalies in the fetus. The National Health Service (NHS)’s Fetal Anomaly Screening Programme specifies that 11 structural abnormalities are detectable at this scan, including one CAKUT (bilateral renal agenesis) [8], but other abnormalities of the kidney and urinary tract can also be identified [4,7]. One such finding is dilatation of the fetal renal pelvis (the collecting system where urine flows from the kidney into the ureter), detected by measuring the anterior-posterior (AP) diameter of the renal pelvis. This is also known as hydronephrosis, pelvicalyceal dilatation, pelviectasis, or pyelectasis [9]. Current UK guidance recommends that this finding should be reported and assessed further [8,10]. However, different classification systems exist, which use different criteria and nomenclature to distinguish between potential pathological dilatation and transient changes that are of limited clinical significance [6,11,12], leading to inconsistent management and parental anxiety [13,14].
Measurement thresholds based on ‘best available evidence…[of] prognostic information’ have been proposed [6], with AP dilatation in the second trimester of 4- to <7-mm classified as mild, 7- to ≤10-mm as moderate, and >10-mm as severe. Third trimester AP thresholds of 7- to <9-mm are classified as mild, 9- to ≤15-mm as moderate, and >15-mm as severe. Severe dilatation is reported to be associated with postnatal pathology in almost 90% of cases [15]. However, previous studies examining the sequelae of mild or moderate antenatal dilatation have found conflicting results, have several methodological limitations, and have been assessed to be of low or moderate quality [15–21]. These have not therefore led to the development of consistent care pathways [12,22,23].
The Welsh Study of Mothers and Babies is a prospective, population-based cohort that was established to examine childhood morbidity associated with ultrasound findings of unknown significance detected at the fetal anomaly scan (FAS). We analysed data from this cohort with the aim of assessing the risk of hospital admission in the first three years of life associated with mild-to-moderate antenatal renal pelvis dilatation (RPD). We also examined how the natural history of the RPD (whether the dilatation persists in later pregnancy or postpartum) or its characteristics (unilateral versus bilateral) changed the risk of hospital admission.
The Welsh Study of Mothers and Babies recruited a cohort of pregnant women receiving antenatal care in Wales between 2008 and 2011, to estimate the prevalence of seven nonstructural findings at the FAS and examine their association with pregnancy outcomes and longer-term health outcomes [24]. Ethical approval for the original study was given by the Multicentre Research Ethics Committee for Wales (reference 08/MRE09/17) on April 16, 2008.
All pregnant women who had a second trimester FAS in six of seven Welsh Health Boards between July 2008 and March 2011 were eligible for inclusion. At recruitment, women were asked to give written consent that the data from their ultrasound scan could be linked with routinely collected data on their child.
The population for this analysis was singleton children who were live-born between January 1, 2009, and December 31, 2011, to mothers in Wales who consented to take part in the study and for whom validated FAS data were available (Fig 1). Pregnancies with an unknown outcome (for example, because the birth happened outside of Wales) were excluded. Children whose information could not be assigned an anonymised linking field (ALF; for example, because they did not receive their healthcare in Wales or did not have a valid NHS number or other identification variables) were also excluded, as linkage with the healthcare datasets was not possible. Children were followed from birth until the occurrence of death, migration out of Wales, 3rd birthday, or December 31, 2014 (end of follow-up). Person-time was censored in cases of death and migration.
RPD was defined as fluid-filled dilatation of the renal pelvis measured on the axial section, with an AP diameter of 5.0 to 9.9 mm (with the callipers to be placed on the inner AP margins of the renal pelvic wall) at the FAS [25]. An additional reporting screen was added to the information system for radiological data storage and reporting in Wales (Radiology Information Service 2 [RadIS2]) to capture the scan data. Data were also collected on whether the dilatation was unilateral or bilateral. An expert quality assurance (QA) panel reviewed FAS images to confirm the presence of dilatation in accordance with the study definition; the number of cases of RPD reduced from 221 to 144 after this process (for detail, see [26]).
We also sought to obtain information from radiological investigations conducted later in pregnancy from RadIS2 and the Congenital Anomaly Register for Wales (CARIS). Data on the presence of dilatation identified later in pregnancy and/or up to 12 months postpartum, plus any measurements recorded, were extracted from radiological reports. In accordance with international guidance [6,12], measurements of 7.1 mm or greater later in pregnancy or postpartum were considered evidence of later dilatation. The exposure groups in the analysis were then stratified into (i) no RPD at FAS or later, (ii) no RPD at FAS but RPD present later, (iii) RPD at FAS but not later, and (iv) RPD at FAS and later (Fig 2).
The outcome was a hospital admission for urinary tract causes in the first three years of life. An admission was defined as a continuous stay using a hospital bed provided by the NHS in Wales under one or more consultants, and included transfers between hospitals. A list of codes for possible causes of admissions in children for which RPD could be considered a possible precursor (including renal and lower urinary tract problems) was agreed by the study steering group, which included a consultant paediatric nephrologist, consultant radiologists, and an academic general practitioner (see S1 Table). The list was based on the International Statistical Classification of Diseases and Related Health Problems, 10th Revision (ICD-10) [27] and the procedure code list used in the NHS (Operating Procedure Codes Classification of Interventions and Procedures version 4 or OPCS-4) [28]. Hospital admissions with any of these codes in any coding position were identified from the Patient Episode Database for Wales (PEDW). Admissions as a day case for postnatal investigations alone are not a part of this dataset, and these admissions would not therefore have been included.
Data from the Welsh Study of Mothers and Babies were exported to the Secure Anonymised Information Linkage (SAIL) Databank [29] to enable the radiological data to be linked with data on hospital admissions (from PEDW), congenital anomalies (from CARIS), deaths (from the Office for National Statistics Annual District Death Extract), and migration (from the Welsh Demographic Service data). For each of these datasets, individuals were assigned a unique identifier (the ALF) provided by the NHS Wales Informatics Service. The linkage system uses a combination of deterministic (based on NHS numbers) and probabilistic record linkage (based on first name, surname, date of birth, gender, and phonex and soundex versions of names); this linkage is more than 99.85% accurate [30]. Second-stage encryption is used by the data bank before storing data, and third-stage encryption is used to create project-specific linked datasets. Approval from the Information Governance Review Panel of the SAIL Databank was obtained for the analysis.
Preliminary analyses of PEDW data showed that 2 per 100 children were admitted to hospital with a urinary tract cause before the age of five years. We therefore estimated that we would need approximately 21,000 children without RPD and 140 children with RPD to detect, at a 5% type 1 error rate, a 3-fold increase in urinary tract hospital admissions with RPD with 80% power. There were enough children in our cohort for this analysis.
The statistical analysis was planned by the investigators of the study, in collaboration with the study steering group, in advance of the conduct of the analysis (for detail, see [24] and the protocol for this analysis, S1 Renal Study Protocol). This included the specification of the ICD-10 codes to be included in the definition of the outcome. An initial analysis of the association between RPD and hospitalisations was presented to the study steering group in 2016, and they recommended that the additional radiological data (from later pregnancy and postpartum) were sought and added to the analyses. We used Cox regression to model time to the first urinary tract hospital admission to three years of age. Our primary outcome was time to first hospital admission because most children with an admission were only admitted once (81.3%). We estimated hazard ratios (HRs) with 95% confidence intervals (CIs) to examine the risk of hospital admissions associated with the presence of RPD at the FAS, and then according to whether the child had RPD and/or later dilatation. The proportional hazards assumption was assessed graphically using log-minus-log plots and was tested based on the Schoenfeld residuals. We originally planned to include follow-up time until the child’s fifth birthday, but there was evidence that the proportional hazards assumption was violated in this analysis because all of the admissions in the RPD group occurred before the age of three years, whereas children without RPD continued to be admitted for the first time after three years of age. Based on this and on peer review comments, we chose a cutoff of three years of age for the follow-up period (instead of the preplanned five years), and formal testing confirmed there was no strong evidence that the proportional hazards assumption was violated in these models (for example, p = 0.11 for the binary RPD variable). Conclusions from models including five years of follow-up were similar to the models including three years of follow-up.
We examined associations in unadjusted models and conditional on other predictors of hospital admissions (sex, maternal age in three categories [<25, 25–34, 35+ years], deprivation quintile based on the UK Townsend Deprivation Score [31], and prematurity). We repeated the analyses to estimate HRs stratified according to whether the RPD was unilateral or bilateral. We also conducted sensitivity analyses: (i) adding information on dilatation identified during hospital admissions into the definition of exposure subgroups (that is, also using codes Q62.0, N13.0, N13.1, N13.2, or N13.3 in any position for a PEDW admission as evidence of later dilatation) and (ii) using an Anderson-Gill model to account for multiple urinary tract admissions during the follow-up period.
Data from radiological investigations to assess for dilatation later in pregnancy or postpartum were missing for 50 of the 138 children with RPD. There was also a low percentage of children with missing data on covariates (0.7% for Townsend score, 0.3% for prematurity). Multiple imputation with chained equations [32] was used to impute values for the missing data (10 imputations) under the missing at random assumption. The imputation model included all covariates, the outcome variable (urinary tract admissions), and the cumulative baseline hazard [33]. Conclusions from a complete case analysis and from the multiple imputation were similar. In response to peer-review comments, we present the results from the multiply imputed datasets.
All analyses were conducted within the SAIL Gateway using Stata version 15.1.
The study is reported as per the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guideline (see S1 STROBE Checklist).
Anomaly scan data were available for 22,045 children (Fig 1). The characteristics and pregnancy outcomes of their mothers were comparable to the general population of pregnant women in Wales [26]. Thirty-two pregnancies were excluded, as their scan images were not available for QA for RPD. There were 81 stillbirths, 50 spontaneous or induced abortions, 607 pregnancies with no outcome data, and 25 babies who could not be assigned an anonymised linkage field identifier, leaving 21,250 children available for analysis after data linkage. Eleven children had dilatation of the renal pelvis measuring 10.0 mm or greater at the FAS. All of these had evidence of a urinary tract hospital admission or significant renal pathology in investigations after birth and were excluded from all further analyses. This analysis is therefore based on 21,239 children (96.3% of those with scan data) who contributed 61,984 child-years of follow-up.
A total of 138 children (0.7%) had confirmed RPD at the FAS (Table 1). RPD was more prevalent in male than female children (0.9% compared with 0.4%) and in children of younger mothers (0.9% when maternal age was <25), but there was no association with area-level social deprivation, prematurity, or low birth weight. Overall, 1.8% of the children had at least one admission for a urinary tract problem during follow-up (Table 2). Girls were more likely than boys to have at least one admission, as were children of younger mothers and premature children.
Of the 21,101 children with no RPD at the FAS, 20,534 (97.3%) had no further investigations and 567 (2.7%) did (see Fig 2). Most of the 567 had no evidence of RPD at the later investigations (n = 523, 92.2%). In the group with no RPD at the FAS and no evidence of later dilatation or later investigations (n = 21,057), there were 333 children with at least one urinary tract hospital admission (1.6%). Most of these had only one admission (86.8%). The median age at first admission was 6 months (interquartile range [IQR] 2–14), and the commonest code linked to the first and all admissions was urinary tract infection (UTI; ICD-10 codes N39.0, N39.1, or P39.3).
A total of 26 of the 44 children (59.1%) with no RPD at the FAS but evidence of dilatation at later investigations had at least one urinary tract hospital admission, and most had more than one admission. The median age at first admission was three months (IQR 1–6), and the commonest code linked to the first and all admissions was hydronephrosis (ICD-10 code Q62.0).
Of the 138 children with RPD at the FAS, 88 (63.8%) had further investigations. A total of 59 of the 88 had no evidence of dilatation at the later investigations, and there were fewer than five hospital admissions in this group. No further investigations were found for 50 of the 138, and there were also fewer than five hospital admissions in this group. As described above, multiple imputation was used to account for the missing data on later dilatation in these children.
A total of 29 children with RPD at the FAS had evidence of later dilatation; 11 of these (37.9%) had at least one hospital admission for a urinary tract cause, and most had more than one admission. The median age at first admission was two months (IQR 0–15), and the commonest code linked to the first and all admissions was hydronephrosis (ICD-10 code Q62.0). Further details on the characteristics of first and all admissions for all groups are summarised in the supporting information (S2 Table and S3 Table).
The risk of at least one hospital admission was seven times greater in those with RPD (n = 21,101) compared with those without (n = 138, conditional HR [cHR] 7.23, 95% CI 4.31–12.15, p < 0.001, Table 3). Table 3 also shows the hazard ratios when the data were further stratified according to whether there was evidence of dilatation in later radiological investigations. Compared to those with no RPD and no later dilatation, the risk of urinary tract hospital admissions was higher in children with no RPD but dilatation detected later (cHR 62.06, 95% CI 41.10–93.71, p < 0.001). It was also higher in children with RPD and later dilatation (cHR 25.13, 95% CI 13.26–47.64, p < 0.001), but not in children with RPD and no later dilatation (cHR 2.16, 95% CI 0.69–6.75, p = 0.185). We also did not detect a difference at the 5% level in hospital admission rates between children with unilateral and bilateral RPD (Table 4). However, compared with children with no RPD and no later dilatation, there was evidence that children with bilateral RPD but no evidence of later dilatation had a higher risk of hospital admission than children with no RPD and no later dilatation (cHR 4.77, 95% CI 1.17–19.47, p = 0.029).
The results of sensitivity analyses are presented in the supporting information (S4 Table and S5 Table). When evidence of dilatation in the hospital admission records was included in the definition of the exposure subgroups, the HRs for children with evidence of later dilatation moved further away from one, whilst the estimate for children with RPD but no later dilatation moved closer to one. Accounting for multiple admissions increased the HRs for children with evidence of later dilatation, as these children were more likely to have multiple admissions, but did not change the conclusion for children with RPD but no later dilatation.
In this study, mild-to-moderate RPD was identified in 7.6 per 1,000 singleton pregnancies at the fetal anomaly ultrasound scan. In most children, this dilatation did not persist, and hospital admission rates for urinary tract problems in those children were similar to those in children with no RPD and no later dilatation. Persistent dilatation in later pregnancy and/or postpartum was rare, but children with this finding had a higher risk of hospital admission for a urinary tract cause before the age of three years. Children whose dilatation was identified later in pregnancy or postpartum had the highest risk of hospital admission.
This was a large population-based study in a cohort that was representative of all pregnant women in Wales. The stringent QA process was a strength of the study, with scan images reviewed by an expert panel to confirm that all cases of RPD conformed to the study definition. Linkage to routinely collected healthcare records was possible for 97% of women and children, ensuring that few were lost to follow-up. Unlike previous studies, we could compare outcomes in children with and without RPD. We used hospital admissions for urinary tract causes (including operations but not investigations) as a proxy for significant morbidity.
The prevalence of RPD in this study was similar to estimates in previous studies. However, the absolute number of cases was small (n = 138), and these numbers reduced further when the data were stratified by the presence or absence of later dilatation. The estimates from the study must therefore be interpreted with caution. Although the small numbers mean that the CIs around the point estimates are wide, the effect sizes for the associations examined are large, suggesting that the sample size for examining these associations is adequate.
The SAIL Databank only includes healthcare information from Wales, which means that we were unable to access information on pregnancy outcomes that occurred in facilities outside of Wales or on mothers and children who ordinarily received their healthcare outside of Wales (for example, because they live along the border with England). The distribution of pregnancy outcomes, congenital anomalies, and premature deliveries in the cohort included in our analysis was comparable with those in published data for Wales [26], suggesting that they were representative of the general obstetric population in Wales.
We used routinely available healthcare records on hospital admissions to capture data on outcomes in this study, as these are available for the whole population in Wales. They are also an indicator of clinically significant morbidity. However, we acknowledge that this means that we have not included less severe outcomes in our analysis, such as UTIs managed in the community or in outpatients, underestimating the total burden of these outcomes in the general population. It is also possible that provider practice may explain some of the increased risk in admissions seen with persistent dilatation, as they may be more likely to admit a child with a known kidney abnormality for treatment and observation. Without access to community or outpatient data, we cannot assess the effect of this potential bias on the estimates obtained.
Obtaining records of radiological investigations later in pregnancy and postpartum was challenging. Although guidelines state that all cases of RPD should receive follow-up investigations, we found later tests for only 88 of the 138 children with RPD (64%). This does not mean that they did not have later investigations, but that we were unable to find a record of these. Previous studies have restricted their samples to children with complete data on later pregnancy or postpartum scans (for example, see [34]), but this leads to an incomplete picture of the natural history of RPD, especially as the characteristics of the excluded sample are usually unreported. We were able to identify additional cases of ‘hydronephrosis’ from hospital admission codes and use this information in a sensitivity analysis to reclassify some children who had no data on later radiological investigations. This strengthened our conclusions, as admission rates for children without persistent dilatation reduced in this analysis.
When information from later radiological investigations was available, it was clear that the number, timing, and type of test was not consistent between children and that different terminology and measurement thresholds were used to identify dilatation in the radiological reports. Our findings suggest that regular training, clear reporting protocols, and frequent audits to monitor follow-up rates and maintain reporting standards are needed if the management of RPD is to be standardised.
Our results are consistent with findings from previous case series that have used repeated prenatal measurement of dilatation to predict which children require postnatal follow-up [35]. However, there were some additional questions that we were unable to answer in this study. For example, RPD was defined as an AP diameter of between 5.0 and 9.9 mm at the FAS, in accordance with the definition used by the antenatal screening programme in Wales when the data were collected [25]. Other classification systems [12] have further defined dilatation at the FAS as ‘mild’ (<7.0 mm) and ‘moderate’ (7.0 to ≤10 mm), and some screening programmes recommend that follow-up is only conducted when an RPD of ≥7.0 mm is identified at the FAS [8]. We could not assess whether there was a difference in hospital admissions between children with ‘mild’ or ‘moderate’ dilatation at the FAS, because the scan data collected for the study only contained information on whether RPD was present or not (but no measurements). Further studies are needed to document the natural history of dilatation and outcomes in children according to the actual measurement of the AP diameter at the FAS. We were also unable to draw firm conclusions about other predictors of poorer outcomes, such as unilateral compared with bilateral dilatation or the gender of the infant, because of sample size constraints. A larger study is needed to compare these subgroups. However, there was some evidence that children with bilateral RPD had a higher risk of hospital admissions even when there was no later evidence of dilatation, which is consistent with how bilateral RPD is currently managed in practice in Wales.
Two retrospective case series of patients known to specialist services have suggested that including other parameters (such as calyceal dilatation, renal parenchymal thickness, renal parenchymal appearance, bladder abnormalities, ureteral abnormalities, and oligohydramnios) in a formalised classification system is helpful in predicting which cases of dilatation will resolve spontaneously and which children require later surgery [36,37]. Further research is needed to fully understand the best combination of factors for predicting later pathology [9], and therefore allow for the development of care pathways with fewer follow-up visits or investigations for children identified as low risk [34]. Small studies have also examined different methods of visualising fetal urinary tract pathology (for example, conventional versus three-dimensional ultrasound) [38] and the use of serum and urinary biomarkers in the postpartum period to predict the risk of obstruction or impairment of renal function [39], but this research is currently inconclusive.
We also found that there were children who did not have RPD at the FAS (either 5.0 to 9.9 mm or ≥10 mm) who were found to have dilatation at later investigations. This is consistent with previous findings that not all cases of congenital dilatation are identified at the FAS [12]. In our study, this was a small group (44 of 21,101, 0.2%), but they had the highest rates of hospital admissions overall. As we did not have access to the full report from the FAS or later medical notes, we do not know why further investigations were conducted for these children. However, their hospital admission codes suggested that they had multiple congenital renal abnormalities and required more renal surgical procedures than children in the other groups. It is therefore likely that these children had other anomalies identified at the FAS, which led to the follow-up investigations and the repeated hospital admissions from an early age.
Clear protocols for reporting and further investigation and management of RPD are being developed (for example, [10]), and regular audits are needed to ensure that these are followed. The effectiveness of continuous prophylactic antibiotics in the prevention of UTI in children with prenatally detected RPD is unclear, with evidence based on observational studies. The first trials of prophylactic antibiotics for antenatal hydronephrosis with postnatally diagnosed vesicoureteral reflux are under way (ClinicalTrials.gov Identifier: NCT01140516). Postnatal infections still occur despite prophylaxis [40], and antibiotic resistance is higher with increased use [41]. Studies have found an increased risk of UTI with severe prenatal dilatation in females, uncircumcised males, and with specific postnatal diagnoses (ureteral dilatation, vesicoureteral reflux, or vesico-ureteric junction obstruction) [42,43]. We did not have access to labour ward, outpatient, or general practitioner records in this study and, as such, did not have information on which children were given antibiotics after birth. Although we are therefore unable to comment on the role of antibiotics in preventing UTI, the main cause of admission in our study was for UTIs and our results support the need for close postnatal follow-up of children with persistent dilatation.
RPD at the FAS is an important finding because it identifies fetuses who require later investigations. When these investigations are normal, we do not yet have sufficient evidence to tell whether there is an increase in hospital admissions for urinary tract problems in childhood. When there is persistent dilatation, hospital admission rates in childhood are higher. These results can be used to improve counselling of parents. Although this was a large study in a representative population of pregnant women in Wales, obtaining records of radiological investigations after the FAS was challenging, and scans were not conducted or reported consistently. Clear protocols for reporting and further investigation of RPD are being developed [10], and regular audits are needed to ensure that these are followed. Further studies should examine whether other characteristics at the FAS could improve the detection of renal pathology during antenatal screening.
|
10.1371/journal.pcbi.1003613 | Pierced Lasso Bundles Are a New Class of Knot-like Motifs | A four-helix bundle is a well-characterized motif often used as a target for designed pharmaceutical therapeutics and nutritional supplements. Recently, we discovered a new structural complexity within this motif created by a disulphide bridge in the long-chain helical bundle cytokine leptin. When oxidized, leptin contains a disulphide bridge creating a covalent-loop through which part of the polypeptide chain is threaded (as seen in knotted proteins). We explored whether other proteins contain a similar intriguing knot-like structure as in leptin and discovered 11 structurally homologous proteins in the PDB. We call this new helical family class the Pierced Lasso Bundle (PLB) and the knot-like threaded structural motif a Pierced Lasso (PL). In the current study, we use structure-based simulation to investigate the threading/folding mechanisms for all the PLBs along with three unthreaded homologs as the covalent loop (or lasso) in leptin is important in folding dynamics and activity. We find that the presence of a small covalent loop leads to a mechanism where structural elements slipknot to thread through the covalent loop. Larger loops use a piercing mechanism where the free terminal plugs through the covalent loop. Remarkably, the position of the loop as well as its size influences the native state dynamics, which can impact receptor binding and biological activity. This previously unrecognized complexity of knot-like proteins within the helical bundle family comprises a completely new class within the knot family, and the hidden complexity we unraveled in the PLBs is expected to be found in other protein structures outside the four-helix bundles. The insights gained here provide critical new elements for future investigation of this emerging class of proteins, where function and the energetic landscape can be controlled by hidden topology, and should be take into account in ab initio predictions of newly identified protein targets.
| We discovered a new class of helical bundle proteins with knot-like structures where part of the polypeptide chain is threaded through a covalently bound loop. We call this unique structural motif a Pierced Lasso Bundle. We discovered 12 structurally homologous proteins in the PDB. Our results indicate that the PLB topology is important for regulating the global native state dynamics, especially for a conserved helix that forms part of the canonical receptor interface surface. As a correctly folded structure is necessary condition for function, we explore kinetic folding data to the active conformation and observe two distinct mechanisms to pierce the lasso on route to the native structure, a slipknotting or a plugging event. Threaded elements predominantly slipknot through small covalent loops (50–63 amino acids) while structural components plug through large covalent loops (68–95 amino acids). This information is important for protein design as the loop length and threading mechanism affect dynamics important for function and ease of folding a native structure for potential therapeutic uses.
| The four-helix bundle is a common motif in nature [1], [2], [3], [4] often used as a target for designed pharmaceutical and nutritional biomolecules [5], [6], [7]. The cytokine subfamily is a family of four-helix bundles that are soluble proteins secreted from different organs/tissues. Cytokines act as chemical messengers important in intercellular communication. They regulate differentiation, proliferation, activation and death of many cell types, with particular involvement in the regulation of the circulatory system and production of immunity and inflammatory responses [8]. Most four-helix bundles also have conserved cysteines within the motif, whose disulphide bonds help maintain their structure and stability [2]. Every protein containing a disulphide bridge forms a covalently closed loop. When the N- and C-termini are covalently linked you have the simplest knotted topology in mathematics, termed a “zero knot” [9]. The “zero knot” is present in the cytokine Interleukin-36 [10], the θ-defensins as well as other lower organism circular proteins as reviewed in [11] (Figure 1, top left). Nonetheless, a true “zero knot” is rare in the case of proteins. More commonly, the covalent loop creates a “cinch” in the polypeptide chain within the central sequence and the N- and C-terminal ends extend from the internal covalent loop (Figure 1A, top right). Occasionally, either an N- or C-terminal cysteine residue participates in forming the closed loop to generate a “lasso-like” structure (Figure 1, bottom left). The size of the covalent loop depends on the sequence separation between the two cysteines forming the covalent loop. If the size of the loop is big enough, it is possible for part of the polypeptide chain to thread through and create what we term a Pierced Lasso (PL, Figure 1, bottom right).
Recently, we discovered complexity in leptin's fold created by a single disulphide bond [12] between residue C96 and the C-terminal cysteine (C146), which creates a lasso as described in Figure 2A. The folding complexity in leptin comes from threading a helical-hairpin through the closed covalent loop in order to reach the native fold [13], [14], [15], [16] to form a PL Bundle (PLB, Figure 1A). This threading is reminiscent of the more common knotted proteins, where a protein terminal must thread across a twisted loop [15], [16], [17]. In the case of leptin, the structure is analogous to slipknotted proteins, where part of the protein adopts a hairpin-like configuration that threads across the covalent loop (Figure 1A). A slipknotted polypeptide geometry (topology) adds folding complexity that was unrecognized until recently [13], [14], [15], [18], [19], [20], [21]. Since the PL in leptin is distinct from knotted/slipknotted proteins, where the protein backbone ties a knot, and from the cystine knot that is created by at least three disulphide bonds [22], [23], [24], [25], we called this new motif a Pierced Lasso Bundle (PLB, Figure 2A) [12]. This new class of proteins is distinct from previously classified cystine knotted proteins. In the PLB case, a closed covalent loop is created from a single disulphide bridge enclosing one of the terminals with one of the loops where part of the amino acid chain threads through and pierces the covalent loop. In the cystine knotted class, the added complexity beyond a closed loop is created by an additional side-chain mediated chemically cross-linked knot through the covalent loop [28].
PLBs, unlike cystine knots, are able to unfold their threaded elements. Furthermore, unlike knotted/slipknotted proteins, PLBs can modulate their complex topology based on the oxidation conditions of the disulphide bridge. Thus, breaking the bond/contact between the two cysteines also breaks the covalent loop and thereby unthreads the structure. Because of the exciting functional consequences these dynamics may have, we searched for other proteins containing a similar PLB topology. A comprehensive search of the Protein Data Bank (PDB) found 11 structures with a similar threaded motif. Leptin has many structurally homologous proteins where disulphide bridges create a covalent loop, but only 11 had a threaded element through the covalent loop. Interestingly, there is a difference between leptin and the other threaded structures in terms of the location of the closed loop. The covalent loop in leptin is located at the C-terminal end while all other structures, found to date, are knotted at the N-terminal end (Figure 2A and B).
The threaded structure of leptin influences the Native State Dynamics (NSD) and thus the biological activity [12]. Here, we explore the effects of a C- versus N-terminal pierced lasso as well as the folding and the NSD in the related structures. Additionally, we investigate the threading mechanism as the effect(s) of loop size. Structure Based Models (SBMs) were used to study the human and murine interleukin 3 and two zebra fish interferons (Figure 2B and C). Additionally, we compare the folding mechanism for three of the unthreaded four-helix bundles (the G-CSF, LIF and hGH, Figure 2B and C), which are members of the leptin family of long-chain helical cytokines. The results show that all PLB proteins stabilize the covalent loop as an initial step in folding (independent of an N- or C-terminal lasso). The disulphide bridge helps stabilize the secondary structure formation that builds the base of the lasso. Remarkably, leptin and mIL-3 mainly slipknot structural components through their lassos, whereas the remaining PLBs thread the C-terminal helix through the N-terminal lasso by a so-called plugging mechanism [26], [27]. We provide, for the first time, direct evidence that the size of the covalent loop influences the threading mechanism. A small loop primarily uses a slipknotting route while the bigger loops are preferentially pierced by a plugging mechanism. In all cases, the N-terminal receptor-binding helix (helix A) is the last element to fold. All PLBs found to date, save leptin, have an N-terminal lasso that pins down the canonical helix A via a covalent linkage, while leptin's helix A has freedom to reorient and fray in the C-terminal PLB. This permutation from the more common N-terminal to C-terminal linkage of the PLBs results in an intriguing switch of the receptor binding helix A from tethered to dynamic and suggests that while the functional landscapes are shared in PLBs, variations in protein-receptor interface dynamics may be needed for signaling activity.
Cytokines are soluble proteins secreted from different organs/tissues that act as chemical messengers important in intercellular communication. All cytokines bind to a subset of homologous membrane bound receptors, activating similar intercellular signaling pathways [29], [30], [31]. The conserved cytokine motif, a four-helix bundle, indicates that the helical cytokines may have evolved from the same ancestral origin (Supporting Figure S1 and Supporting Table S1). Despite the structural identities, there are little or no sequence similarities within the family due to co-evolution, where each ligand and its specific receptor have diverged in sequence from its ancestors. Therefore, recognition by commonly used sequence homology methods is not possible [32]. Instead, structural methods are used to classify these four-helix bundles as cytokines. Furthermore, all cytokines share a characteristic up-up-down-down fold, forming a two-layer packing of anti-parallel helix pairs were helix A and D packs against helix C and B. The superfamily of helical cytokines is divided into three families: long-chain helical cytokines, short-chain helical cytokines and interferons/interleukin 10 (Figure 3) [32]. While the overall geometry of the cytokines is conserved, there are differences in structure such as chain length and secondary structural elements (Supporting Table S1).
The PLB protein motif in leptin is a unique fold for proteins in general. A search of the PDB lead to the discovery of an additional 11 proteins with a similar threaded motif. Here, we compare leptin dynamics and threading to four PLBs, two Zebra fish interferons, human (hIL-3) and murine (mIL-3) interleukin 3 (Supporting Table S1). Additionally, three unthreaded four-helix bundles were investigated as controls, namely Granulocyte colony-stimulating factor (G-CSF), Leukemia inhibitory factor (LIF) and human Growth Hormone (hGH). Figure 2 displays the various structures as well as a cartoon describing the position of cysteines (yellow) creating the two types of lassos, i.e. the N-terminal loop (light blue) and the C-terminal loop (dark blue). The four canonical helices making up the core of each protein are labeled A–D. Additional helices are numbered from the position in sequence; for example, the extra helix in leptin is the fourth helix in sequence and is labeled 4′.
Structure-based simulations were used to investigate the folding mechanism of the PLB proteins. Two different oxidation states were investigated for the disulphide bridge involved in the lasso, i.e. the reduced state (blue) and the oxidized sate (red) (details in Section Methods). Three unthreaded helical bundles were used as controls and their reduced states (DD, details in Section Methods) are plotted in black. Additionally, both the reduced and oxidized state of hGH were plotted as a control for an unthreaded structure, as it has a large “empty” covalent loop, where nothing is threaded through this element. The folding transition is monitored by the fraction of native contacts formed (Q) along the folding trajectory. A native contact is a contact formed between two residues that are close in the native state. Q varies from 0, completely denatured, to 1, completely native. The folding mechanism is monitored via q(segment), the fraction of native contacts formed by a secondary structure element. q(segment) versus Q shows the average number of contacts a segment makes as a function of the overall folding progress, and therefore discerns the average order of events during folding. The results are plotted in Figure 4 and Supplementary Figure S2, S3, S4. The diagonal dashed gray line shows where q(segment) is tracking the overall folding progress. The boxes represent the actual positioning of the covalent loop from Figure 2, where light blue represents the N-terminal loop, dark blue the C-terminal loop and gray the unthreaded structures.
The NSD for leptin together with in vitro activity assays revealed that the disulphide-bond plays an important role in controlling receptor binding [40] and thus biological activity by controlling local motions on distal receptor-binding sites far removed from the disulphide-bridge (Figure 6). These shifts are seen, for example, in helix A as well as in loop four, despite leptin is a C-terminal PLB [12]. To quantify the NSD for the PLBs we performed all-atom structure- based simulations far below the folding temperature, where the protein is effectively always in the folded basin. We calculated the essential dynamics, of the backbone, by projecting the trajectory onto the first four principle components. Oxidation has a significant effect on the amplitude of fluctuation of individual amino acids along the sequence for the PLBs due to topological constraints introduced by the threaded element [41] (Figure 7 and Supporting Figure S5). These modulations in fluctuations are not limited to the regions in the vicinity of the disulphide. Since the disulphide bridge mobilizes helix A, the dynamics of the N-terminal PLBs show the largest shifts. Both interferons show additional small increased NSD in the reduced state around the helical hairpin (loop 2, helix B and C), which is part of the closed loop. The interleukins show increased dynamics in both terminals in the reduced state. Indeed, in hIL-3 helix A even unfolds completely up on reduction. Taken together, the NSD data indicate that both the interferons and interleukins are more dynamic in the reduced state. However, leptin is unique in that it introduces increased dynamics in the oxidized state. As a control, we performed NSD for the empty covalent loop homolog protein, hGH. This protein shows no significant changes between the oxidized and reduced protein, with the exception of the expected increased local dynamics in the vicinity of the disulphide bridge upon reduction. Taken together, this data indicates that the observed long-range changes in dynamics in the PLBs are a direct result of the effect of a closed loop in the presence of a threaded element of the polypeptide chain. In the case of leptin, increased dynamics distal to the knot in the oxidized state is likely a consequence of the inability to relax the tension introduced into the vicinity of the small closed loop in the lasso. In the other cases discovered to date, local motions are enhanced by increasing the size of the loop and the expected reduction in dynamics upon disulphide bond formation are observed. That is, the simple formation of the closed loop effects only the local but not the global dynamics in the helical bundles, while the presence of a PL alters the global NSDs in both expected and unexpected ways.
Frustrated surface regions have been proposed as sites relevant for allostery [42]. NSD and frustration in proteins have shown to be essential to protein function [43], [44]. The conserved region for receptor binding within the cytokine family is helix A [45], [46], [47], [48], [49], [50], [51]. The late formation of helix A as well as the fraying of helix A in the reduced state of the PLBs implies that the dynamics is important to receptor binding and activity of the PLBs [12]. As an example, we show the receptor complex of leptin and the changed dynamics between the reduced and oxidized state in Figure 6. This suggests that the malleability of the substrate and receptor interface has been conserved throughout evolution (Supporting Figure S1). More importantly, leptin has kept the malleability in helix A by forming its covalent loop at the C-terminal end. Interestingly, comparing the oxidized and reduced NSD data for leptin reveals a altered dynamics of helix A where the oxidized protein of leptin has greater dynamics then the reduced state. In vitro activity assays showed that oxidized leptin is more active than the reduced protein, and supports that a dynamic structure is of importance for biological activity [12]. Nevertheless, all unthreaded helical bundles show reduced dynamic of helix A in general than the PLBs suggesting an overall more rigid structure. Our results suggest that a threaded topology is an important factor designating function. Here we focus on the knot-like four helix bundle class of proteins, however based on the current results we expect that other proteins will have a similar complex topology.
We have found a new class of knotted proteins, namely the Pierced Lasso Bundles (PLBs). The PLB topology is defined as a four-helix bundle where a disulphide bridge closes a covalent loop, and part of the polypeptide chain is slipknotted/plugged through this covalent loop. All PLBs discovered, save leptin, have their covalent loop at the N-terminal end and plug the C-terminal helix through the covalent loop. In contrast, leptin mainly slipknots a helical-hairpin through its C-terminal loop. The closed loop also changes the dynamics of helix A which is important for activity and receptor interaction [45], [46], [47], [48], [49], [50], [51]. The NSD reveals that oxidized leptin is more flexible than the reduced state, implying that dynamics in helix A is biologically important for leptin. Interestingly, sequence alignments of leptin homologs reveal that chicken and turkey leptin has three cysteines, one at the C-terminal end, one at position 100 and one at the N-terminal end [52], [53], [54]. Additionally, natural genetic polymorphism in bovine leptin has developed a sequence with a single cysteine substitution at the N-terminal end (R4C) [52], [53], [54]. This actually allows for three different combinations of covalent loops: (1) Closure of the C-terminal covalent loop, as seen in wild-type leptin, forming a disulphide bridge between residue C96 and C146. (2) The formation of an almost completely circularized protein where residue C4 and C146 form a disulphide bridge, creating a “zero knot”. (3) The formation of an N-terminal covalent loop where residue C4 binds to residue C96. In the latter case, the C-terminal helix could either slipknot or plug through the closed loop. Future in vitro experiments can distinguish the three states from each other and the effect of a long N-terminal PLB (90 residues) versus the shorter C-terminal PLB as well as investigate fully circularized “zero knot” protein as a template for understanding knot formation and threading control of function in the PLBs (Figure 1). The folding landscape of these three states of leptin could additionally be studied by traditional mutagenic analyses [55], [56], [57]. However, the analysis of the full landscape is complicated by the early threading of the covalent loop that occurs at the level of the transition state. While threading mechanisms have previously been investigated through Fluorescence Resonance Energy Transfer (FRET) [58] this is not an optimal technique in the case of leptin as the loop is 50 residues long and big probes could compromise the threading event. On the other hand, leptin is an optimal system for pulling experiments. For example, a major issue in the field is the inability to untie knots with denaturant [59]. In the case of leptin, simply reducing the disulphide bridge unthreads the structure and we are assured that we are comparing the energetics of the fully threaded and fully unthreaded states. Additionally, reversibly knotting proteins with pulling experiments is extremely complicated [60], while for leptin the experiment is straightforward and can be used to investigate rate of loop threading. Nevertheless, understanding the topological constraints will lead to a broader understanding of the exotic shapes of the free energy landscapes in the growing class of knot-like PLB proteins. Moreover, one should point out that there are probably other undiscovered PL structures deposited in the PDB, where they proteins are bold up of β-strands and/or mixed α/β proteins where the loop is a cinch instead of a lasso. Finally, knowledge about topological constraints in the PLBs could increase the interest of researchers to pharmacologically modulate the pleiotropic hormone leptin [61] and other cytokines, as they have become a hotspot for many medical disorders as cancer, reproduction, diabetes, obesity among others [62], [63], [64], [65]. Taken into account that the PLBs identified here show a different behavior than the unthreaded four-helix bundles suggests that the importance of the threaded element should be considered in modulating receptor ligand interactions for therapeutic development.
In this work we used a Cα SBMs [66], [67] to investigate the folding of eight helical cytokines, including five PLB proteins (PDB code 1AX8, 3PIV, 3PIW, 1JLI and 2L3O) and three unthreaded four-helix bundles (1RHG, 1EMR and 1HGU). Each amino acid is represented as a single bead and attractive interactions are given to residue pairs close in the native state. These native interactions are identified based on a shadow map [68], [69]. The basic Hamiltonian is,
Native interactions have a repulsive term plus an attractive Gaussian term. The R(rij)Gij(rij) term is a correction that anchors the minimum of each contact -ε (where the last two corresponds respectively to attractive and repulsive non-bonded interactions [70]. denotes the native distance between atoms i and j along the sequence. The local topology of the chain is described by the native angles between the bonds connecting residue pairs ij and jk, and by the native dihedrals or torsional angles between the planes defined by atoms ijk and jkl. The strengths of the interactions are given in reduced energy units by the constants kb = 2×104 ε/nm2, ka = 40 ε/rad2, k1d = ε and k2d = 0.5ε, where ε is the reduced energy unit. Σ = 4 Å. The details of the model are characterised elsewhere [70], [71].
We used the web server SMOG (http://smog-server.org/) to create the input files for our simulations [66], [68], [69]. The GROMACS 4.5.3 package was used to perform the molecular dynamics simulations [72]. The integration steps were t = 0.005, stochastic dynamics with coupling constant 2 was used to maintain temperature. The apparent folding temperatures are estimated from each maximum peak in each specific heat curve. For a formed native contact the energy gain is measured by epsilon (ε), and thus the temperatures and energies reported in this paper are measured in units of ε. For sufficient sampling of the transition states some proteins required umbrella sampling along Q as in [73]. Corrected folding mechanisms (Q versus q(segment)) were then created with the Weighted Histogram Analysis Method (WHAM) [74], [75].
To mimic the experimental conditions/environment for the disulphide bridge building up the covalent loop we used two comparable in silico models, i.e. a reduced state (blue in all plots), an oxidized state (red in all plots). The ability of the disulphide to make and break during folding also was employed to mimic the conditions where folding takes place at the respective reduction potential of the disulphide bridge. This state, the DynamicDisulphide, is best studied in silico where it can be explicitly defined. This state was also simulated for the unthreaded structures (G-CSF and LIF, black in all plots), see Haglund et al for a full description of the states of the disulphide bridge [12]. The hGH was simulated as a control for covalent loop formation, as one of the disulphide bridges (C53–C163, forming a so called “cinch”) forms a large “empty” covalent loop of 112 residues. This loop is classified as “empty” as no part of the polypeptide chain is threaded through the loop. This construct can help show the effects of the threaded element in PLBs.
All-atom structure-based simulations [66], [76] were performed to characterize the NSDs. To investigate the contribution of the threaded topology we performed simulations of both reduced and oxidized states for all PLBs as well as for hGH. A reduced state was simulated for the unthreaded structures (G-CSF and LIF). The slow component of the dynamics described by the first four eigenvector was analysed as described in Haglund et al [12].
Some of the crystal structures have gaps in the sequence. Therefore, the Arch pred server [77] was used to recreate the spaces in the structure of leptin, G-CSF and hGH. Due to problems with aggregation the interleukins are truncated at the N-terminal end [49], [50] (Supporting Table S1). Also, most of the proteins do not show complete density for the entire sequence as is stated in Supporting Table S1. To align all four-helix bundles with leptin we used the PDB tool “Compare Structures” using the comparison method jFATCAT-ridged and jFATCAT-flexible (http://www.pdb.org/pdb/workbench/workbench.do) [78]. The sequence alignment tools used to align all sequences to leptin were ALIGN Query (http://xylian.igh.cnrs.fr/bin/align-guess.cgi, for sequence identity) and ClustalW multiple sequence alignment (http://www.ebi.ac.uk/Tools/msa/clustalw2/, for sequence similarity). The results from the structural and sequence alignments are shown in Supporting Table S1. To find other proteins with a PLB topology we performed geometrical threading on precompiled all verses all input based on the structure of leptin given by jfatcat server. We used a 4 Å rmsd threshold during trace of the fragment matrix. In the second step, we analyzed the discovered structures with P-values lower than 4.0E−8 with two conditions: (1) Four-helix motif (all possible combination – motif to thread). (2) Distance along sequence for amino acids which form cysteine bridge has to be bigger than 40 amino acids but shorter than 200. The final set of structures were visually inspected and new motifs were used to repeat the same procedure. Other PL topologies/configurations likely exist; however, they are the subject of future studies as they would reside in a different fold family.
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.